{ "version": "1.0", "date_publication": "2026-10-21", "date_maj": "2026-10-21", "taux_bce_eur_usd": 1.1403, "date_taux_bce": "2026-09-25", "kwh_eur_ht": 0.1624, "amortissement_mois": 36, "heures_par_mois": 730, "pipeline_rafraichissement": { "moteur_orchestration": "save-agent-rs (http://127.0.0.1:8787)", "modele_analyse": "opus-5.5 (Claude Opus 5.5 via LiteLLM :4100)", "lecteur_primaire": "sar-reader read --render", "repli_antibot": "android_scrape (Chrome Android Redroid)", "controle_visuel": "valider-publication.py --visuel + sar-visual" }, "fournisseurs": { "nebius": { "id": "nebius", "nom": "Nebius AI Cloud", "pays_fr": "Union européenne (Finlande, France) et États-Unis", "pays_en": "European Union (Finland, France) & US", "url_tarifs": "https://nebius.com/prices", "url_rent_fr": "https://console.eu.nebius.com/", "url_rent_en": "https://console.eu.nebius.com/", "referral_actif": false, "referral_type": "b2b_partner", "referral_programme_fr": "Nebius Partner Program (B2B, enregistrement d'opportunités)", "referral_programme_en": "Nebius Partner Program (B2B deal registration)", "bonus_filleul_fr": "Tarifs dégressifs sur engagement réservé (3 à 12 mois) et instances Spot dès 0,79 $/h", "bonus_filleul_en": "Volume discounts on reserved commitments (3 to 12 months) and Spot instances from $0.79/h", "rel_attr": "external noopener", "dispo_nebius": true, "dispo_brev": true }, "brev": { "id": "brev", "nom": "NVIDIA Brev", "pays_fr": "Multi-cloud (Nebius, Crusoe, Lambda, CoreWeave, AWS, GCP)", "pays_en": "Multi-cloud (Nebius, Crusoe, Lambda, CoreWeave, AWS, GCP)", "url_tarifs": "https://developer.nvidia.com/brev", "url_rent_fr": "https://brev.nvidia.com/environment/new", "url_rent_en": "https://brev.nvidia.com/environment/new", "referral_actif": false, "referral_type": "credits_developer", "referral_programme_fr": "Crédits développeurs via NVIDIA Developer et NVIDIA Inception", "referral_programme_en": "Developer credits via NVIDIA Developer and NVIDIA Inception", "bonus_filleul_fr": "Déploiement en un clic (CUDA, Docker, Jupyter, VS Code, SSH) facturé à la seconde", "bonus_filleul_en": "One-click deployment (CUDA, Docker, Jupyter, VS Code, SSH) billed per second", "rel_attr": "external noopener", "dispo_nebius": false, "dispo_brev": true }, "runpod": { "id": "runpod", "nom": "RunPod", "pays_fr": "États-Unis et Europe (31 régions mondiales)", "pays_en": "US & Europe (31 global regions)", "url_tarifs": "https://www.runpod.io/pricing", "url_rent_fr": "https://www.runpod.io/console/gpu-cloud?ref=qdna", "url_rent_en": "https://www.runpod.io/console/gpu-cloud?ref=qdna", "referral_actif": true, "referral_type": "self_serve_and_affiliate", "referral_programme_fr": "Parrainage RunPod (3 % Pods / 5 % Serverless en crédits, ou 10 % cash PartnerStack)", "referral_programme_en": "RunPod Referral (3% Pods / 5% Serverless credits, or 10% cash via PartnerStack)", "bonus_filleul_fr": "Bonus de 5 $ à 500 $ de crédits à l'inscription après premier rechargement (10 $ min.)", "bonus_filleul_en": "$5 to $500 signup credit bonus after first top-up ($10 min.)", "rel_attr": "external nofollow noopener sponsored", "dispo_nebius": false, "dispo_brev": false }, "vast": { "id": "vast", "nom": "Vast.ai", "pays_fr": "Place de marché mondiale (centres de données certifiés et hôtes)", "pays_en": "Global marketplace (verified datacenters & hosts)", "url_tarifs": "https://vast.ai/pricing", "url_rent_fr": "https://cloud.vast.ai/create/?ref_id=qdna", "url_rent_en": "https://cloud.vast.ai/create/?ref_id=qdna", "referral_actif": true, "referral_type": "self_serve_and_affiliate", "referral_programme_fr": "Programme d'affiliation Vast.ai (3 % récurrents à vie, 75 % convertibles en virement)", "referral_programme_en": "Vast.ai Affiliate Program (3% lifetime recurring, 75% cash withdrawable)", "bonus_filleul_fr": "Tarifs planchers mondiaux à la seconde sur GPU grand public et serveur", "bonus_filleul_en": "Per-second global floor pricing across consumer and datacenter GPUs", "rel_attr": "external nofollow noopener sponsored", "dispo_nebius": false, "dispo_brev": false }, "vultr": { "id": "vultr", "nom": "Vultr Cloud GPU", "pays_fr": "États-Unis, Europe (Paris, Francfort, Amsterdam) et Asie (32 régions)", "pays_en": "US, Europe (Paris, Frankfurt, Amsterdam) & Asia (32 regions)", "url_tarifs": "https://www.vultr.com/products/cloud-gpu/", "url_rent_fr": "https://www.vultr.com/products/cloud-gpu/?ref=qdna", "url_rent_en": "https://www.vultr.com/products/cloud-gpu/?ref=qdna", "referral_actif": true, "referral_type": "self_serve_and_affiliate", "referral_programme_fr": "Programme affilié Vultr (jusqu'à 100 $ CPA par nouveau compte vérifié)", "referral_programme_en": "Vultr Referral Program (up to $100 CPA per verified new user)", "bonus_filleul_fr": "Jusqu'à 300 $ de crédits d'essai (valables 30 jours) pour les nouveaux inscrits", "bonus_filleul_en": "Up to $300 free trial credits (valid 30 days) for new signups", "rel_attr": "external nofollow noopener sponsored", "dispo_nebius": false, "dispo_brev": false }, "scaleway": { "id": "scaleway", "nom": "Scaleway (Groupe iliad)", "pays_fr": "France (Paris fr-par-1 / fr-par-2, cloud souverain européen)", "pays_en": "France (Paris fr-par-1 / fr-par-2, European sovereign cloud)", "url_tarifs": "https://www.scaleway.com/en/pricing/gpu/", "url_rent_fr": "https://console.scaleway.com/gpu-instances/instances", "url_rent_en": "https://console.scaleway.com/gpu-instances/instances", "referral_actif": false, "referral_type": "b2b_partner", "referral_programme_fr": "Scaleway Partner Program (volet Referral rewards B2B)", "referral_programme_en": "Scaleway Partner Program (B2B Referral rewards track)", "bonus_filleul_fr": "Hébergement 100 % français (Paris), facturation directe en euros HT à l'heure", "bonus_filleul_en": "100% French hosting (Paris), direct hourly billing in EUR excl. VAT", "rel_attr": "external noopener", "dispo_nebius": false, "dispo_brev": false }, "ovhcloud": { "id": "ovhcloud", "nom": "OVHcloud", "pays_fr": "France (Gravelines, Strasbourg) et Europe", "pays_en": "France (Gravelines, Strasbourg) & Europe", "url_tarifs": "https://www.ovhcloud.com/fr/public-cloud/prices/", "url_rent_fr": "https://www.ovhcloud.com/fr/public-cloud/gpu/", "url_rent_en": "https://www.ovhcloud.com/en/public-cloud/gpu/", "referral_actif": false, "referral_type": "affiliate_awin", "referral_programme_fr": "Programme affilié OVHcloud (réseau Awin) et OVHcloud Partner Program", "referral_programme_en": "OVHcloud Affiliate Program (Awin network) & Partner Program", "bonus_filleul_fr": "200 € de crédits Public Cloud offerts à l'ouverture d'un nouveau compte", "bonus_filleul_en": "€200 free Public Cloud credits upon opening a new account", "rel_attr": "external noopener", "dispo_nebius": false, "dispo_brev": false }, "lambda": { "id": "lambda", "nom": "Lambda", "pays_fr": "États-Unis (Texas, Utah, Virginie)", "pays_en": "United States (Texas, Utah, Virginia)", "url_tarifs": "https://lambda.ai/service/gpu-cloud", "url_rent_fr": "https://cloud.lambdalabs.com/instances", "url_rent_en": "https://cloud.lambdalabs.com/instances", "referral_actif": false, "referral_type": "b2b_partner", "referral_programme_fr": "Lambda AI Cloud Partner Program (B2B) et intégration native NVIDIA Brev", "referral_programme_en": "Lambda AI Cloud Partner Program (B2B) & native NVIDIA Brev integration", "bonus_filleul_fr": "Facturation à la minute, trafic réseau entrant/sortant gratuit, disponible via NVIDIA Brev", "bonus_filleul_en": "Per-minute billing, zero egress fees, available directly via NVIDIA Brev", "rel_attr": "external noopener", "dispo_nebius": false, "dispo_brev": true }, "crusoe": { "id": "crusoe", "nom": "Crusoe Cloud", "pays_fr": "Islande (géothermie) et États-Unis", "pays_en": "Iceland (geothermal) & United States", "url_tarifs": "https://www.crusoe.ai/cloud/pricing", "url_rent_fr": "https://console.crusoecloud.com/", "url_rent_en": "https://console.crusoecloud.com/", "referral_actif": false, "referral_type": "b2b_partner", "referral_programme_fr": "Crusoe Partner Program (Referral partnerships B2B) et intégration NVIDIA Brev", "referral_programme_en": "Crusoe Partner Program (B2B referral partnerships) & NVIDIA Brev integration", "bonus_filleul_fr": "Énergie décarbonée, disponible en 1 clic via NVIDIA Brev", "bonus_filleul_en": "Clean energy datacenters, 1-click deployment via NVIDIA Brev", "rel_attr": "external noopener", "dispo_nebius": false, "dispo_brev": true }, "coreweave": { "id": "coreweave", "nom": "CoreWeave", "pays_fr": "États-Unis et Europe (Royaume-Uni, Suède, Espagne)", "pays_en": "US & Europe (UK, Sweden, Spain)", "url_tarifs": "https://www.coreweave.com/pricing", "url_rent_fr": "https://console.coreweave.com/", "url_rent_en": "https://console.coreweave.com/", "referral_actif": false, "referral_type": "b2b_partner", "referral_programme_fr": "Partenariat B2B CoreWeave et intégration NVIDIA Brev", "referral_programme_en": "CoreWeave B2B Partner & NVIDIA Brev integration", "bonus_filleul_fr": "Clusters Kubernetes bare-metal InfiniBand NDR/XDR, accessible via NVIDIA Brev", "bonus_filleul_en": "Bare-metal InfiniBand NDR/XDR Kubernetes clusters, accessible via NVIDIA Brev", "rel_attr": "external noopener", "dispo_nebius": false, "dispo_brev": true }, "hyperstack": { "id": "hyperstack", "nom": "Hyperstack (NexGen Cloud)", "pays_fr": "Norvège, Royaume-Uni et Canada (100 % énergies renouvelables)", "pays_en": "Norway, UK & Canada (100% renewable energy)", "url_tarifs": "https://www.hyperstack.cloud/gpu-pricing", "url_rent_fr": "https://console.hyperstack.cloud/deploy-virtual-machine", "url_rent_en": "https://console.hyperstack.cloud/deploy-virtual-machine", "referral_actif": false, "referral_type": "partner_negotiated", "referral_programme_fr": "Partenariat NexGen Cloud et intégration directe dans NVIDIA Brev", "referral_programme_en": "NexGen Cloud Partner & direct NVIDIA Brev integration", "bonus_filleul_fr": "Réductions pour les membres NVIDIA Inception, déployable depuis NVIDIA Brev", "bonus_filleul_en": "Discounts for NVIDIA Inception members, deployable from NVIDIA Brev", "rel_attr": "external noopener", "dispo_nebius": false, "dispo_brev": true }, "tensorwave": { "id": "tensorwave", "nom": "TensorWave", "pays_fr": "États-Unis (spécialiste exclusif AMD Instinct CDNA 3 et CDNA 4)", "pays_en": "United States (exclusive AMD Instinct CDNA 3 & CDNA 4 