NVIDIA · 192 GB HBM3e · 8000 GB/s bandwidth ·
Alibaba · 27.8B · model ctx up to 256K
Yes — the B200 192GB SXM (192 GB HBM3e) can run Qwen3.6-27B. At Q4 with the default 8K context it needs ≈ 21.85 GB VRAM (weights 15.29 GB + KV cache 5.56 GB + 1 GB runtime overhead), which fits within 192 GB. Decode at Q4/8K: ≈ 183.1 tok/s (estimated, batch 1). FP16 (≈ 62.16 GB) also fits, at ≈ 50.4 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
Quant
Context
VRAM needed
Fits?
tok/s (decode)
Q4
4K
19.07 GB
Yes
183.1est.
Q4
8K default
21.85 GB
Yes
183.1est.
Q4
32K
38.53 GB
Yes
183.1est.
FP16
4K
59.38 GB
Yes
50.4est.
FP16
8K default
62.16 GB
Yes
50.4est.
FP16
32K
78.84 GB
Yes
50.4est.
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).