← AI Hashrate

Can B200 192GB SXM run Qwen2.5-32B-Instruct?

NVIDIA · 192 GB HBM3e · 8000 GB/s bandwidth · Alibaba · 32.8B · model ctx up to 128K

Yes — the B200 192GB SXM (192 GB HBM3e) can run Qwen2.5-32B-Instruct. At Q4 with the default 8K context it needs ≈ 25.6 GB VRAM (weights 18.04 GB + KV cache 6.56 GB + 1 GB runtime overhead), which fits within 192 GB. Decode at Q4/8K: ≈ 155.2 tok/s (estimated, batch 1). FP16 (≈ 73.16 GB) also fits, at ≈ 42.7 tok/s (estimated). Estimates come from memory-bandwidth math; rows tagged measured override estimates. Relative ranking is more reliable than absolute tok/s. Methodology.

Fit & speed by quant and context

QuantContextVRAM neededFits?tok/s (decode)
Q44K22.32 GBYes155.2 est.
Q48K default25.6 GBYes155.2 est.
Q432K45.28 GBYes155.2 est.
FP164K69.88 GBYes42.7 est.
FP168K default73.16 GBYes42.7 est.
FP1632K92.84 GBYes42.7 est.

Fits = weights + KV(ctx) + 1 GB overhead ≤ 95% of VRAM. Measured anchors are context-agnostic; the fit verdict is recomputed per context. A missing tok/s means the model is far beyond this card (offload-only territory).

VRAM breakdown at 8K context

QuantWeightsKV cacheOverheadTotal neededB200 192GB SXM VRAM
Q418.04 GB6.56 GB1 GB25.6 GB192 GB
FP1665.6 GB6.56 GB1 GB73.16 GB192 GB

Other GPUs that run Qwen2.5-32B-Instruct

All GPUs for Qwen2.5-32B-Instruct →

Other models for the B200 192GB SXM

All models on the B200 192GB SXM →