specialist)", "url_tarifs": "https://tensorwave.com/products/accelerators/amd-mi325x", "url_rent_fr": "https://cloud.tensorwave.com/", "url_rent_en": "https://cloud.tensorwave.com/", "referral_actif": false, "referral_type": "none_public", "referral_programme_fr": "Accès direct sans intermédiaire aux clusters bare-metal AMD MI355X, MI325X et MI300X", "referral_programme_en": "Direct bare-metal access to AMD MI355X, MI325X and MI300X clusters", "bonus_filleul_fr": "Nœuds 8x OAM 1,5 à 2,3 To HBM3/HBM3e avec réseau RoCEv2 3,2 Tbit/s", "bonus_filleul_en": "8x OAM nodes with 1.5 to 2.3 TB HBM3/HBM3e and 3.2 Tbps RoCEv2 fabric", "rel_attr": "external noopener", "dispo_nebius": false, "dispo_brev": false }, "hotaisle": { "id": "hotaisle", "nom": "Hot Aisle", "pays_fr": "États-Unis (Michigan, clusters Dell PowerEdge XE9680 AMD MI300X)", "pays_en": "United States (Michigan, Dell PowerEdge XE9680 AMD MI300X clusters)", "url_tarifs": "https://hotaisle.xyz/pricing/", "url_rent_fr": "https://hotaisle.xyz/pricing/", "url_rent_en": "https://hotaisle.xyz/pricing/", "referral_actif": false, "referral_type": "none_public", "referral_programme_fr": "Accès libre-service par SSH aux machines Dell XE9680 MI300X", "referral_programme_en": "Self-serve SSH access to Dell XE9680 MI300X machines", "bonus_filleul_fr": "Stockage local massif (15,36 To NVMe par GPU, soit 122 To par nœud 8x)", "bonus_filleul_en": "Massive local storage (15.36 TB NVMe per GPU, 122 TB per 8x node)", "rel_attr": "external noopener", "dispo_nebius": false, "dispo_brev": false }, "fluidstack": { "id": "fluidstack", "nom": "FluidStack (Verda)", "pays_fr": "Europe (Islande, Norvège) et États-Unis", "pays_en": "Europe (Iceland, Norway) & United States", "url_tarifs": "https://www.fluidstack.io/pricing", "url_rent_fr": "https://console.fluidstack.io/", "url_rent_en": "https://console.fluidstack.io/", "referral_actif": false, "referral_type": "b2b_partner", "referral_programme_fr": "Fournisseur intégré au plan de contrôle NVIDIA Brev", "referral_programme_en": "Provider integrated into the NVIDIA Brev control plane", "bonus_filleul_fr": "Instances NVIDIA et AMD à la demande ou Spot, intégrées à NVIDIA Brev", "bonus_filleul_en": "On-demand and Spot NVIDIA and AMD instances, integrated into NVIDIA Brev", "rel_attr": "external noopener", "dispo_nebius": false, "dispo_brev": true }, "denvr": { "id": "denvr", "nom": "Denvr Dataworks", "pays_fr": "Canada et États-Unis (refroidissement liquide par immersion)", "pays_en": "Canada & United States (liquid immersion cooling)", "url_tarifs": "https://www.denvrdata.com/pricing", "url_rent_fr": "https://www.denvrdata.com/pricing", "url_rent_en": "https://www.denvrdata.com/pricing", "referral_actif": false, "referral_type": "none_public", "referral_programme_fr": "Location directe de nœuds NVIDIA H100/H200/A100 et Intel Gaudi 2", "referral_programme_en": "Direct rental of NVIDIA H100/H200/A100 and Intel Gaudi 2 nodes", "bonus_filleul_fr": "Tarifs très compétitifs sur Intel Gaudi 2 (1,10 $/h) et NVIDIA H100 SXM5 (2,45 $/h)", "bonus_filleul_en": "Competitive pricing on Intel Gaudi 2 ($1.10/h) and NVIDIA H100 SXM5 ($2.45/h)", "rel_attr": "external noopener", "dispo_nebius": false, "dispo_brev": false }, "tensordock": { "id": "tensordock", "nom": "TensorDock", "pays_fr": "États-Unis, Europe et Asie (plus de 100 centres de données)", "pays_en": "US, Europe & Asia (100+ datacenters)", "url_tarifs": "https://tensordock.com/", "url_rent_fr": "https://dashboard.tensordock.com/deploy", "url_rent_en": "https://dashboard.tensordock.com/deploy", "referral_actif": false, "referral_type": "none_by_policy", "referral_programme_fr": "Politique zéro commission d'affiliation pour maintenir les prix horaires au plancher", "referral_programme_en": "Zero-affiliate policy to keep hourly GPU rates at floor level", "bonus_filleul_fr": "Facturation à la seconde sans frais d'intermédiation", "bonus_filleul_en": "Per-second billing with zero intermediation markup", "rel_attr": "external noopener", "dispo_nebius": false, "dispo_brev": false }, "inteltiber": { "id": "inteltiber", "nom": "Intel Tiber AI Cloud", "pays_fr": "États-Unis et Europe (cloud officiel Intel Gaudi)", "pays_en": "US & Europe (official Intel Gaudi cloud)", "url_tarifs": "https://cloud.intel.com/compute-instances", "url_rent_fr": "https://cloud.intel.com/compute-instances", "url_rent_en": "https://cloud.intel.com/compute-instances", "referral_actif": false, "referral_type": "credits_developer", "referral_programme_fr": "Crédits développeurs et programme startups Intel Liftoff", "referral_programme_en": "Developer credits & Intel Liftoff startup program", "bonus_filleul_fr": "Accès officiel aux nœuds bare-metal 8x Intel Gaudi 3 (1 024 Go) et Gaudi 2 (768 Go)", "bonus_filleul_en": "Official access to 8x Intel Gaudi 3 (1,024 GB) and Gaudi 2 (768 GB) bare-metal nodes", "rel_attr": "external noopener", "dispo_nebius": false, "dispo_brev": false }, "ibmcloud": { "id": "ibmcloud", "nom": "IBM Cloud", "pays_fr": "Europe (Francfort) et États-Unis (Dallas, Washington)", "pays_en": "Europe (Frankfurt) & US (Dallas, Washington)", "url_tarifs": "https://www.ibm.com/products/gpu-ai-accelerator/intel-gaudi3", "url_rent_fr": "https://www.ibm.com/products/gpu-ai-accelerator/intel-gaudi3", "url_rent_en": "https://www.ibm.com/products/gpu-ai-accelerator/intel-gaudi3", "referral_actif": false, "referral_type": "b2b_partner", "referral_programme_fr": "IBM Partner Plus (B2B)", "referral_programme_en": "IBM Partner Plus (B2B)", "bonus_filleul_fr": "Instances VPC d'entreprise 8x Intel Gaudi 3 en Europe (Francfort) et aux États-Unis", "bonus_filleul_en": "Enterprise VPC 8x Intel Gaudi 3 instances in Europe (Frankfurt) and the US", "rel_attr": "external noopener", "dispo_nebius": false, "dispo_brev": false } }, "serveurs_sur_site": [ { "id": "rtx-5090-1x", "nom_fr": "Station 1x NVIDIA RTX 5090 (32 Go GDDR7)", "nom_en": "1x NVIDIA RTX 5090 Workstation (32 GB GDDR7)", "marque": "NVIDIA", "vram_go": 32, "prix_achat_eur_ht": 6500, "puissance_w": 850 }, { "id": "dgx-spark", "nom_fr": "NVIDIA DGX Spark GB10 (128 Go unifiés)", "nom_en": "NVIDIA DGX Spark GB10 (128 GB unified)", "marque": "NVIDIA", "vram_go": 128, "prix_achat_eur_ht": 7000, "puissance_w": 240 }, { "id": "rtx-pro-6000-1x", "nom_fr": "Station 1x NVIDIA RTX PRO 6000 Blackwell (96 Go GDDR7)", "nom_en": "1x NVIDIA RTX PRO 6000 Blackwell Workstation (96 GB GDDR7)", "marque": "NVIDIA", "vram_go": 96, "prix_achat_eur_ht": 23000, "puissance_w": 900 }, { "id": "rtx-pro-6000-2x", "nom_fr": "Station 2x NVIDIA RTX PRO 6000 Blackwell (192 Go GDDR7)", "nom_en": "2x NVIDIA RTX PRO 6000 Blackwell Workstation (192 GB GDDR7)", "marque": "NVIDIA", "vram_go": 192, "prix_achat_eur_ht": 43000, "puissance_w": 1600 }, { "id": "rtx-pro-6000-8x", "nom_fr": "Serveur 8x NVIDIA RTX PRO 6000 Blackwell (768 Go GDDR7)", "nom_en": "8x NVIDIA RTX PRO 6000 Blackwell Server (768 GB GDDR7)", "marque": "NVIDIA", "vram_go": 768, "prix_achat_eur_ht": 192000, "puissance_w": 5800 }, { "id": "mi300x-8x", "nom_fr": "Serveur 8x AMD Instinct MI300X OAM (1 536 Go HBM3)", "nom_en": "8x AMD Instinct MI300X OAM Server (1,536 GB HBM3)", "marque": "AMD", "vram_go": 1536, "prix_achat_eur_ht": 277000, "puissance_w": 8500 }, { "id": "h100-8x", "nom_fr": "Serveur 8x NVIDIA H100 SXM5 (640 Go HBM3)", "nom_en": "8x NVIDIA H100 SXM5 Server (640 GB HBM3)", "marque": "NVIDIA", "vram_go": 640, "prix_achat_eur_ht": 335000, "puissance_w": 10200 }, { "id": "mi325x-8x", "nom_fr": "Serveur 8x AMD Instinct MI325X OAM (2 048 Go HBM3e)", "nom_en": "8x AMD Instinct MI325X OAM Server (2,048 GB HBM3e)", "marque": "AMD", "vram_go": 2048, "prix_achat_eur_ht": 335000, "puissance_w": 10000 }, { "id": "dgx-h200", "nom_fr": "NVIDIA DGX H200 8x SXM5 (1 128 Go HBM3e)", "nom_en": "NVIDIA DGX H200 8x SXM5 (1,128 GB HBM3e)", "marque": "NVIDIA", "vram_go": 1128, "prix_achat_eur_ht": 429000, "puissance_w": 10200 }, { "id": "dgx-b200", "nom_fr": "NVIDIA DGX B200 8x SXM6 (1 440 Go HBM3e)", "nom_en": "NVIDIA DGX B200 8x SXM6 (1,440 GB HBM3e)", "marque": "NVIDIA", "vram_go": 1440, "prix_achat_eur_ht": 527000, "puissance_w": 14300 }, { "id": "dgx-b300", "nom_fr": "NVIDIA DGX B300 8x SXM6 (2 304 Go HBM3e)", "nom_en": "NVIDIA DGX B300 8x SXM6 (2,304 GB HBM3e)", "marque": "NVIDIA", "vram_go": 2304, "prix_achat_eur_ht": 736000, "puissance_w": 14500 } ], "offres": [ { "id": "tensordock-rtx4090-1x", "marque": "NVIDIA", "famille": "Ada Lovelace", "carte_gpu": "NVIDIA GeForce RTX 4090", "architecture": "Ada Lovelace", "nb_gpu": 1, "vram_gpu_go": 24, "vram_totale_go": 24, "type_vram": "GDDR6X", "bp_memoire_gpu_tos": 1.01, "bp_memoire_totale_tos": 1.01, "tflops_fp16": 165, "tflops_fp8": 330, "fournisseur_id": "tensordock", "fournisseur": "TensorDock", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Place de marché (Vast.ai dès 0,15 $/h)", "mode_tarif_en": "Marketplace (Vast.ai from $0.15/h)", "machine_ref": "TensorDock RTX 4090", "cpu_modele": "AMD EPYC", "vcpu_total": 16, "ram_systeme_go": 32, "stockage_nvme_to": 1.0, "interconnexion": "PCIe Gen4 x16", "reseau_cluster": "Ethernet 10 Gbps", "prix_heure_gpu_usd": 0.35, "prix_heure_gpu_eur_ht": 0.31, "prix_heure_serveur_usd": 0.35, "prix_heure_serveur_eur_ht": 0.31, "cout_mensuel_730h_serveur_eur_ht": 226, "cout_mensuel_730h_serveur_usd": 255, "prix_spot_gpu_usd": 0.15, "prix_spot_gpu_eur_ht": 0.13, "vram_par_euro_heure": 77.4, "url_source": "https://tensordock.com/", "url_rent_fr": "https://dashboard.tensordock.com/deploy", "url_rent_en": "https://dashboard.tensordock.com/deploy", "url_brev": null, "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "rtx-5090-1x", "date_releve": "2026-10-21" }, { "id": "vast-rtx5090-1x", "marque": "NVIDIA", "famille": "Blackwell", "carte_gpu": "NVIDIA GeForce RTX 5090", "architecture": "Blackwell", "nb_gpu": 1, "vram_gpu_go": 32, "vram_totale_go": 32, "type_vram": "GDDR7", "bp_memoire_gpu_tos": 1.79, "bp_memoire_totale_tos": 1.79, "tflops_fp16": 209, "tflops_fp8": 419, "fournisseur_id": "vast", "fournisseur": "Vast.ai", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Plancher à la demande (Spot : 0,25 $/h)", "mode_tarif_en": "On-demand floor (Spot: $0.25/h)", "machine_ref": "Vast RTX 5090 32GB", "cpu_modele": "AMD EPYC / Ryzen 9", "vcpu_total": 16, "ram_systeme_go": 64, "stockage_nvme_to": 1.0, "interconnexion": "PCIe Gen5 x16", "reseau_cluster": "Ethernet 10 Gbps", "prix_heure_gpu_usd": 0.41, "prix_heure_gpu_eur_ht": 0.36, "prix_heure_serveur_usd": 0.41, "prix_heure_serveur_eur_ht": 0.36, "cout_mensuel_730h_serveur_eur_ht": 263, "cout_mensuel_730h_serveur_usd": 299, "prix_spot_gpu_usd": 0.25, "prix_spot_gpu_eur_ht": 0.22, "vram_par_euro_heure": 