NVIDIA · 192 GB HBM3e · 8000 GB/s bandwidth ·
Alibaba · 0.8B · model ctx up to 40K
Yes — the B200 192GB SXM (192 GB HBM3e) can run Qwen3-0.6B. At Q4 with the default 8K context it needs ≈ 1.6 GB VRAM (weights 0.44 GB + KV cache 0.16 GB + 1 GB runtime overhead), which fits within 192 GB. Decode at Q4/8K: ≈ 5000 tok/s (estimated, batch 1). FP16 (≈ 2.76 GB) also fits, at ≈ 1750.0 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
1.52 GB
Yes
5000est.
Q4
8K default
1.6 GB
Yes
5000est.
Q4
32K
2.08 GB
Yes
5000est.
FP16
4K
2.68 GB
Yes
1750.0est.
FP16
8K default
2.76 GB
Yes
1750.0est.
FP16
32K
3.24 GB
Yes
1750.0est.
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).