NVIDIA · 141 GB HBM3e · 4800 GB/s bandwidth ·
StepFun · 196.8B (active 11.0B) · MoE · model ctx up to 256K
Yes — the H200 141GB SXM5 (141 GB HBM3e) can run Step-3.5-Flash. At Q4 with the default 8K context it needs ≈ 111.44 GB VRAM (weights 108.24 GB + KV cache 2.2 GB + 1 GB runtime overhead), which fits within 141 GB. Decode at Q4/8K: ≈ 277.7 tok/s (estimated, batch 1). FP16 needs ≈ 396.8 GB — does not fit on this card. 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
110.34 GB
Yes
277.7est.
Q4
8K default
111.44 GB
Yes
277.7est.
Q4
32K
118.04 GB
Yes
277.7est.
FP16
4K
395.7 GB
No
27.7est.
FP16
8K default
396.8 GB
No
27.5est.
FP16
32K
403.4 GB
No
26.7est.
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).