88.9, "url_source": "https://vast.ai/pricing", "url_rent_fr": "https://cloud.vast.ai/create/?ref_id=qdna", "url_rent_en": "https://cloud.vast.ai/create/?ref_id=qdna", "url_brev": null, "rel_attr": "external nofollow noopener sponsored", "referral_actif": true, "serveur_site_ref": "rtx-5090-1x", "date_releve": "2026-10-21" }, { "id": "vast-rtxpro6000-1x", "marque": "NVIDIA", "famille": "Blackwell", "carte_gpu": "NVIDIA RTX PRO 6000 Blackwell", "architecture": "Blackwell", "nb_gpu": 1, "vram_gpu_go": 96, "vram_totale_go": 96, "type_vram": "GDDR7 ECC", "bp_memoire_gpu_tos": 1.79, "bp_memoire_totale_tos": 1.79, "tflops_fp16": 500, "tflops_fp8": 1000, "fournisseur_id": "vast", "fournisseur": "Vast.ai", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Plancher vérifié (médiane : 1,24 $/h)", "mode_tarif_en": "Verified floor (median: $1.24/h)", "machine_ref": "Vast RTX PRO 6000 96GB", "cpu_modele": "AMD EPYC / Threadripper PRO", "vcpu_total": 24, "ram_systeme_go": 128, "stockage_nvme_to": 2.0, "interconnexion": "PCIe Gen5 x16", "reseau_cluster": "Ethernet 10 à 100 Gbps", "prix_heure_gpu_usd": 0.64, "prix_heure_gpu_eur_ht": 0.56, "prix_heure_serveur_usd": 0.64, "prix_heure_serveur_eur_ht": 0.56, "cout_mensuel_730h_serveur_eur_ht": 409, "cout_mensuel_730h_serveur_usd": 467, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 171.4, "url_source": "https://vast.ai/pricing", "url_rent_fr": "https://cloud.vast.ai/create/?ref_id=qdna", "url_rent_en": "https://cloud.vast.ai/create/?ref_id=qdna", "url_brev": null, "rel_attr": "external nofollow noopener sponsored", "referral_actif": true, "serveur_site_ref": "rtx-pro-6000-1x", "date_releve": "2026-10-21" }, { "id": "tensordock-rtx6000ada-1x", "marque": "NVIDIA", "famille": "Ada Lovelace", "carte_gpu": "NVIDIA RTX 6000 Ada Generation", "architecture": "Ada Lovelace", "nb_gpu": 1, "vram_gpu_go": 48, "vram_totale_go": 48, "type_vram": "GDDR6 ECC", "bp_memoire_gpu_tos": 0.96, "bp_memoire_totale_tos": 0.96, "tflops_fp16": 364, "tflops_fp8": 728, "fournisseur_id": "tensordock", "fournisseur": "TensorDock", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Place de marché à la seconde", "mode_tarif_en": "Per-second marketplace", "machine_ref": "TensorDock RTX 6000 Ada", "cpu_modele": "AMD EPYC Milan / Genoa", "vcpu_total": 16, "ram_systeme_go": 32, "stockage_nvme_to": 1.0, "interconnexion": "PCIe Gen4 x16", "reseau_cluster": "Ethernet 10 Gbps", "prix_heure_gpu_usd": 0.75, "prix_heure_gpu_eur_ht": 0.66, "prix_heure_serveur_usd": 0.75, "prix_heure_serveur_eur_ht": 0.66, "cout_mensuel_730h_serveur_eur_ht": 482, "cout_mensuel_730h_serveur_usd": 548, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 72.7, "url_source": "https://tensordock.com/", "url_rent_fr": "https://dashboard.tensordock.com/deploy", "url_rent_en": "https://dashboard.tensordock.com/deploy", "url_brev": null, "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "dgx-spark", "date_releve": "2026-10-21" }, { "id": "ovh-l4-1x", "marque": "NVIDIA", "famille": "Ada Lovelace", "carte_gpu": "NVIDIA L4", "architecture": "Ada Lovelace", "nb_gpu": 1, "vram_gpu_go": 24, "vram_totale_go": 24, "type_vram": "GDDR6", "bp_memoire_gpu_tos": 0.3, "bp_memoire_totale_tos": 0.3, "tflops_fp16": 121, "tflops_fp8": 242, "fournisseur_id": "ovhcloud", "fournisseur": "OVHcloud", "dispo_nebius": false, "dispo_brev": true, "mode_tarif_fr": "Public Cloud France (Scaleway : 0,79 €/h)", "mode_tarif_en": "France Public Cloud (Scaleway: €0.79/h)", "machine_ref": "l4-90", "cpu_modele": "AMD EPYC", "vcpu_total": 22, "ram_systeme_go": 90, "stockage_nvme_to": 0.4, "interconnexion": "PCIe Gen4 x16", "reseau_cluster": "Ethernet 25 Gbps", "prix_heure_gpu_usd": 0.86, "prix_heure_gpu_eur_ht": 0.75, "prix_heure_serveur_usd": 0.86, "prix_heure_serveur_eur_ht": 0.75, "cout_mensuel_730h_serveur_eur_ht": 548, "cout_mensuel_730h_serveur_usd": 628, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 32.0, "url_source": "https://www.ovhcloud.com/fr/public-cloud/prices/", "url_rent_fr": "https://www.ovhcloud.com/fr/public-cloud/gpu/", "url_rent_en": "https://www.ovhcloud.com/en/public-cloud/gpu/", "url_brev": "https://brev.nvidia.com/environment/new/public?gpu=L4", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "rtx-5090-1x", "date_releve": "2026-10-21" }, { "id": "crusoe-l40s-1x", "marque": "NVIDIA", "famille": "Ada Lovelace", "carte_gpu": "NVIDIA L40S", "architecture": "Ada Lovelace", "nb_gpu": 1, "vram_gpu_go": 48, "vram_totale_go": 48, "type_vram": "GDDR6 ECC", "bp_memoire_gpu_tos": 0.86, "bp_memoire_totale_tos": 0.86, "tflops_fp16": 362, "tflops_fp8": 733, "fournisseur_id": "crusoe", "fournisseur": "Crusoe Cloud", "dispo_nebius": false, "dispo_brev": true, "mode_tarif_fr": "À la demande (disponible sur Brev)", "mode_tarif_en": "On-demand (available on Brev)", "machine_ref": "l40s-48gb.1x", "cpu_modele": "AMD EPYC Genoa", "vcpu_total": 16, "ram_systeme_go": 96, "stockage_nvme_to": 1.0, "interconnexion": "PCIe Gen4 x16", "reseau_cluster": "Ethernet 100 Gbps", "prix_heure_gpu_usd": 1.0, "prix_heure_gpu_eur_ht": 0.88, "prix_heure_serveur_usd": 1.0, "prix_heure_serveur_eur_ht": 0.88, "cout_mensuel_730h_serveur_eur_ht": 642, "cout_mensuel_730h_serveur_usd": 730, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 54.5, "url_source": "https://www.crusoe.ai/cloud/pricing", "url_rent_fr": "https://console.crusoecloud.com/", "url_rent_en": "https://console.crusoecloud.com/", "url_brev": "https://brev.nvidia.com/environment/new/public?gpu=L40S", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "dgx-spark", "date_releve": "2026-10-21" }, { "id": "fluidstack-mi250x-1x", "marque": "AMD", "famille": "Instinct CDNA 2", "carte_gpu": "AMD Instinct MI250X OAM", "architecture": "CDNA 2", "nb_gpu": 1, "vram_gpu_go": 128, "vram_totale_go": 128, "type_vram": "HBM2e", "bp_memoire_gpu_tos": 3.28, "bp_memoire_totale_tos": 3.28, "tflops_fp16": 383, "tflops_fp8": 0, "fournisseur_id": "fluidstack", "fournisseur": "FluidStack (Verda)", "dispo_nebius": false, "dispo_brev": true, "mode_tarif_fr": "À la demande (Spot : 0,79 $/h)", "mode_tarif_en": "On-demand (Spot: $0.79/h)", "machine_ref": "FluidStack MI250X 128GB", "cpu_modele": "AMD EPYC Milan 7003", "vcpu_total": 16, "ram_systeme_go": 128, "stockage_nvme_to": 1.0, "interconnexion": "AMD Infinity Fabric 400 Go/s", "reseau_cluster": "Ethernet 100 Gbps", "prix_heure_gpu_usd": 1.09, "prix_heure_gpu_eur_ht": 0.96, "prix_heure_serveur_usd": 1.09, "prix_heure_serveur_eur_ht": 0.96, "cout_mensuel_730h_serveur_eur_ht": 701, "cout_mensuel_730h_serveur_usd": 796, "prix_spot_gpu_usd": 0.79, "prix_spot_gpu_eur_ht": 0.69, "vram_par_euro_heure": 133.3, "url_source": "https://inferencebench.io/gpus/amd-mi250x/", "url_rent_fr": "https://console.fluidstack.io/", "url_rent_en": "https://console.fluidstack.io/", "url_brev": "https://brev.nvidia.com/environment/new", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "dgx-spark", "date_releve": "2026-10-21" }, { "id": "denvr-gaudi2-8x", "marque": "Intel", "famille": "Gaudi 2", "carte_gpu": "Intel Gaudi 2 HL-225 OAM", "architecture": "Gaudi 2", "nb_gpu": 8, "vram_gpu_go": 96, "vram_totale_go": 768, "type_vram": "HBM2e", "bp_memoire_gpu_tos": 2.45, "bp_memoire_totale_tos": 19.6, "tflops_fp16": 432, "tflops_fp8": 865, "fournisseur_id": "denvr", "fournisseur": "Denvr Dataworks", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Nœud bare-metal 8x (768 Go HBM2e)", "mode_tarif_en": "8x Bare-metal node (768 GB HBM2e)", "machine_ref": "Denvr 8x Gaudi 2 HL-225", "cpu_modele": "2x Intel Xeon Platinum 8380 Ice Lake", "vcpu_total": 160, "ram_systeme_go": 1024, "stockage_nvme_to": 30.4, "interconnexion": "21x 100GbE RDMA RoCEv2 sur puce", "reseau_cluster": "3x 100GbE RoCEv2 par puce (2,4 Tbit/s)", "prix_heure_gpu_usd": 1.1, "prix_heure_gpu_eur_ht": 0.96, "prix_heure_serveur_usd": 8.8, "prix_heure_serveur_eur_ht": 7.68, "cout_mensuel_730h_serveur_eur_ht": 5606, "cout_mensuel_730h_serveur_usd": 6424, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 100.0, "url_source": "https://www.denvrdata.com/pricing", "url_rent_fr": "https://www.denvrdata.com/pricing", "url_rent_en": "https://www.denvrdata.com/pricing", "url_brev": null, "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "rtx-pro-6000-8x", "date_releve": "2026-10-21" }, { "id": "denvr-a100-8x", "marque": "NVIDIA", "famille": "Ampere", "carte_gpu": "NVIDIA HGX A100 SXM4", "architecture": "Ampere", "nb_gpu": 8, "vram_gpu_go": 80, "vram_totale_go": 640, "type_vram": "HBM2e", "bp_memoire_gpu_tos": 2.0, "bp_memoire_totale_tos": 16.0, "tflops_fp16": 312, "tflops_fp8": 0, "fournisseur_id": "denvr", "fournisseur": "Denvr Dataworks", "dispo_nebius": false, "dispo_brev": true, "mode_tarif_fr": "Nœud 8x A100 80 Go SXM4", "mode_tarif_en": "8x A100 80 GB SXM4 node", "machine_ref": "Denvr A100-80G-8x", "cpu_modele": "2x AMD EPYC Milan", "vcpu_total": 208, "ram_systeme_go": 1024, "stockage_nvme_to": 22.8, "interconnexion": "NVLink 3 (600 Go/s)", "reseau_cluster": "InfiniBand HDR 1,6 Tbit/s", "prix_heure_gpu_usd": 1.15, "prix_heure_gpu_eur_ht": 1.01, "prix_heure_serveur_usd": 9.2, "prix_heure_serveur_eur_ht": 8.08, "cout_mensuel_730h_serveur_eur_ht": 5898, "cout_mensuel_730h_serveur_usd": 6716, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 79.2, "url_source": "https://www.denvrdata.com/pricing", "url_rent_fr": "https://www.denvrdata.com/pricing", "url_rent_en": "https://www.denvrdata.com/pricing", "url_brev": "https://brev.nvidia.com/environment/new/public?gpu=A100", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "h100-8x", "date_releve": "2026-10-21" }, { "id": "runpod-rtx5090-1x", "marque": "NVIDIA", "famille": "Blackwell", "carte_gpu": "NVIDIA GeForce RTX 5090", "architecture": "Blackwell", "nb_gpu": 1, "vram_gpu_go": 32, "vram_totale_go": 32, "type_vram": "GDDR7", "bp_memoire_gpu_tos": 1.79, "bp_memoire_totale_tos": 1.79, "tflops_fp16": 209, "tflops_fp8": 419, "fournisseur_id": "runpod", "fournisseur": "RunPod", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Secure Cloud (Community : 0,69 $/h)", "mode_tarif_en": "Secure Cloud (Community: $0.69/h)", "machine_ref": "RunPod RTX 5090 Pod", "cpu_modele": "AMD EPYC", "vcpu_total": 15, "ram_systeme_go": 62, "stockage_nvme_to": 1.0, "interconnexion": "PCIe Gen5 x16", "reseau_cluster": "Ethernet 25 Gbps", "prix_heure_gpu_usd": 1.19, "prix_heure_gpu_eur_ht": 1.04, "prix_heure_serveur_usd": 1.19, "prix_heure_serveur_eur_ht": 1.04, "cout_mensuel_730h_serveur_eur_ht": 759, "cout_mensuel_730h_serveur_usd": 869, "prix_spot_gpu_usd": 