NVIDIA · 80 GB HBM3 · 3350 GB/s bandwidth ·
Alibaba · 4.0B · model ctx up to 256K
Yes — the H100 80GB SXM5 (80 GB HBM3) can run Qwen3-4B-Instruct-2507. At Q4 with the default 8K context it needs ≈ 4.0 GB VRAM (weights 2.2 GB + KV cache 0.8 GB + 1 GB runtime overhead), which fits within 80 GB. Decode at Q4/8K: ≈ 533.0 tok/s (estimated, batch 1). FP16 (≈ 9.8 GB) also fits, at ≈ 146.6 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
3.6 GB
Yes
533.0est.
Q4
8K default
4.0 GB
Yes
533.0est.
Q4
32K
6.4 GB
Yes
533.0est.
FP16
4K
9.4 GB
Yes
146.6est.
FP16
8K default
9.8 GB
Yes
146.6est.
FP16
32K
12.2 GB
Yes
146.6est.
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).