llama.cpp on the AMD RX 9070 XT 16GB
On the AMD RX 9070 XT 16GB, llama.cpp runs Qwen3 8B at 4-bit at a median 98.8 tokens per second for generation, from 1 plain run (512 context).
Speed by model on the AMD RX 9070 XT 16GB
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 8B 4-bit | 98.8 t/sone report, 512 context | no other engine reported |
| Qwen3.5 9B 4-bit | 7.9 t/sone report, 32K context | no other engine reported |
| Qwen3.5 9B 6-bit | 62.0 t/sone report, 64K context | no other engine reported |
10 llama.cpp reports on the AMD RX 9070 XT 16GB
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 8B · Q4_K_M98.8 t/s
- Qwen3.5 9B · Q4_K_M7.9 t/s
- Qwen3.6 35B · 3B active62.0 t/s(part in system RAM)
- Qwen3.8 27B42.0 t/s(speculative decoding)
- Qwen3.8 27B · Q650.0 t/s(speculative decoding, part in system RAM)
- Qwen3.8 27B · IQ3_S43.7 t/s(speculative decoding)
- Qwen3.8 27B · Q4_K_M24.7 t/s(2 cards)
- Qwen3.6 35B · 3B active · Q4_K_MPP 706.0 t/s(part in system RAM)
- Qwen3.5 9B · Q6_K_XL62.0 t/s
- Qwen3.8 27B · Q650.0 t/s(2 cards, speculative decoding)
Frequently asked
How fast is llama.cpp on the AMD RX 9070 XT 16GB?
On the AMD RX 9070 XT 16GB, llama.cpp runs Qwen3 8B at 4-bit at a median 98.8 tokens per second for generation, from 1 plain run (512 context).