0.69, "prix_spot_gpu_eur_ht": 0.61, "vram_par_euro_heure": 30.8, "url_source": "https://www.runpod.io/pricing", "url_rent_fr": "https://www.runpod.io/console/gpu-cloud?ref=qdna", "url_rent_en": "https://www.runpod.io/console/gpu-cloud?ref=qdna", "url_brev": null, "rel_attr": "external nofollow noopener sponsored", "referral_actif": true, "serveur_site_ref": "rtx-5090-1x", "date_releve": "2026-10-21" }, { "id": "nebius-l40s-1x", "marque": "NVIDIA", "famille": "Ada Lovelace", "carte_gpu": "NVIDIA L40S", "architecture": "Ada Lovelace", "nb_gpu": 1, "vram_gpu_go": 48, "vram_totale_go": 48, "type_vram": "GDDR6 ECC", "bp_memoire_gpu_tos": 0.86, "bp_memoire_totale_tos": 0.86, "tflops_fp16": 362, "tflops_fp8": 733, "fournisseur_id": "nebius", "fournisseur": "Nebius AI Cloud", "dispo_nebius": true, "dispo_brev": true, "mode_tarif_fr": "À la demande (Préemptible : 0,74 $/h)", "mode_tarif_en": "On-demand (Preemptible: $0.74/h)", "machine_ref": "gpu-l40s-d (1gpu-8vcpu-32gb)", "cpu_modele": "Intel Xeon Gold 6338 Ice Lake (ou EPYC 9654)", "vcpu_total": 8, "ram_systeme_go": 32, "stockage_nvme_to": 1.0, "interconnexion": "PCIe Gen4 x16", "reseau_cluster": "Ethernet 100 Gbps", "prix_heure_gpu_usd": 1.55, "prix_heure_gpu_eur_ht": 1.36, "prix_heure_serveur_usd": 1.55, "prix_heure_serveur_eur_ht": 1.36, "cout_mensuel_730h_serveur_eur_ht": 993, "cout_mensuel_730h_serveur_usd": 1132, "prix_spot_gpu_usd": 0.74, "prix_spot_gpu_eur_ht": 0.65, "vram_par_euro_heure": 35.3, "url_source": "https://nebius.com/prices", "url_rent_fr": "https://console.eu.nebius.com/", "url_rent_en": "https://console.eu.nebius.com/", "url_brev": "https://brev.nvidia.com/environment/new/public?gpu=L40S", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "dgx-spark", "date_releve": "2026-10-21" }, { "id": "ovh-l40s-1x", "marque": "NVIDIA", "famille": "Ada Lovelace", "carte_gpu": "NVIDIA L40S", "architecture": "Ada Lovelace", "nb_gpu": 1, "vram_gpu_go": 48, "vram_totale_go": 48, "type_vram": "GDDR6 ECC", "bp_memoire_gpu_tos": 0.86, "bp_memoire_totale_tos": 0.86, "tflops_fp16": 362, "tflops_fp8": 733, "fournisseur_id": "ovhcloud", "fournisseur": "OVHcloud", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Public Cloud France (Gravelines)", "mode_tarif_en": "France Public Cloud (Gravelines)", "machine_ref": "l40s-90", "cpu_modele": "AMD EPYC", "vcpu_total": 15, "ram_systeme_go": 90, "stockage_nvme_to": 0.4, "interconnexion": "PCIe Gen4 x16", "reseau_cluster": "Ethernet 25 Gbps", "prix_heure_gpu_usd": 1.6, "prix_heure_gpu_eur_ht": 1.4, "prix_heure_serveur_usd": 1.6, "prix_heure_serveur_eur_ht": 1.4, "cout_mensuel_730h_serveur_eur_ht": 1022, "cout_mensuel_730h_serveur_usd": 1168, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 34.3, "url_source": "https://www.ovhcloud.com/fr/public-cloud/prices/", "url_rent_fr": "https://www.ovhcloud.com/fr/public-cloud/gpu/", "url_rent_en": "https://www.ovhcloud.com/en/public-cloud/gpu/", "url_brev": null, "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "dgx-spark", "date_releve": "2026-10-21" }, { "id": "inteltiber-gaudi2-8x", "marque": "Intel", "famille": "Gaudi 2", "carte_gpu": "Intel Gaudi 2 HL-225 OAM", "architecture": "Gaudi 2", "nb_gpu": 8, "vram_gpu_go": 96, "vram_totale_go": 768, "type_vram": "HBM2e", "bp_memoire_gpu_tos": 2.45, "bp_memoire_totale_tos": 19.6, "tflops_fp16": 432, "tflops_fp8": 865, "fournisseur_id": "inteltiber", "fournisseur": "Intel Tiber AI Cloud", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Nœud bare-metal 8x officiel Intel", "mode_tarif_en": "Official Intel 8x bare-metal node", "machine_ref": "UBB 8x Intel Gaudi 2 HL-225", "cpu_modele": "2x Intel Xeon Platinum 8380 Ice Lake", "vcpu_total": 160, "ram_systeme_go": 1024, "stockage_nvme_to": 30.72, "interconnexion": "21x 100GbE RDMA RoCEv2 sur puce", "reseau_cluster": "24x 100GbE RoCEv2 intégrés", "prix_heure_gpu_usd": 1.65, "prix_heure_gpu_eur_ht": 1.45, "prix_heure_serveur_usd": 13.2, "prix_heure_serveur_eur_ht": 11.6, "cout_mensuel_730h_serveur_eur_ht": 8468, "cout_mensuel_730h_serveur_usd": 9636, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 66.2, "url_source": "https://cloud.intel.com/compute-instances", "url_rent_fr": "https://cloud.intel.com/compute-instances", "url_rent_en": "https://cloud.intel.com/compute-instances", "url_brev": null, "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "rtx-pro-6000-8x", "date_releve": "2026-10-21" }, { "id": "tensorwave-mi300x-8x", "marque": "AMD", "famille": "Instinct CDNA 3", "carte_gpu": "AMD Instinct MI300X OAM", "architecture": "CDNA 3", "nb_gpu": 8, "vram_gpu_go": 192, "vram_totale_go": 1536, "type_vram": "HBM3", "bp_memoire_gpu_tos": 5.3, "bp_memoire_totale_tos": 42.4, "tflops_fp16": 1307, "tflops_fp8": 2615, "fournisseur_id": "tensorwave", "fournisseur": "TensorWave", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Nœud bare-metal 8x (1 536 Go HBM3)", "mode_tarif_en": "8x Bare-metal node (1,536 GB HBM3)", "machine_ref": "TensorWave 8x MI300X", "cpu_modele": "2x AMD EPYC Genoa 9654 (192 cœurs / 384 threads)", "vcpu_total": 384, "ram_systeme_go": 2304, "stockage_nvme_to": 17.28, "interconnexion": "AMD Infinity Fabric 896 Go/s", "reseau_cluster": "8x 400GbE RoCEv2 (3,2 Tbit/s)", "prix_heure_gpu_usd": 1.71, "prix_heure_gpu_eur_ht": 1.5, "prix_heure_serveur_usd": 13.68, "prix_heure_serveur_eur_ht": 12.0, "cout_mensuel_730h_serveur_eur_ht": 8760, "cout_mensuel_730h_serveur_usd": 9986, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 128.0, "url_source": "https://tensorwave.com/products/accelerators/amd-mi300x", "url_rent_fr": "https://cloud.tensorwave.com/", "url_rent_en": "https://cloud.tensorwave.com/", "url_brev": null, "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "mi300x-8x", "date_releve": "2026-10-21" }, { "id": "nebius-rtxpro6000-1x", "marque": "NVIDIA", "famille": "Blackwell", "carte_gpu": "NVIDIA RTX PRO 6000 Blackwell Server Edition", "architecture": "Blackwell", "nb_gpu": 1, "vram_gpu_go": 96, "vram_totale_go": 96, "type_vram": "GDDR7 ECC", "bp_memoire_gpu_tos": 1.79, "bp_memoire_totale_tos": 1.79, "tflops_fp16": 500, "tflops_fp8": 1000, "fournisseur_id": "nebius", "fournisseur": "Nebius AI Cloud", "dispo_nebius": true, "dispo_brev": true, "mode_tarif_fr": "À la demande (Spot : 0,79 $/h)", "mode_tarif_en": "On-demand (Spot: $0.79/h)", "machine_ref": "gpu-rtx6000 (1gpu-24vcpu-218gb)", "cpu_modele": "Intel Xeon 6776P Granite Rapids", "vcpu_total": 24, "ram_systeme_go": 218, "stockage_nvme_to": 2.0, "interconnexion": "PCIe Gen5 x16", "reseau_cluster": "NVIDIA BlueField-3 400 Gbps", "prix_heure_gpu_usd": 1.8, "prix_heure_gpu_eur_ht": 1.58, "prix_heure_serveur_usd": 1.8, "prix_heure_serveur_eur_ht": 1.58, "cout_mensuel_730h_serveur_eur_ht": 1153, "cout_mensuel_730h_serveur_usd": 1314, "prix_spot_gpu_usd": 0.79, "prix_spot_gpu_eur_ht": 0.69, "vram_par_euro_heure": 60.8, "url_source": "https://nebius.com/prices", "url_rent_fr": "https://console.eu.nebius.com/", "url_rent_en": "https://console.eu.nebius.com/", "url_brev": "https://brev.nvidia.com/environment/new", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "rtx-pro-6000-1x", "date_releve": "2026-10-21" }, { "id": "nebius-rtxpro6000-8x", "marque": "NVIDIA", "famille": "Blackwell", "carte_gpu": "NVIDIA RTX PRO 6000 Blackwell Server Edition", "architecture": "Blackwell", "nb_gpu": 8, "vram_gpu_go": 96, "vram_totale_go": 768, "type_vram": "GDDR7 ECC", "bp_memoire_gpu_tos": 1.79, "bp_memoire_totale_tos": 14.32, "tflops_fp16": 500, "tflops_fp8": 1000, "fournisseur_id": "nebius", "fournisseur": "Nebius AI Cloud", "dispo_nebius": true, "dispo_brev": true, "mode_tarif_fr": "Nœud 8x à la demande (768 Go GDDR7)", "mode_tarif_en": "8x On-demand node (768 GB GDDR7)", "machine_ref": "gpu-rtx6000 (8gpu-192vcpu-1744gb)", "cpu_modele": "2x Intel Xeon 6776P Granite Rapids", "vcpu_total": 192, "ram_systeme_go": 1744, "stockage_nvme_to": 16.0, "interconnexion": "PCIe Gen5 x16", "reseau_cluster": "NVIDIA BlueField-3 400 Gbps", "prix_heure_gpu_usd": 1.8, "prix_heure_gpu_eur_ht": 1.58, "prix_heure_serveur_usd": 14.4, "prix_heure_serveur_eur_ht": 12.64, "cout_mensuel_730h_serveur_eur_ht": 9227, "cout_mensuel_730h_serveur_usd": 10512, "prix_spot_gpu_usd": 0.79, "prix_spot_gpu_eur_ht": 0.69, "vram_par_euro_heure": 60.8, "url_source": "https://nebius.com/prices", "url_rent_fr": "https://console.eu.nebius.com/", "url_rent_en": "https://console.eu.nebius.com/", "url_brev": "https://brev.nvidia.com/environment/new", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "rtx-pro-6000-8x", "date_releve": "2026-10-21" }, { "id": "vultr-mi300x-8x", "marque": "AMD", "famille": "Instinct CDNA 3", "carte_gpu": "AMD Instinct MI300X OAM", "architecture": "CDNA 3", "nb_gpu": 8, "vram_gpu_go": 192, "vram_totale_go": 1536, "type_vram": "HBM3", "bp_memoire_gpu_tos": 5.3, "bp_memoire_totale_tos": 42.4, "tflops_fp16": 1307, "tflops_fp8": 2615, "fournisseur_id": "vultr", "fournisseur": "Vultr Cloud GPU", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Nœud 8x Spot / Réservé (1 536 Go HBM3)", "mode_tarif_en": "8x Spot / Reserved node (1,536 GB HBM3)", "machine_ref": "Vultr 8x MI300X", "cpu_modele": "2x AMD EPYC Genoa (256 vCPU)", "vcpu_total": 256, "ram_systeme_go": 2048, "stockage_nvme_to": 15.36, "interconnexion": "AMD Infinity Fabric 896 Go/s", "reseau_cluster": "8x 400GbE RoCEv2 (3,2 Tbit/s)", "prix_heure_gpu_usd": 1.85, "prix_heure_gpu_eur_ht": 1.62, "prix_heure_serveur_usd": 14.8, "prix_heure_serveur_eur_ht": 12.96, "cout_mensuel_730h_serveur_eur_ht": 9461, "cout_mensuel_730h_serveur_usd": 10804, "prix_spot_gpu_usd": 1.75, "prix_spot_gpu_eur_ht": 1.53, "vram_par_euro_heure": 118.5, "url_source": "https://www.vultr.com/products/cloud-gpu/amd-mi300x/", "url_rent_fr": "https://www.vultr.com/products/cloud-gpu/?ref=qdna", "url_rent_en": "https://www.vultr.com/products/cloud-gpu/?ref=qdna", "url_brev": null, "rel_attr": "external nofollow noopener sponsored", "referral_actif": true, "serveur_site_ref": "mi300x-8x", "date_releve": "2026-10-21" }, { "id": "vultr-mi325x-8x", "marque": "AMD", "famille": "Instinct CDNA 3", "carte_gpu": "AMD Instinct MI325X OAM", "architecture": "CDNA 3", "nb_gpu": 8, "vram_gpu_go": 256, "vram_totale_go": 2048, "type_vram": "HBM3e", "bp_memoire_gpu_tos": 6.0, "bp_memoire_totale_tos": 48.0, "tflops_fp16": 1307, "tflops_fp8": 2615, "fournisseur_id": "vultr", "fournisseur": "Vultr Cloud GPU", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Nœud 8x (2 048 Go HBM3e, tarif de départ)", "mode_tarif_en": "8x node (2,048 GB HBM3e, starting rate)", "machine_ref": "Vultr 