Running MiniMax M3 on B200 SXM
MiniMax M3 has 428 billion parameters. B200 SXM offers 1,440 GB of HBM3e across 8 GPUs memory. This page puts the two side by side.
How much memory does MiniMax M3 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: 428 billion parameters × 2 bytes (FP16) = 856.0 GB of weights; × 1.15 runtime margin = 984.4 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 | 984 GB | fits | full precision, the quality reference |
| FP8 | 492 GB | fits | negligible loss on most tasks |
| NVFP4 | 246 GB | fits | Blackwell format, native FP4 compute |
| Q4 | 295 GB | fits | block quantisation, broadly supported |
Model sheet: 428 billion parameters, 23 billion active per token, 1,048,576-token context, MiniMax Community licence. About 428 billion announced, 23 active; the BF16 repository holds 427 billion elements (854 GB).
MiniMax M3 activates 23 billion parameters per token out of 428 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?
MiniMax M3 fits comfortably on B200 SXM. In FP16, its weights take about 984 GB of the 1,440 GB available, leaving 456 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].
B200 SXM: price on quotation. NVIDIA publishes no price; Aivres quoted about $340,000 excl. VAT for an HGX B200 server (Arc Compute, updated 30 July 2026, flagged as outdated by the author). The machine targets the GE segment.
Where else can MiniMax M3 run?
- MiniMax M3 on Mac Studio Ultra
- MiniMax M3 on DGX Station
- MiniMax M3 on RTX PRO 6000 server
- MiniMax M3 on H200 SXM server
- MiniMax M3 on B300 SXM
See the MiniMax M3 and B200 SXM fact sheets.
How much memory does MiniMax M3 need?
In FP16, weights take about 984 GB including the runtime margin. The attention cache sits on top and depends on context.
Does MiniMax M3 fit on B200 SXM?
MiniMax M3 fits comfortably on B200 SXM. In FP16, its weights take about 984 GB of the 1,440 GB available, leaving 456 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.
- MiniMax M3: 428 billion parameters, 23 billion active, 1,048,576-token context, MiniMax Community licence. Hugging Face repository (config.json, API, model card), checked on 2 September 2026. About 428 billion announced, 23 active; the BF16 repository holds 427 billion elements (854 GB).
- B200 SXM: 1,440 GB of memory; HGX B200, 8 Blackwell GPUs, 1.4 TB total memory, 144 PFLOPS FP4 with sparsity (72 dense), 1.8 TB/s NVLink per GPU. Manufacturer sheet checked on 2 September 2026.
- B200 SXM price: on quotation. NVIDIA publishes no price; Aivres quoted about $340,000 excl. VAT for an HGX B200 server (Arc Compute, updated 30 July 2026, flagged as outdated by the author).
- 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.