Can V100 16GB PCIe run gpt-oss-20b?

NVIDIA · 16 GB HBM2 · 900 GB/s bandwidth · OpenAI · 21.5B (active 3.6B) · MoE · model ctx up to 128K

Fit & speed by quant and context

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

Other GPUs that run gpt-oss-20b

• B200 192GB SXM — ≈2411.7 tok/s Q4 est.
• MI300X 192GB — ≈1511.3 tok/s Q4 est.
• H200 141GB SXM5 — ≈1447.0 tok/s Q4 est.
• H100 80GB SXM5 — ≈1009.9 tok/s Q4 est.
• Gaudi 3 128GB — ≈934.2 tok/s Q4 est.
• A100 80GB SXM4 — ≈614.7 tok/s Q4 est.
• RTX 4090D 24GB (China) — ≈300.9 tok/s Q4 est.
• RTX 3090 Ti 24GB — ≈300.9 tok/s Q4 est.

All GPUs for gpt-oss-20b →

Other models for the V100 16GB PCIe

• MiniCPM5-1B — ≈895.6 tok/s Q4 est.
• Qwen3.5-2B — ≈483.6 tok/s Q4 est.
• DeepSeek-R1-0528-Qwen3-8B-layer-mix-bpw-3.8-mlx — ≈420.5 tok/s Q4 est.
• Llama-3.2-3B-Instruct — ≈302.3 tok/s Q4 est.
• gemma-4-26B-A4B-it — ≈254.5 tok/s Q4 est.
• Qwen3-4B-Instruct-2507 — ≈241.8 tok/s Q4 est.
• NVIDIA-Nemotron-3-Nano-4B-BF16 — ≈241.8 tok/s Q4 est.
• Qwen3-4B-Thinking-2507 — ≈241.8 tok/s Q4 est.

All models on the V100 16GB PCIe →