8x MI325X Bare Metal", "cpu_modele": "2x AMD EPYC Turin / Genoa (256 vCPU)", "vcpu_total": 256, "ram_systeme_go": 3072, "stockage_nvme_to": 15.36, "interconnexion": "AMD Infinity Fabric 896 Go/s", "reseau_cluster": "8x 400GbE RoCEv2 (3,2 Tbit/s)", "prix_heure_gpu_usd": 2.0, "prix_heure_gpu_eur_ht": 1.75, "prix_heure_serveur_usd": 16.0, "prix_heure_serveur_eur_ht": 14.0, "cout_mensuel_730h_serveur_eur_ht": 10220, "cout_mensuel_730h_serveur_usd": 11680, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 146.3, "url_source": "https://getdeploying.com/gpus/amd-mi325x", "url_rent_fr": "https://www.vultr.com/products/cloud-gpu/?ref=qdna", "url_rent_en": "https://www.vultr.com/products/cloud-gpu/?ref=qdna", "url_brev": null, "rel_attr": "external nofollow noopener sponsored", "referral_actif": true, "serveur_site_ref": "mi325x-8x", "date_releve": "2026-10-21" }, { "id": "tensorwave-mi325x-8x", "marque": "AMD", "famille": "Instinct CDNA 3", "carte_gpu": "AMD Instinct MI325X OAM", "architecture": "CDNA 3", "nb_gpu": 8, "vram_gpu_go": 256, "vram_totale_go": 2048, "type_vram": "HBM3e", "bp_memoire_gpu_tos": 6.0, "bp_memoire_totale_tos": 48.0, "tflops_fp16": 1307, "tflops_fp8": 2615, "fournisseur_id": "tensorwave", "fournisseur": "TensorWave", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Nœud bare-metal 8x refroidissement liquide", "mode_tarif_en": "8x Direct-liquid-cooled bare-metal node", "machine_ref": "TensorWave 8x MI325X DLC", "cpu_modele": "2x AMD EPYC Turin 9575F (128 cœurs / 256 threads)", "vcpu_total": 256, "ram_systeme_go": 3072, "stockage_nvme_to": 17.15, "interconnexion": "AMD Infinity Fabric 8x 128 Go/s (896 Go/s)", "reseau_cluster": "8x 400GbE RoCEv2 (3,2 Tbit/s)", "prix_heure_gpu_usd": 2.25, "prix_heure_gpu_eur_ht": 1.97, "prix_heure_serveur_usd": 18.0, "prix_heure_serveur_eur_ht": 15.76, "cout_mensuel_730h_serveur_eur_ht": 11505, "cout_mensuel_730h_serveur_usd": 13140, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 129.9, "url_source": "https://tensorwave.com/products/accelerators/amd-mi325x", "url_rent_fr": "https://cloud.tensorwave.com/", "url_rent_en": "https://cloud.tensorwave.com/", "url_brev": null, "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "mi325x-8x", "date_releve": "2026-10-21" }, { "id": "lambda-gh200-1x", "marque": "NVIDIA", "famille": "Hopper", "carte_gpu": "NVIDIA GH200 Grace Hopper Superchip", "architecture": "Grace Hopper", "nb_gpu": 1, "vram_gpu_go": 96, "vram_totale_go": 96, "type_vram": "HBM3 (+ 432 Go LPDDR5X)", "bp_memoire_gpu_tos": 4.0, "bp_memoire_totale_tos": 4.0, "tflops_fp16": 989, "tflops_fp8": 1979, "fournisseur_id": "lambda", "fournisseur": "Lambda", "dispo_nebius": false, "dispo_brev": true, "mode_tarif_fr": "À la demande (disponible sur Brev)", "mode_tarif_en": "On-demand (available on Brev)", "machine_ref": "gpu_1x_gh200", "cpu_modele": "NVIDIA Grace Arm Neoverse V2 (64 vCPU)", "vcpu_total": 64, "ram_systeme_go": 432, "stockage_nvme_to": 4.0, "interconnexion": "NVLink-C2C (900 Go/s CPU-GPU)", "reseau_cluster": "InfiniBand NDR 400 Gbps", "prix_heure_gpu_usd": 2.29, "prix_heure_gpu_eur_ht": 2.01, "prix_heure_serveur_usd": 2.29, "prix_heure_serveur_eur_ht": 2.01, "cout_mensuel_730h_serveur_eur_ht": 1467, "cout_mensuel_730h_serveur_usd": 1672, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 47.8, "url_source": "https://lambda.ai/service/gpu-cloud", "url_rent_fr": "https://cloud.lambdalabs.com/instances", "url_rent_en": "https://cloud.lambdalabs.com/instances", "url_brev": "https://brev.nvidia.com/environment/new", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "rtx-pro-6000-1x", "date_releve": "2026-10-21" }, { "id": "runpod-mi300x-1x", "marque": "AMD", "famille": "Instinct CDNA 3", "carte_gpu": "AMD Instinct MI300X OAM", "architecture": "CDNA 3", "nb_gpu": 1, "vram_gpu_go": 192, "vram_totale_go": 192, "type_vram": "HBM3", "bp_memoire_gpu_tos": 5.3, "bp_memoire_totale_tos": 5.3, "tflops_fp16": 1307, "tflops_fp8": 2615, "fournisseur_id": "runpod", "fournisseur": "RunPod", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Secure Cloud à la seconde (1x à 8x GPU)", "mode_tarif_en": "Per-second Secure Cloud (1x to 8x GPUs)", "machine_ref": "RunPod MI300X 192GB Pod", "cpu_modele": "AMD EPYC Genoa", "vcpu_total": 24, "ram_systeme_go": 120, "stockage_nvme_to": 1.0, "interconnexion": "AMD Infinity Fabric OAM", "reseau_cluster": "RoCEv2 400 Gbps", "prix_heure_gpu_usd": 2.39, "prix_heure_gpu_eur_ht": 2.1, "prix_heure_serveur_usd": 2.39, "prix_heure_serveur_eur_ht": 2.1, "cout_mensuel_730h_serveur_eur_ht": 1533, "cout_mensuel_730h_serveur_usd": 1745, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 91.4, "url_source": "https://www.runpod.io/gpu-models/mi300x", "url_rent_fr": "https://www.runpod.io/console/gpu-cloud?ref=qdna", "url_rent_en": "https://www.runpod.io/console/gpu-cloud?ref=qdna", "url_brev": null, "rel_attr": "external nofollow noopener sponsored", "referral_actif": true, "serveur_site_ref": "mi300x-8x", "date_releve": "2026-10-21" }, { "id": "hyperstack-h100-1x", "marque": "NVIDIA", "famille": "Hopper", "carte_gpu": "NVIDIA H100 PCIe", "architecture": "Hopper", "nb_gpu": 1, "vram_gpu_go": 80, "vram_totale_go": 80, "type_vram": "HBM3", "bp_memoire_gpu_tos": 2.0, "bp_memoire_totale_tos": 2.0, "tflops_fp16": 756, "tflops_fp8": 1513, "fournisseur_id": "hyperstack", "fournisseur": "Hyperstack (NexGen Cloud)", "dispo_nebius": false, "dispo_brev": true, "mode_tarif_fr": "À la demande (disponible sur Brev)", "mode_tarif_en": "On-demand (available on Brev)", "machine_ref": "H100-80G-PCIe", "cpu_modele": "AMD EPYC Genoa", "vcpu_total": 28, "ram_systeme_go": 180, "stockage_nvme_to": 1.5, "interconnexion": "PCIe Gen5 x16", "reseau_cluster": "Ethernet 100 Gbps", "prix_heure_gpu_usd": 2.4, "prix_heure_gpu_eur_ht": 2.1, "prix_heure_serveur_usd": 2.4, "prix_heure_serveur_eur_ht": 2.1, "cout_mensuel_730h_serveur_eur_ht": 1533, "cout_mensuel_730h_serveur_usd": 1752, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 38.1, "url_source": "https://www.hyperstack.cloud/gpu-pricing", "url_rent_fr": "https://console.hyperstack.cloud/deploy-virtual-machine", "url_rent_en": "https://console.hyperstack.cloud/deploy-virtual-machine", "url_brev": "https://brev.nvidia.com/environment/new/public?gpu=H100", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "h100-8x", "date_releve": "2026-10-21" }, { "id": "denvr-h100-8x", "marque": "NVIDIA", "famille": "Hopper", "carte_gpu": "NVIDIA HGX H100 SXM5", "architecture": "Hopper", "nb_gpu": 8, "vram_gpu_go": 80, "vram_totale_go": 640, "type_vram": "HBM3", "bp_memoire_gpu_tos": 3.35, "bp_memoire_totale_tos": 26.8, "tflops_fp16": 989, "tflops_fp8": 1979, "fournisseur_id": "denvr", "fournisseur": "Denvr Dataworks", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Nœud 8x SXM5 (refroidissement par immersion)", "mode_tarif_en": "8x SXM5 node (immersion cooling)", "machine_ref": "Denvr H100-SXM5-8x", "cpu_modele": "2x Intel Xeon Sapphire Rapids", "vcpu_total": 208, "ram_systeme_go": 1024, "stockage_nvme_to": 22.8, "interconnexion": "NVLink 4 (900 Go/s)", "reseau_cluster": "InfiniBand NDR 3,2 Tbit/s", "prix_heure_gpu_usd": 2.45, "prix_heure_gpu_eur_ht": 2.15, "prix_heure_serveur_usd": 19.6, "prix_heure_serveur_eur_ht": 17.2, "cout_mensuel_730h_serveur_eur_ht": 12556, "cout_mensuel_730h_serveur_usd": 14308, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 37.2, "url_source": "https://www.denvrdata.com/pricing", "url_rent_fr": "https://www.denvrdata.com/pricing", "url_rent_en": "https://www.denvrdata.com/pricing", "url_brev": null, "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "h100-8x", "date_releve": "2026-10-21" }, { "id": "runpod-rtxpro6000-1x", "marque": "NVIDIA", "famille": "Blackwell", "carte_gpu": "NVIDIA RTX PRO 6000 Blackwell Server Edition", "architecture": "Blackwell", "nb_gpu": 1, "vram_gpu_go": 96, "vram_totale_go": 96, "type_vram": "GDDR7 ECC", "bp_memoire_gpu_tos": 1.79, "bp_memoire_totale_tos": 1.79, "tflops_fp16": 500, "tflops_fp8": 1000, "fournisseur_id": "runpod", "fournisseur": "RunPod", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Secure Cloud (Community : 1,64 $/h)", "mode_tarif_en": "Secure Cloud (Community: $1.64/h)", "machine_ref": "RunPod RTX PRO 6000", "cpu_modele": "AMD EPYC Genoa / Turin", "vcpu_total": 16, "ram_systeme_go": 188, "stockage_nvme_to": 1.0, "interconnexion": "PCIe Gen5 x16", "reseau_cluster": "Ethernet 100 Gbps", "prix_heure_gpu_usd": 2.49, "prix_heure_gpu_eur_ht": 2.18, "prix_heure_serveur_usd": 2.49, "prix_heure_serveur_eur_ht": 2.18, "cout_mensuel_730h_serveur_eur_ht": 1591, "cout_mensuel_730h_serveur_usd": 1818, "prix_spot_gpu_usd": 1.64, "prix_spot_gpu_eur_ht": 1.44, "vram_par_euro_heure": 44.0, "url_source": "https://www.runpod.io/pricing", "url_rent_fr": "https://www.runpod.io/console/gpu-cloud?ref=qdna", "url_rent_en": "https://www.runpod.io/console/gpu-cloud?ref=qdna", "url_brev": null, "rel_attr": "external nofollow noopener sponsored", "referral_actif": true, "serveur_site_ref": "rtx-pro-6000-1x", "date_releve": "2026-10-21" }, { "id": "tensorwave-mi355x-8x", "marque": "AMD", "famille": "Instinct CDNA 4", "carte_gpu": "AMD Instinct MI355X OAM", "architecture": "CDNA 4", "nb_gpu": 8, "vram_gpu_go": 288, "vram_totale_go": 2304, "type_vram": "HBM3e", "bp_memoire_gpu_tos": 8.0, "bp_memoire_totale_tos": 64.0, "tflops_fp16": 2500, "tflops_fp8": 5000, "fournisseur_id": "tensorwave", "fournisseur": "TensorWave", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Nœud bare-metal 8x (2 304 Go HBM3e)", "mode_tarif_en": "8x Bare-metal node (2,304 GB HBM3e)", "machine_ref": "TensorWave 8x MI355X DLC", "cpu_modele": "2x AMD EPYC Turin 9575F (128 cœurs / 256 threads)", "vcpu_total": 256, "ram_systeme_go": 3072, "stockage_nvme_to": 17.2, "interconnexion": "AMD Infinity Fabric (1,0 To/s par GPU)", "reseau_cluster": "8x 400GbE RoCEv2 (3,2 Tbit/s)", "prix_heure_gpu_usd": 2.95, "prix_heure_gpu_eur_ht": 2.59, "prix_heure_serveur_usd": 23.6, "prix_heure_serveur_eur_ht": 20.72, "cout_mensuel_730h_serveur_eur_ht": 15126, "cout_mensuel_730h_serveur_usd": 17228, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 111.2, "url_source": "https://tensorwave.com/", "url_rent_fr": "https://cloud.tensorwave.com/", "url_rent_en": "https://cloud.tensorwave.com/", "url_brev": null, "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "mi325x-8x", "date_releve": "2026-10-21" }, { "id": "hotaisle-mi300x-1x", "marque": "AMD", "famille": "Instinct CDNA 3", "carte_gpu": "AMD Instinct MI300X OAM", "architecture": "CDNA 