Running GLM 5.2 on B300 SXM
GLM 5.2 has 744 billion parameters. B300 SXM offers 2,304 GB of HBM3e across 8 GPUs memory. This page puts the two side by side.
How much memory does GLM 5.2 need?
Weights take the parameter count multiplied by the format size, plus a 15% runtime margin for activations and buffers. In FP16 the calculation reads: 744 billion parameters × 2 bytes (FP16) = 1,488.0 GB of weights; × 1.15 runtime margin = 1,711.2 GB. The attention cache sits on top: it grows with context length and with the number of concurrent requests, so it is sized case by case.
| Format | Weights in memory | On this machine | What the format costs |
|---|---|---|---|
| FP16 | 1,711 GB | fits | full precision, the quality reference |
| FP8 | 856 GB | fits | negligible loss on most tasks |
| NVFP4 | 428 GB | fits | Blackwell format, native FP4 compute |
| Q4 | 513 GB | fits | block quantisation, broadly supported |
Model sheet: 744 billion parameters, 40 billion active per token, 1,048,576-token context, MIT licence. 744 billion announced by Z.AI; the repository holds 753 billion tensor elements in BF16 (1,507 GB).
GLM 5.2 activates 40 billion parameters per token out of 744 billion. Active parameters govern speed, not footprint: every expert stays resident in memory. Confusing the two under-sizes the machine by an order of magnitude.
What is left to serve requests?
GLM 5.2 fits comfortably on B300 SXM. In FP16, its weights take about 1,711 GB of the 2,304 GB available, leaving 593 GB for the attention cache and concurrency. The remaining margin decides how many concurrent requests and how much context the machine sustains. By QDNA convention a machine “fits” when the weights take at most 75% of its memory; beyond that it “barely fits” and forces a limit on context or on concurrency. The attention cache is not calculated on this page. These models’ architectures (compressed latent, linear or sparse attention, Mamba layers) share no per-token formula; it is measured on the machine, with the target context and concurrency [TO BE MEASURED].
B300 SXM: indicative price about €430k to €690k excl. VAT, not contractual. Source: BIZON X9000 G5 at $496,728 (November 2026 delivery) and Supermicro AS-8126GS-NB3RT at $795,000 from Vipera, both listed on 2 September 2026, converted at the ECB rate of $1.159 per €1. The machine targets the GE segment.
Where else can GLM 5.2 run?
- GLM 5.2 on Mac Studio Ultra
- GLM 5.2 on DGX Station
- GLM 5.2 on RTX PRO 6000 server
- GLM 5.2 on H200 SXM server
- GLM 5.2 on B200 SXM
See the GLM 5.2 and B300 SXM fact sheets.
How much memory does GLM 5.2 need?
In FP16, weights take about 1,711 GB including the runtime margin. The attention cache sits on top and depends on context.
Does GLM 5.2 fit on B300 SXM?
GLM 5.2 fits comfortably on B300 SXM. In FP16, its weights take about 1,711 GB of the 2,304 GB available, leaving 593 GB for the attention cache and concurrency.
Method and sources
Memory footprints are calculated, not measured: a reading on real hardware may differ depending on the engine and the exact weight format. Every figure below carries its source and the date of its last check.
- GLM 5.2: 744 billion parameters, 40 billion active, 1,048,576-token context, MIT licence. Hugging Face repository (config.json, API, model card), checked on 2 September 2026. 744 billion announced by Z.AI; the repository holds 753 billion tensor elements in BF16 (1,507 GB).
- B300 SXM: 2,304 GB of memory; HGX B300, 8 Blackwell Ultra GPUs at 288 GB, hence 2,304 GB (the HGX sheet rounds to 2.1 TB), 144 PFLOPS FP4 with sparsity (108 dense), 1.8 TB/s NVLink per GPU. Manufacturer sheet checked on 2 September 2026.
- B300 SXM price: about €430k to €690k excl. VAT, indicative and not contractual. Source: BIZON X9000 G5 at $496,728 (November 2026 delivery) and Supermicro AS-8126GS-NB3RT at $795,000 from Vipera, both listed on 2 September 2026, converted at the ECB rate of $1.159 per €1.
- Bytes per parameter: FP16 2 (16 bits per parameter, hence 2 bytes); FP8 1 (8 bits per parameter (E4M3), hence 1 byte, block scales not counted); NVFP4 0.5 (4 bits per parameter, hence 0.5 byte; 4-bit published weights weigh 0.54 to 0.56 byte per parameter with their scales (DeepSeek V4 Pro 865 GB for 1,599 billion, Kimi K3 1,561 GB for 2,780 billion)); Q4 0.6 (taken here as llama.cpp Q4_K_M, 0.6 byte per parameter including scales; Q4_K_S weighs 0.56, AWQ and GPTQ 0.55). Source: connaissance/faits.yaml, quantifications family, checked on 2026-08-31.
- Runtime margin × 1.15: QDNA operating assumption, not measured: 15% above the weights for activations, buffers and fragmentation.
- “Fits” threshold at 75% of memory: QDNA assumption, not measured, which keeps the remaining quarter for the attention cache and concurrency.