llama.cpp on the AMD MI50 32GB
On the AMD MI50 32GB, llama.cpp runs Qwen3-Coder 30B · 3B active at 4-bit at a median 66.1 tokens per second for generation, from 1 plain run (context not stated).
Speed by model on the AMD MI50 32GB
Plain runs only: one device, one request, no speculative decoding, the whole model in memory. Each row is one model size at one bit class, with the other engines' plain runs of the same model on this card beside it. Context length and build still differ between runs, and each side shows its own.
| Model | llama.cpp | Other engines |
|---|---|---|
| Qwen3-Coder 30B · 3B active 4-bit | 66.1 t/sone report, context not stated | no other engine reported |
5 llama.cpp reports on the AMD MI50 32GB
Newest first. A run on several cards, with speculative decoding, with part of the model in system RAM or with batched requests says so, since each describes a different setup.
- Qwen3.8 27B · Q6_K20.2 t/s(2 cards, speculative decoding)
- Qwen3-Coder 30B · 3B active · Q4_K_M66.1 t/s
- Qwen3.8 125B · 6B active · UD-Q4_K_XL15.0 t/s(3 cards)
- Model not identified15.1 t/s
- Qwen3.8 27B13.6 t/s(speculative decoding)
Frequently asked
How fast is llama.cpp on the AMD MI50 32GB?
On the AMD MI50 32GB, llama.cpp runs Qwen3-Coder 30B · 3B active at 4-bit at a median 66.1 tokens per second for generation, from 1 plain run (context not stated).