3", "nb_gpu": 1, "vram_gpu_go": 192, "vram_totale_go": 192, "type_vram": "HBM3", "bp_memoire_gpu_tos": 5.3, "bp_memoire_totale_tos": 5.3, "tflops_fp16": 1307, "tflops_fp8": 2615, "fournisseur_id": "hotaisle", "fournisseur": "Hot Aisle", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "VM sur Dell PowerEdge XE9680 (8x BM : 3,39 $/h)", "mode_tarif_en": "VM on Dell PowerEdge XE9680 (8x BM: $3.39/h)", "machine_ref": "Dell PowerEdge XE9680 (1x MI300X)", "cpu_modele": "Intel Xeon Platinum 8468 Sapphire Rapids", "vcpu_total": 13, "ram_systeme_go": 224, "stockage_nvme_to": 15.36, "interconnexion": "AMD Infinity Fabric OAM", "reseau_cluster": "RoCEv2 400 Gbps", "prix_heure_gpu_usd": 2.99, "prix_heure_gpu_eur_ht": 2.62, "prix_heure_serveur_usd": 2.99, "prix_heure_serveur_eur_ht": 2.62, "cout_mensuel_730h_serveur_eur_ht": 1913, "cout_mensuel_730h_serveur_usd": 2183, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 73.3, "url_source": "https://hotaisle.xyz/pricing/", "url_rent_fr": "https://hotaisle.xyz/pricing/", "url_rent_en": "https://hotaisle.xyz/pricing/", "url_brev": null, "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "mi300x-8x", "date_releve": "2026-10-21" }, { "id": "ovh-h100-1x", "marque": "NVIDIA", "famille": "Hopper", "carte_gpu": "NVIDIA H100 PCIe", "architecture": "Hopper", "nb_gpu": 1, "vram_gpu_go": 80, "vram_totale_go": 80, "type_vram": "HBM3", "bp_memoire_gpu_tos": 2.0, "bp_memoire_totale_tos": 2.0, "tflops_fp16": 756, "tflops_fp8": 1513, "fournisseur_id": "ovhcloud", "fournisseur": "OVHcloud", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Public Cloud France (1x à 4x GPU)", "mode_tarif_en": "France Public Cloud (1x to 4x GPUs)", "machine_ref": "h100-380", "cpu_modele": "AMD EPYC Genoa", "vcpu_total": 30, "ram_systeme_go": 380, "stockage_nvme_to": 3.84, "interconnexion": "PCIe Gen5 x16", "reseau_cluster": "3,84 To NVMe Passthrough + 25 Gbps", "prix_heure_gpu_usd": 3.19, "prix_heure_gpu_eur_ht": 2.8, "prix_heure_serveur_usd": 3.19, "prix_heure_serveur_eur_ht": 2.8, "cout_mensuel_730h_serveur_eur_ht": 2044, "cout_mensuel_730h_serveur_usd": 2329, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 28.6, "url_source": "https://www.ovhcloud.com/fr/public-cloud/prices/", "url_rent_fr": "https://www.ovhcloud.com/fr/public-cloud/gpu/", "url_rent_en": "https://www.ovhcloud.com/en/public-cloud/gpu/", "url_brev": null, "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "h100-8x", "date_releve": "2026-10-21" }, { "id": "crusoe-mi300x-8x", "marque": "AMD", "famille": "Instinct CDNA 3", "carte_gpu": "AMD Instinct MI300X OAM", "architecture": "CDNA 3", "nb_gpu": 8, "vram_gpu_go": 192, "vram_totale_go": 1536, "type_vram": "HBM3", "bp_memoire_gpu_tos": 5.3, "bp_memoire_totale_tos": 42.4, "tflops_fp16": 1307, "tflops_fp8": 2615, "fournisseur_id": "crusoe", "fournisseur": "Crusoe Cloud", "dispo_nebius": false, "dispo_brev": true, "mode_tarif_fr": "Nœud 8x à la demande (Spot : 1,20 $/h)", "mode_tarif_en": "8x On-demand node (Spot: $1.20/h)", "machine_ref": "mi300x-192gb-ib.8x", "cpu_modele": "2x AMD EPYC Genoa (240 vCPU)", "vcpu_total": 240, "ram_systeme_go": 2048, "stockage_nvme_to": 24.0, "interconnexion": "AMD Infinity Fabric 896 Go/s", "reseau_cluster": "InfiniBand NDR 400 Gbps (3,2 Tbit/s)", "prix_heure_gpu_usd": 3.45, "prix_heure_gpu_eur_ht": 3.03, "prix_heure_serveur_usd": 27.6, "prix_heure_serveur_eur_ht": 24.24, "cout_mensuel_730h_serveur_eur_ht": 17695, "cout_mensuel_730h_serveur_usd": 20148, "prix_spot_gpu_usd": 1.2, "prix_spot_gpu_eur_ht": 1.05, "vram_par_euro_heure": 63.4, "url_source": "https://www.crusoe.ai/cloud/pricing", "url_rent_fr": "https://console.crusoecloud.com/", "url_rent_en": "https://console.crusoecloud.com/", "url_brev": "https://brev.nvidia.com/environment/new", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "mi300x-8x", "date_releve": "2026-10-21" }, { "id": "scaleway-h100-sxm-8x", "marque": "NVIDIA", "famille": "Hopper", "carte_gpu": "NVIDIA HGX H100 SXM5", "architecture": "Hopper", "nb_gpu": 8, "vram_gpu_go": 80, "vram_totale_go": 640, "type_vram": "HBM3", "bp_memoire_gpu_tos": 3.35, "bp_memoire_totale_tos": 26.8, "tflops_fp16": 989, "tflops_fp8": 1979, "fournisseur_id": "scaleway", "fournisseur": "Scaleway (Groupe iliad)", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Nœud 8x souverain FR (Paris fr-par-2)", "mode_tarif_en": "8x FR sovereign node (Paris fr-par-2)", "machine_ref": "H100-SXM-8-80G", "cpu_modele": "2x AMD EPYC Genoa", "vcpu_total": 128, "ram_systeme_go": 960, "stockage_nvme_to": 12.8, "interconnexion": "NVLink 4 (900 Go/s)", "reseau_cluster": "InfiniBand NDR 400 Gbps", "prix_heure_gpu_usd": 3.61, "prix_heure_gpu_eur_ht": 3.17, "prix_heure_serveur_usd": 28.88, "prix_heure_serveur_eur_ht": 25.36, "cout_mensuel_730h_serveur_eur_ht": 18513, "cout_mensuel_730h_serveur_usd": 21082, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 25.2, "url_source": "https://www.scaleway.com/en/pricing/gpu/?zone=fr-par-2", "url_rent_fr": "https://console.scaleway.com/gpu-instances/instances", "url_rent_en": "https://console.scaleway.com/gpu-instances/instances", "url_brev": null, "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "h100-8x", "date_releve": "2026-10-21" }, { "id": "crusoe-h200-8x", "marque": "NVIDIA", "famille": "Hopper", "carte_gpu": "NVIDIA HGX H200 SXM5", "architecture": "Hopper", "nb_gpu": 8, "vram_gpu_go": 141, "vram_totale_go": 1128, "type_vram": "HBM3e", "bp_memoire_gpu_tos": 4.8, "bp_memoire_totale_tos": 38.4, "tflops_fp16": 989, "tflops_fp8": 1979, "fournisseur_id": "crusoe", "fournisseur": "Crusoe Cloud", "dispo_nebius": false, "dispo_brev": true, "mode_tarif_fr": "Nœud 8x à la demande (disponible sur Brev)", "mode_tarif_en": "8x On-demand node (available on Brev)", "machine_ref": "h200-141gb-sxm-ib.8x", "cpu_modele": "2x Intel Xeon Sapphire Rapids", "vcpu_total": 176, "ram_systeme_go": 1920, "stockage_nvme_to": 24.0, "interconnexion": "NVLink 4 (900 Go/s)", "reseau_cluster": "InfiniBand NDR 400 Gbps (3,2 Tbit/s)", "prix_heure_gpu_usd": 4.29, "prix_heure_gpu_eur_ht": 3.76, "prix_heure_serveur_usd": 34.32, "prix_heure_serveur_eur_ht": 30.08, "cout_mensuel_730h_serveur_eur_ht": 21958, "cout_mensuel_730h_serveur_usd": 25054, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 37.5, "url_source": "https://www.crusoe.ai/cloud/pricing", "url_rent_fr": "https://console.crusoecloud.com/", "url_rent_en": "https://console.crusoecloud.com/", "url_brev": "https://brev.nvidia.com/environment/new/public?gpu=H200", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "dgx-h200", "date_releve": "2026-10-21" }, { "id": "crusoe-b200-8x", "marque": "NVIDIA", "famille": "Blackwell", "carte_gpu": "NVIDIA HGX B200 SXM6", "architecture": "Blackwell", "nb_gpu": 8, "vram_gpu_go": 180, "vram_totale_go": 1440, "type_vram": "HBM3e", "bp_memoire_gpu_tos": 8.0, "bp_memoire_totale_tos": 64.0, "tflops_fp16": 2250, "tflops_fp8": 4500, "fournisseur_id": "crusoe", "fournisseur": "Crusoe Cloud", "dispo_nebius": false, "dispo_brev": true, "mode_tarif_fr": "Nœud 8x à la demande (disponible sur Brev)", "mode_tarif_en": "8x On-demand node (available on Brev)", "machine_ref": "b200-180gb-sxm-ib.8x", "cpu_modele": "2x Intel Xeon Emerald Rapids", "vcpu_total": 192, "ram_systeme_go": 2048, "stockage_nvme_to": 24.0, "interconnexion": "NVLink 5 (1,8 To/s)", "reseau_cluster": "InfiniBand NDR 400 Gbps (3,2 Tbit/s)", "prix_heure_gpu_usd": 4.49, "prix_heure_gpu_eur_ht": 3.94, "prix_heure_serveur_usd": 35.92, "prix_heure_serveur_eur_ht": 31.52, "cout_mensuel_730h_serveur_eur_ht": 23010, "cout_mensuel_730h_serveur_usd": 26222, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 45.7, "url_source": "https://www.crusoe.ai/cloud/pricing", "url_rent_fr": "https://console.crusoecloud.com/", "url_rent_en": "https://console.crusoecloud.com/", "url_brev": "https://brev.nvidia.com/environment/new/public?gpu=B200", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "dgx-b200", "date_releve": "2026-10-21" }, { "id": "nebius-h100-1x", "marque": "NVIDIA", "famille": "Hopper", "carte_gpu": "NVIDIA HGX H100 SXM5", "architecture": "Hopper", "nb_gpu": 1, "vram_gpu_go": 80, "vram_totale_go": 80, "type_vram": "HBM3", "bp_memoire_gpu_tos": 3.35, "bp_memoire_totale_tos": 3.35, "tflops_fp16": 989, "tflops_fp8": 1979, "fournisseur_id": "nebius", "fournisseur": "Nebius AI Cloud", "dispo_nebius": true, "dispo_brev": true, "mode_tarif_fr": "À la demande (Spot : 0,79 $/h)", "mode_tarif_en": "On-demand (Spot: $0.79/h)", "machine_ref": "gpu-h100-sxm (1gpu-16vcpu-200gb)", "cpu_modele": "Intel Xeon Platinum 8468 Sapphire Rapids", "vcpu_total": 16, "ram_systeme_go": 200, "stockage_nvme_to": 3.2, "interconnexion": "NVLink 4 (900 Go/s)", "reseau_cluster": "InfiniBand NDR 400 Gbps (ConnectX-7)", "prix_heure_gpu_usd": 4.5, "prix_heure_gpu_eur_ht": 3.95, "prix_heure_serveur_usd": 4.5, "prix_heure_serveur_eur_ht": 3.95, "cout_mensuel_730h_serveur_eur_ht": 2884, "cout_mensuel_730h_serveur_usd": 3285, "prix_spot_gpu_usd": 0.79, "prix_spot_gpu_eur_ht": 0.69, "vram_par_euro_heure": 20.3, "url_source": "https://nebius.com/prices", "url_rent_fr": "https://console.eu.nebius.com/", "url_rent_en": "https://console.eu.nebius.com/", "url_brev": "https://brev.nvidia.com/environment/new/public?gpu=H100", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "h100-8x", "date_releve": "2026-10-21" }, { "id": "nebius-h100-8x", "marque": "NVIDIA", "famille": "Hopper", "carte_gpu": "NVIDIA HGX H100 SXM5", "architecture": "Hopper", "nb_gpu": 8, "vram_gpu_go": 80, "vram_totale_go": 640, "type_vram": "HBM3", "bp_memoire_gpu_tos": 3.35, "bp_memoire_totale_tos": 26.8, "tflops_fp16": 989, "tflops_fp8": 1979, "fournisseur_id": "nebius", "fournisseur": "Nebius AI Cloud", "dispo_nebius": true, "dispo_brev": true, "mode_tarif_fr": "Nœud 8x à la demande (Spot : 0,79 $/h)", "mode_tarif_en": "8x On-demand node (Spot: $0.79/h)", "machine_ref": "gpu-h100-sxm (8gpu-128vcpu-1600gb)", "cpu_modele": "2x Intel Xeon Platinum 8468 Sapphire Rapids", "vcpu_total": 128, "ram_systeme_go": 1600, "stockage_nvme_to": 25.6, "interconnexion": "NVLink 4 (900 Go/s par GPU)", "reseau_cluster": "8x 400 Gbps InfiniBand NDR (3,2 Tbit/s)", "prix_heure_gpu_usd": 4.5, "prix_heure_gpu_eur_ht": 3.95, "prix_heure_serveur_usd": 36.0, "prix_heure_serveur_eur_ht": 31.6, "cout_mensuel_730h_serveur_eur_ht": 23068, "cout_mensuel_730h_serveur_usd": 26280, "prix_spot_gpu_usd": 0.79, "prix_spot_gpu_eur_ht": 0.69, "vram_par_euro_heure": 