Steam Deck OLED · Llama 3.2 3B Instruct · Q4_K_M
Yes. Steam Deck OLED runs Llama 3.2 3B Instruct at Q4_K_M with room left at 8,192 tokens, roughly 19–41 tokens per second.
| Piece | Value | Note |
|---|---|---|
| Weights | 1.88 GiB | Llama-3.2-3B-Instruct-Q4_K_M.gguf |
| KV cache | 0.88 GiB | 8,192 tokens |
| Engine overhead | 0.50 GiB | Flat 0.5 GiB allowance |
| 5% headroom | 0.16 GiB | On the sum above |
| Total | 3.42 GiB | Fits |
| Fast memory | 12.0 GiB | After OS reserve |
| Speed | 19–41 tok/s | Dense bandwidth range |
ollama run llama3.2:3b
Steam Deck OLED · Llama 3.2 3B Instruct ·GGUF file · Methodology
FAQ
How was this fit calculated?
Weights 1.88 GiB, KV cache 0.88 GiB at 8,192 tokens, engine overhead 0.50 GiB, then 5% headroom, for 3.42 GiB. Fast memory available is 12.0 GiB.
What if I change the quant or the context?
This page is Q4_K_M at 8,192 tokens. A smaller quant or a shorter context lowers the total. The homepage calculator changes both without a new page.
Can I trust the tokens per second?
The range 19–41 tok/s is a bandwidth roofline, 55% to 80% efficient. This dense model reads the whole weight file each token. Measure your own box before you buy around it.
What do I type?
ollama run llama3.2:3b
Is Steam Deck OLED assumed to have extra RAM?
No separate RAM pool. The chip's unified memory is the whole budget, minus the OS reserve described on the methodology page.