Yes — with 1.6 GB to spare
Olmo 3 7B Instruct at Q4_K_M fits your Arc B570 entirely on the GPU at 8K context, at an estimated 56 tokens per second. Past 20K the KV cache pushes it over — quantise the cache to q8_0, or step down a quantisation, to go longer.
Fully on GPU
8K context
Q4_K_M · 4.1 GB
Apache 2.0
Released 20 Nov 2025
Fully open — training data and code included. Full multi-head attention, so its cache is 4× a GQA 7B at the same context.
The VRAM budget
weights 4.1 GB
Weights 4.1 GB
KV cache @ 8K 2.50 GB
Runtime overhead 0.6 GB
Free 1.6 GB of 8.8 GB
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| F16 | 13.6 GB | 16.7 GB | — | ~4.1 | Reference | 7.9 GB over |
| Q8_0 | 7.2 GB | 10.3 GB | 1K | ~15 | −0.1% ppl | 1.5 GB over |
| Q6_K | 5.6 GB | 8.7 GB | 9K | 41 | −0.4% ppl | Fits |
| Q5_K_M | 4.8 GB | 7.9 GB | 15K | 48 | −0.8% ppl | Fits |
| Q4_K_M | 4.1 GB | 7.2 GB | 20K | 56 | −1.9% ppl | Recommended |
| Q3_K_M | 3.3 GB | 6.4 GB | 27K | 69 | −5.4% ppl | Long context |
| Q2_K | 2.8 GB | 5.9 GB | 30K | 81 | −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. This model interleaves sliding-window layers (4096 tokens, 8 of 32 layers global), which is why its cache barely grows with context.
How to run it
$ ollama pull olmo-3:7b $ OLLAMA_CONTEXT_LENGTH=8192 \ ollama run olmo-3:7b
The default. One binary, a model registry, an OpenAI-compatible port. More on Ollama.
01Download is 4.1 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 20K context on this card.