20.3, "url_source": "https://nebius.com/prices", "url_rent_fr": "https://console.eu.nebius.com/", "url_rent_en": "https://console.eu.nebius.com/", "url_brev": "https://brev.nvidia.com/environment/new/public?gpu=H100", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "h100-8x", "date_releve": "2026-10-21" }, { "id": "runpod-h200-1x", "marque": "NVIDIA", "famille": "Hopper", "carte_gpu": "NVIDIA HGX H200 SXM5", "architecture": "Hopper", "nb_gpu": 1, "vram_gpu_go": 141, "vram_totale_go": 141, "type_vram": "HBM3e", "bp_memoire_gpu_tos": 4.8, "bp_memoire_totale_tos": 4.8, "tflops_fp16": 989, "tflops_fp8": 1979, "fournisseur_id": "runpod", "fournisseur": "RunPod", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Secure Cloud (Community : 3,59 $/h)", "mode_tarif_en": "Secure Cloud (Community: $3.59/h)", "machine_ref": "RunPod H200 SXM", "cpu_modele": "Intel Xeon Sapphire Rapids", "vcpu_total": 24, "ram_systeme_go": 276, "stockage_nvme_to": 2.0, "interconnexion": "NVLink 4 (900 Go/s)", "reseau_cluster": "InfiniBand NDR 400 Gbps", "prix_heure_gpu_usd": 5.29, "prix_heure_gpu_eur_ht": 4.64, "prix_heure_serveur_usd": 5.29, "prix_heure_serveur_eur_ht": 4.64, "cout_mensuel_730h_serveur_eur_ht": 3387, "cout_mensuel_730h_serveur_usd": 3862, "prix_spot_gpu_usd": 3.59, "prix_spot_gpu_eur_ht": 3.15, "vram_par_euro_heure": 30.4, "url_source": "https://www.runpod.io/pricing", "url_rent_fr": "https://www.runpod.io/console/gpu-cloud?ref=qdna", "url_rent_en": "https://www.runpod.io/console/gpu-cloud?ref=qdna", "url_brev": null, "rel_attr": "external nofollow noopener sponsored", "referral_actif": true, "serveur_site_ref": "dgx-h200", "date_releve": "2026-10-21" }, { "id": "nebius-h200-1x", "marque": "NVIDIA", "famille": "Hopper", "carte_gpu": "NVIDIA HGX H200 SXM5", "architecture": "Hopper", "nb_gpu": 1, "vram_gpu_go": 141, "vram_totale_go": 141, "type_vram": "HBM3e", "bp_memoire_gpu_tos": 4.8, "bp_memoire_totale_tos": 4.8, "tflops_fp16": 989, "tflops_fp8": 1979, "fournisseur_id": "nebius", "fournisseur": "Nebius AI Cloud", "dispo_nebius": true, "dispo_brev": true, "mode_tarif_fr": "À la demande (Spot : 1,40 $/h)", "mode_tarif_en": "On-demand (Spot: $1.40/h)", "machine_ref": "gpu-h200-sxm (1gpu-16vcpu-200gb)", "cpu_modele": "Intel Xeon Platinum 8468 Sapphire Rapids", "vcpu_total": 16, "ram_systeme_go": 200, "stockage_nvme_to": 3.2, "interconnexion": "NVLink 4 (900 Go/s)", "reseau_cluster": "InfiniBand NDR 400 Gbps (ConnectX-7)", "prix_heure_gpu_usd": 5.4, "prix_heure_gpu_eur_ht": 4.74, "prix_heure_serveur_usd": 5.4, "prix_heure_serveur_eur_ht": 4.74, "cout_mensuel_730h_serveur_eur_ht": 3460, "cout_mensuel_730h_serveur_usd": 3942, "prix_spot_gpu_usd": 1.4, "prix_spot_gpu_eur_ht": 1.23, "vram_par_euro_heure": 29.7, "url_source": "https://nebius.com/prices", "url_rent_fr": "https://console.eu.nebius.com/", "url_rent_en": "https://console.eu.nebius.com/", "url_brev": "https://brev.nvidia.com/environment/new/public?gpu=H200", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "dgx-h200", "date_releve": "2026-10-21" }, { "id": "nebius-h200-8x", "marque": "NVIDIA", "famille": "Hopper", "carte_gpu": "NVIDIA HGX H200 SXM5", "architecture": "Hopper", "nb_gpu": 8, "vram_gpu_go": 141, "vram_totale_go": 1128, "type_vram": "HBM3e", "bp_memoire_gpu_tos": 4.8, "bp_memoire_totale_tos": 38.4, "tflops_fp16": 989, "tflops_fp8": 1979, "fournisseur_id": "nebius", "fournisseur": "Nebius AI Cloud", "dispo_nebius": true, "dispo_brev": true, "mode_tarif_fr": "Nœud 8x à la demande (Spot dès 0,79 $/h)", "mode_tarif_en": "8x On-demand node (Spot from $0.79/h)", "machine_ref": "gpu-h200-sxm (8gpu-128vcpu-1600gb)", "cpu_modele": "2x Intel Xeon Platinum 8468 Sapphire Rapids", "vcpu_total": 128, "ram_systeme_go": 1600, "stockage_nvme_to": 25.6, "interconnexion": "NVLink 4 (900 Go/s par GPU)", "reseau_cluster": "8x 400 Gbps InfiniBand NDR (3,2 Tbit/s)", "prix_heure_gpu_usd": 5.4, "prix_heure_gpu_eur_ht": 4.74, "prix_heure_serveur_usd": 43.2, "prix_heure_serveur_eur_ht": 37.92, "cout_mensuel_730h_serveur_eur_ht": 27682, "cout_mensuel_730h_serveur_usd": 31536, "prix_spot_gpu_usd": 1.4, "prix_spot_gpu_eur_ht": 1.23, "vram_par_euro_heure": 29.7, "url_source": "https://nebius.com/prices", "url_rent_fr": "https://console.eu.nebius.com/", "url_rent_en": "https://console.eu.nebius.com/", "url_brev": "https://brev.nvidia.com/environment/new/public?gpu=H200", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "dgx-h200", "date_releve": "2026-10-21" }, { "id": "ovh-h200-8x", "marque": "NVIDIA", "famille": "Hopper", "carte_gpu": "NVIDIA HGX H200 SXM5", "architecture": "Hopper", "nb_gpu": 8, "vram_gpu_go": 141, "vram_totale_go": 1128, "type_vram": "HBM3e", "bp_memoire_gpu_tos": 4.8, "bp_memoire_totale_tos": 38.4, "tflops_fp16": 989, "tflops_fp8": 1979, "fournisseur_id": "ovhcloud", "fournisseur": "OVHcloud", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Nœud 8x Public Cloud France (Gravelines)", "mode_tarif_en": "8x France Public Cloud node (Gravelines)", "machine_ref": "h200-1920", "cpu_modele": "2x AMD EPYC / Intel Xeon", "vcpu_total": 224, "ram_systeme_go": 1920, "stockage_nvme_to": 30.72, "interconnexion": "NVLink 4 (900 Go/s)", "reseau_cluster": "4x 7,68 To NVMe Passthrough + 50 Gbps", "prix_heure_gpu_usd": 5.99, "prix_heure_gpu_eur_ht": 5.25, "prix_heure_serveur_usd": 47.92, "prix_heure_serveur_eur_ht": 42.0, "cout_mensuel_730h_serveur_eur_ht": 30660, "cout_mensuel_730h_serveur_usd": 34982, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 26.9, "url_source": "https://www.ovhcloud.com/fr/public-cloud/prices/", "url_rent_fr": "https://www.ovhcloud.com/fr/public-cloud/gpu/", "url_rent_en": "https://www.ovhcloud.com/en/public-cloud/gpu/", "url_brev": null, "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "dgx-h200", "date_releve": "2026-10-21" }, { "id": "inteltiber-gaudi3-8x", "marque": "Intel", "famille": "Gaudi 3", "carte_gpu": "Intel Gaudi 3 HL-325L OAM", "architecture": "Gaudi 3", "nb_gpu": 8, "vram_gpu_go": 128, "vram_totale_go": 1024, "type_vram": "HBM2e", "bp_memoire_gpu_tos": 3.7, "bp_memoire_totale_tos": 29.6, "tflops_fp16": 1835, "tflops_fp8": 3670, "fournisseur_id": "inteltiber", "fournisseur": "Intel Tiber AI Cloud", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Nœud bare-metal 8x (Réservé 1 an : 4,50 $/h)", "mode_tarif_en": "8x Bare-metal node (1-yr reserved: $4.50/h)", "machine_ref": "UBB 8x Intel Gaudi 3 HL-325L", "cpu_modele": "2x Intel Xeon Platinum 8580 Emerald Rapids", "vcpu_total": 224, "ram_systeme_go": 2048, "stockage_nvme_to": 30.72, "interconnexion": "21x 200GbE RDMA RoCEv2 sur puce (all-to-all)", "reseau_cluster": "6x 800GbE RoCEv2 inter-nœuds", "prix_heure_gpu_usd": 6.0, "prix_heure_gpu_eur_ht": 5.26, "prix_heure_serveur_usd": 48.0, "prix_heure_serveur_eur_ht": 42.08, "cout_mensuel_730h_serveur_eur_ht": 30718, "cout_mensuel_730h_serveur_usd": 35040, "prix_spot_gpu_usd": 4.5, "prix_spot_gpu_eur_ht": 3.95, "vram_par_euro_heure": 24.3, "url_source": "https://cloud.intel.com/compute-instances", "url_rent_fr": "https://cloud.intel.com/compute-instances", "url_rent_en": "https://cloud.intel.com/compute-instances", "url_brev": null, "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "dgx-h200", "date_releve": "2026-10-21" }, { "id": "lambda-b200-8x", "marque": "NVIDIA", "famille": "Blackwell", "carte_gpu": "NVIDIA HGX B200 SXM6", "architecture": "Blackwell", "nb_gpu": 8, "vram_gpu_go": 180, "vram_totale_go": 1440, "type_vram": "HBM3e", "bp_memoire_gpu_tos": 8.0, "bp_memoire_totale_tos": 64.0, "tflops_fp16": 2250, "tflops_fp8": 4500, "fournisseur_id": "lambda", "fournisseur": "Lambda", "dispo_nebius": false, "dispo_brev": true, "mode_tarif_fr": "Nœud 8x à la demande (1x à 6,99 $/h)", "mode_tarif_en": "8x On-demand node (1x at $6.99/h)", "machine_ref": "gpu_8x_b200_sxm6", "cpu_modele": "2x Intel Xeon Emerald Rapids", "vcpu_total": 208, "ram_systeme_go": 2900, "stockage_nvme_to": 22.0, "interconnexion": "NVLink 5 (1,8 To/s)", "reseau_cluster": "InfiniBand NDR 400 Gbps (3,2 Tbit/s)", "prix_heure_gpu_usd": 6.69, "prix_heure_gpu_eur_ht": 5.87, "prix_heure_serveur_usd": 53.52, "prix_heure_serveur_eur_ht": 46.96, "cout_mensuel_730h_serveur_eur_ht": 34281, "cout_mensuel_730h_serveur_usd": 39070, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 30.7, "url_source": "https://lambda.ai/service/gpu-cloud", "url_rent_fr": "https://cloud.lambdalabs.com/instances", "url_rent_en": "https://cloud.lambdalabs.com/instances", "url_brev": "https://brev.nvidia.com/environment/new/public?gpu=B200", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "dgx-b200", "date_releve": "2026-10-21" }, { "id": "ibmcloud-gaudi3-8x", "marque": "Intel", "famille": "Gaudi 3", "carte_gpu": "Intel Gaudi 3 HL-325L OAM", "architecture": "Gaudi 3", "nb_gpu": 8, "vram_gpu_go": 128, "vram_totale_go": 1024, "type_vram": "HBM2e", "bp_memoire_gpu_tos": 3.7, "bp_memoire_totale_tos": 29.6, "tflops_fp16": 1835, "tflops_fp8": 3670, "fournisseur_id": "ibmcloud", "fournisseur": "IBM Cloud", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Instance VPC 8x (Europe Francfort / US)", "mode_tarif_en": "8x VPC instance (Europe Frankfurt / US)", "machine_ref": "gx3d-160x1792x8gaudi3", "cpu_modele": "2x Intel Xeon 8568Y+ Emerald Rapids (80 cœurs)", "vcpu_total": 160, "ram_systeme_go": 1792, "stockage_nvme_to": 25.6, "interconnexion": "21x 200GbE OAM-to-OAM RoCEv2", "reseau_cluster": "200 Gbps VPC Network", "prix_heure_gpu_usd": 7.35, "prix_heure_gpu_eur_ht": 6.45, "prix_heure_serveur_usd": 58.8, "prix_heure_serveur_eur_ht": 51.6, "cout_mensuel_730h_serveur_eur_ht": 37668, "cout_mensuel_730h_serveur_usd": 42924, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 19.8, "url_source": "https://www.ibm.com/products/gpu-ai-accelerator/intel-gaudi3", "url_rent_fr": "https://www.ibm.com/products/gpu-ai-accelerator/intel-gaudi3", "url_rent_en": "https://www.ibm.com/products/gpu-ai-accelerator/intel-gaudi3", "url_brev": null, "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "dgx-h200", "date_releve": "2026-10-21" }, { "id": "runpod-b200-1x", "marque": "NVIDIA", "famille": "Blackwell", "carte_gpu": "NVIDIA HGX B200 SXM6", "architecture": "Blackwell", "nb_gpu": 1, "vram_gpu_go": 180, "vram_totale_go": 180, "type_vram": "HBM3e", "bp_memoire_gpu_tos": 8.0, "bp_memoire_totale_tos": 8.0, "tflops_fp16": 2250, "tflops_fp8": 4500, "fournisseur_id": "runpod", "fournisseur": "RunPod", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Secure Cloud (Community : 5,98 $/h)", "mode_tarif_en": "Secure Cloud (Community: $5.98/h)", "machine_ref": "RunPod B200 SXM", "cpu_modele": "Intel Xeon Emerald Rapids", "vcpu_total": 