Yes — with 74.4 GB to spare
Qwen3 4B at Q4_K_M fits your H100 SXM entirely on the GPU at 8K context, at an estimated 897 tokens per second. There is room for its full 32K window.
Fully on GPU
8K context
Q4_K_M · 2.3 GB
Apache 2.0
Released Apr 2025
The 2025 sweet spot for 8 GB cards with reasoning traces. Qwen3.5 4B adds vision and 8× the context.
The VRAM budget
weights 2.3 GB
Weights 2.3 GB
KV cache @ 8K 1.13 GB
Runtime overhead 0.6 GB
Free 74.4 GB of 78.4 GB
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| F16 | 7.5 GB | 9.2 GB | 32K | 271 | Reference | Long context |
| Q8_0 | 4.0 GB | 5.7 GB | 32K | 510 | −0.1% ppl | Long context |
| Q6_K | 3.1 GB | 4.8 GB | 32K | 661 | −0.4% ppl | Long context |
| Q5_K_M | 2.7 GB | 4.4 GB | 32K | 764 | −0.8% ppl | Long context |
| Q4_K_M | 2.3 GB | 4.0 GB | 32K | 897 | −1.9% ppl | Recommended |
| Q3_K_M | 1.8 GB | 3.6 GB | 32K | 1108 | −5.4% ppl | Long context |
| Q2_K | 1.6 GB | 3.3 GB | 32K | 1294 | −15% ppl | Long context |
Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it.
How to run it
$ ollama pull qwen3:4b $ OLLAMA_CONTEXT_LENGTH=8192 \ ollama run qwen3:4b
The default. One binary, a model registry, an OpenAI-compatible port. More on Ollama.
01Download is 2.3 GB. Keep it on an SSD — a first load off a spinning disk takes minutes.
02Close anything else holding VRAM. A browser with hardware acceleration can sit on 1–2 GB.
03There is room to go to the model's full 32K context on this card.