28, "ram_systeme_go": 283, "stockage_nvme_to": 2.0, "interconnexion": "NVLink 5 (1,8 To/s)", "reseau_cluster": "InfiniBand NDR 400 Gbps", "prix_heure_gpu_usd": 7.99, "prix_heure_gpu_eur_ht": 7.01, "prix_heure_serveur_usd": 7.99, "prix_heure_serveur_eur_ht": 7.01, "cout_mensuel_730h_serveur_eur_ht": 5117, "cout_mensuel_730h_serveur_usd": 5833, "prix_spot_gpu_usd": 5.98, "prix_spot_gpu_eur_ht": 5.24, "vram_par_euro_heure": 25.7, "url_source": "https://www.runpod.io/pricing", "url_rent_fr": "https://www.runpod.io/console/gpu-cloud?ref=qdna", "url_rent_en": "https://www.runpod.io/console/gpu-cloud?ref=qdna", "url_brev": null, "rel_attr": "external nofollow noopener sponsored", "referral_actif": true, "serveur_site_ref": "dgx-b200", "date_releve": "2026-10-21" }, { "id": "nebius-b200-1x", "marque": "NVIDIA", "famille": "Blackwell", "carte_gpu": "NVIDIA HGX B200 SXM6", "architecture": "Blackwell", "nb_gpu": 1, "vram_gpu_go": 180, "vram_totale_go": 180, "type_vram": "HBM3e", "bp_memoire_gpu_tos": 8.0, "bp_memoire_totale_tos": 8.0, "tflops_fp16": 2250, "tflops_fp8": 4500, "fournisseur_id": "nebius", "fournisseur": "Nebius AI Cloud", "dispo_nebius": true, "dispo_brev": true, "mode_tarif_fr": "À la demande (Spot : 4,50 $/h)", "mode_tarif_en": "On-demand (Spot: $4.50/h)", "machine_ref": "gpu-b200-sxm (1gpu-20vcpu-224gb)", "cpu_modele": "Intel Xeon Platinum 8580 Emerald Rapids", "vcpu_total": 20, "ram_systeme_go": 224, "stockage_nvme_to": 3.84, "interconnexion": "NVLink 5 (1,8 To/s)", "reseau_cluster": "InfiniBand NDR 400 Gbps (ConnectX-7)", "prix_heure_gpu_usd": 8.5, "prix_heure_gpu_eur_ht": 7.45, "prix_heure_serveur_usd": 8.5, "prix_heure_serveur_eur_ht": 7.45, "cout_mensuel_730h_serveur_eur_ht": 5438, "cout_mensuel_730h_serveur_usd": 6205, "prix_spot_gpu_usd": 4.5, "prix_spot_gpu_eur_ht": 3.95, "vram_par_euro_heure": 24.2, "url_source": "https://nebius.com/prices", "url_rent_fr": "https://console.eu.nebius.com/", "url_rent_en": "https://console.eu.nebius.com/", "url_brev": "https://brev.nvidia.com/environment/new/public?gpu=B200", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "dgx-b200", "date_releve": "2026-10-21" }, { "id": "nebius-b200-8x", "marque": "NVIDIA", "famille": "Blackwell", "carte_gpu": "NVIDIA HGX B200 SXM6", "architecture": "Blackwell", "nb_gpu": 8, "vram_gpu_go": 180, "vram_totale_go": 1440, "type_vram": "HBM3e", "bp_memoire_gpu_tos": 8.0, "bp_memoire_totale_tos": 64.0, "tflops_fp16": 2250, "tflops_fp8": 4500, "fournisseur_id": "nebius", "fournisseur": "Nebius AI Cloud", "dispo_nebius": true, "dispo_brev": true, "mode_tarif_fr": "Nœud 8x à la demande (Spot : 4,50 $/h)", "mode_tarif_en": "8x On-demand node (Spot: $4.50/h)", "machine_ref": "gpu-b200-sxm (8gpu-160vcpu-1792gb)", "cpu_modele": "2x Intel Xeon Platinum 8580 Emerald Rapids", "vcpu_total": 160, "ram_systeme_go": 1792, "stockage_nvme_to": 30.72, "interconnexion": "NVLink 5 (1,8 To/s par GPU)", "reseau_cluster": "8x 400 Gbps InfiniBand NDR (3,2 Tbit/s)", "prix_heure_gpu_usd": 8.5, "prix_heure_gpu_eur_ht": 7.45, "prix_heure_serveur_usd": 68.0, "prix_heure_serveur_eur_ht": 59.6, "cout_mensuel_730h_serveur_eur_ht": 43508, "cout_mensuel_730h_serveur_usd": 49640, "prix_spot_gpu_usd": 4.5, "prix_spot_gpu_eur_ht": 3.95, "vram_par_euro_heure": 24.2, "url_source": "https://nebius.com/prices", "url_rent_fr": "https://console.eu.nebius.com/", "url_rent_en": "https://console.eu.nebius.com/", "url_brev": "https://brev.nvidia.com/environment/new/public?gpu=B200", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "dgx-b200", "date_releve": "2026-10-21" }, { "id": "scaleway-b300-8x", "marque": "NVIDIA", "famille": "Blackwell Ultra", "carte_gpu": "NVIDIA HGX B300 SXM6", "architecture": "Blackwell Ultra", "nb_gpu": 8, "vram_gpu_go": 288, "vram_totale_go": 2304, "type_vram": "HBM3e", "bp_memoire_gpu_tos": 8.0, "bp_memoire_totale_tos": 64.0, "tflops_fp16": 2250, "tflops_fp8": 4500, "fournisseur_id": "scaleway", "fournisseur": "Scaleway (Groupe iliad)", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Nœud 8x souverain FR (Paris fr-par-2)", "mode_tarif_en": "8x FR sovereign node (Paris fr-par-2)", "machine_ref": "B300-SXM-8-288G", "cpu_modele": "2x Intel Xeon Granite Rapids", "vcpu_total": 224, "ram_systeme_go": 3840, "stockage_nvme_to": 24.0, "interconnexion": "NVLink 5 (1,8 To/s)", "reseau_cluster": "RoCEv2 / InfiniBand 800G", "prix_heure_gpu_usd": 8.55, "prix_heure_gpu_eur_ht": 7.5, "prix_heure_serveur_usd": 68.4, "prix_heure_serveur_eur_ht": 60.0, "cout_mensuel_730h_serveur_eur_ht": 43800, "cout_mensuel_730h_serveur_usd": 49932, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 38.4, "url_source": "https://www.scaleway.com/en/pricing/gpu/?zone=fr-par-2", "url_rent_fr": "https://console.scaleway.com/gpu-instances/instances", "url_rent_en": "https://console.scaleway.com/gpu-instances/instances", "url_brev": null, "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "dgx-b300", "date_releve": "2026-10-21" }, { "id": "runpod-b300-1x", "marque": "NVIDIA", "famille": "Blackwell Ultra", "carte_gpu": "NVIDIA HGX B300 SXM6", "architecture": "Blackwell Ultra", "nb_gpu": 1, "vram_gpu_go": 288, "vram_totale_go": 288, "type_vram": "HBM3e", "bp_memoire_gpu_tos": 8.0, "bp_memoire_totale_tos": 8.0, "tflops_fp16": 2250, "tflops_fp8": 4500, "fournisseur_id": "runpod", "fournisseur": "RunPod", "dispo_nebius": false, "dispo_brev": false, "mode_tarif_fr": "Secure Cloud (à la seconde)", "mode_tarif_en": "Secure Cloud (per-second)", "machine_ref": "RunPod B300 SXM Pod", "cpu_modele": "Intel Xeon Granite Rapids", "vcpu_total": 28, "ram_systeme_go": 360, "stockage_nvme_to": 2.0, "interconnexion": "NVLink 5 (1,8 To/s)", "reseau_cluster": "InfiniBand 800 Gbps", "prix_heure_gpu_usd": 8.99, "prix_heure_gpu_eur_ht": 7.88, "prix_heure_serveur_usd": 8.99, "prix_heure_serveur_eur_ht": 7.88, "cout_mensuel_730h_serveur_eur_ht": 5752, "cout_mensuel_730h_serveur_usd": 6563, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 36.5, "url_source": "https://www.runpod.io/pricing", "url_rent_fr": "https://www.runpod.io/console/gpu-cloud?ref=qdna", "url_rent_en": "https://www.runpod.io/console/gpu-cloud?ref=qdna", "url_brev": null, "rel_attr": "external nofollow noopener sponsored", "referral_actif": true, "serveur_site_ref": "dgx-b300", "date_releve": "2026-10-21" }, { "id": "nebius-b300-1x", "marque": "NVIDIA", "famille": "Blackwell Ultra", "carte_gpu": "NVIDIA HGX B300 SXM6", "architecture": "Blackwell Ultra", "nb_gpu": 1, "vram_gpu_go": 288, "vram_totale_go": 288, "type_vram": "HBM3e", "bp_memoire_gpu_tos": 8.0, "bp_memoire_totale_tos": 8.0, "tflops_fp16": 2250, "tflops_fp8": 4500, "fournisseur_id": "nebius", "fournisseur": "Nebius AI Cloud", "dispo_nebius": true, "dispo_brev": true, "mode_tarif_fr": "À la demande (Spot : 4,39 $/h)", "mode_tarif_en": "On-demand (Spot: $4.39/h)", "machine_ref": "gpu-b300-sxm (1gpu-24vcpu-346gb)", "cpu_modele": "Intel Xeon 6776P Granite Rapids", "vcpu_total": 24, "ram_systeme_go": 346, "stockage_nvme_to": 3.84, "interconnexion": "NVLink 5 (1,8 To/s)", "reseau_cluster": "InfiniBand XDR 800 Gbps (ConnectX-8)", "prix_heure_gpu_usd": 9.5, "prix_heure_gpu_eur_ht": 8.33, "prix_heure_serveur_usd": 9.5, "prix_heure_serveur_eur_ht": 8.33, "cout_mensuel_730h_serveur_eur_ht": 6081, "cout_mensuel_730h_serveur_usd": 6935, "prix_spot_gpu_usd": 4.39, "prix_spot_gpu_eur_ht": 3.85, "vram_par_euro_heure": 34.6, "url_source": "https://nebius.com/prices", "url_rent_fr": "https://console.eu.nebius.com/", "url_rent_en": "https://console.eu.nebius.com/", "url_brev": "https://brev.nvidia.com/environment/new", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "dgx-b300", "date_releve": "2026-10-21" }, { "id": "nebius-b300-8x", "marque": "NVIDIA", "famille": "Blackwell Ultra", "carte_gpu": "NVIDIA HGX B300 SXM6", "architecture": "Blackwell Ultra", "nb_gpu": 8, "vram_gpu_go": 288, "vram_totale_go": 2304, "type_vram": "HBM3e", "bp_memoire_gpu_tos": 8.0, "bp_memoire_totale_tos": 64.0, "tflops_fp16": 2250, "tflops_fp8": 4500, "fournisseur_id": "nebius", "fournisseur": "Nebius AI Cloud", "dispo_nebius": true, "dispo_brev": true, "mode_tarif_fr": "Nœud 8x à la demande (Spot dès 0,99 $/h)", "mode_tarif_en": "8x On-demand node (Spot from $0.99/h)", "machine_ref": "gpu-b300-sxm (8gpu-192vcpu-2768gb)", "cpu_modele": "2x Intel Xeon 6776P Granite Rapids", "vcpu_total": 192, "ram_systeme_go": 2768, "stockage_nvme_to": 23.04, "interconnexion": "NVLink 5 (1,8 To/s par GPU)", "reseau_cluster": "8x 800 Gbps InfiniBand XDR (6,4 Tbit/s)", "prix_heure_gpu_usd": 9.5, "prix_heure_gpu_eur_ht": 8.33, "prix_heure_serveur_usd": 76.0, "prix_heure_serveur_eur_ht": 66.64, "cout_mensuel_730h_serveur_eur_ht": 48647, "cout_mensuel_730h_serveur_usd": 55480, "prix_spot_gpu_usd": 4.39, "prix_spot_gpu_eur_ht": 3.85, "vram_par_euro_heure": 34.6, "url_source": "https://nebius.com/prices", "url_rent_fr": "https://console.eu.nebius.com/", "url_rent_en": "https://console.eu.nebius.com/", "url_brev": "https://brev.nvidia.com/environment/new", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "dgx-b300", "date_releve": "2026-10-21" }, { "id": "coreweave-gb200-4x", "marque": "NVIDIA", "famille": "Blackwell", "carte_gpu": "NVIDIA GB200 NVL72 (nœud 4x GPU)", "architecture": "Grace Blackwell", "nb_gpu": 4, "vram_gpu_go": 186, "vram_totale_go": 744, "type_vram": "HBM3e", "bp_memoire_gpu_tos": 8.0, "bp_memoire_totale_tos": 32.0, "tflops_fp16": 2250, "tflops_fp8": 4500, "fournisseur_id": "coreweave", "fournisseur": "CoreWeave", "dispo_nebius": false, "dispo_brev": true, "mode_tarif_fr": "Nœud 2x Grace + 4x B200 (rack NVL72)", "mode_tarif_en": "2x Grace + 4x B200 node (NVL72 rack)", "machine_ref": "GB200-NVL72-4x", "cpu_modele": "2x NVIDIA Grace Arm Neoverse V2 (144 cœurs)", "vcpu_total": 144, "ram_systeme_go": 960, "stockage_nvme_to": 15.36, "interconnexion": "NVLink-C2C 900 Go/s + NVLink 5 (1,8 To/s)", "reseau_cluster": "InfiniBand NDR/XDR 800 Gbps", "prix_heure_gpu_usd": 10.5, "prix_heure_gpu_eur_ht": 9.21, "prix_heure_serveur_usd": 42.0, "prix_heure_serveur_eur_ht": 36.84, "cout_mensuel_730h_serveur_eur_ht": 26893, "cout_mensuel_730h_serveur_usd": 30660, "prix_spot_gpu_usd": null, "prix_spot_gpu_eur_ht": null, "vram_par_euro_heure": 20.2, "url_source": "https://getdeploying.com/gpus/nvidia-gb200", "url_rent_fr": "https://console.coreweave.com/", "url_rent_en": "https://console.coreweave.com/", "url_brev": "https://brev.nvidia.com/environment/new", "rel_attr": "external noopener", "referral_actif": false, "serveur_site_ref": "dgx-b200", "date_releve": "2026-10-21" } ] }