llama.cpp on the NVIDIA RTX 5060 Ti 16GB
On the NVIDIA RTX 5060 Ti 16GB, llama.cpp runs Qwen3.8 27B at 3-bit at a median 28.8 tokens per second for generation, from 2 plain runs (4K to 200K context).
Speed by model on the NVIDIA RTX 5060 Ti 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 |
|---|---|---|
| Qwen3.8 27B 3-bit | 28.8 t/smedian of 2, 4K to 200K context | no other engine reported |
| Gemma 4 26B · 4B active 4-bit | 95.2 t/sone report, 64K context | no other engine reported |
| Mimo 2.6 9B 8-bit | 41.6 t/sone report, 120K context | no other engine reported |
| Muse 30B 4-bit | 18.0 t/sone report, 132K context | no other engine reported |
| Qwen2.5 7B 4-bit | 77.0 t/sone report, context not stated | no other engine reported |
| Qwen3.6 27B 4-bit | 22.0 t/sone report, 75K context | no other engine reported |
24 llama.cpp reports on the NVIDIA RTX 5060 Ti 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.
- Qwen2.5 7B · Q477.0 t/s
- Qwen3.8 27B · IQ3_S28.5 t/s
- Gemma 4 26B · 4B active · MXFP4-MOE95.2 t/s
- Qwen3.8 125B · 6B active · IQ1_M30.5 t/s(2 cards, part in system RAM)
- Qwen3.8 27B · UD-IQ3_XXS29.2 t/s
- Qwen3.8 27B · IQ3_XXS35.0 t/s(speculative decoding)
- Qwen3.5 35B · 3B active · UD-IQ3_XXSPP 1168.0 t/s
- Nemotron 3.5 Lightning 30B · 3B active · NVFP480.0 t/s(2 cards)
- Mimo 2.6 9B · Q8_041.6 t/s
- Qwen3.8 27B76.0 t/s(2 cards, speculative decoding, part in system RAM)
- Model not identified100.0 t/s(2 cards)
- Qwen3.8 27B · IQ45.3 t/s(part in system RAM)
- Qwen3.8 · Q418.3 t/s(part in system RAM)
- Qwen3.6 35B · 3B active37.2 t/s
- Muse 30B · Q4_K_XL18.0 t/s
- Qwen3.6 35B · 3B active · Q4_K_XL40.0 t/s(part in system RAM)
- DeepSeek V4 Flash 284B · 13B active11.0 t/s(part in system RAM)
- Qwen3.8 27B · Q6_K68.3 t/s(speculative decoding, part in system RAM)
- Qwen3.8 27B · IQ4_XS31.3 t/s(speculative decoding)
- Qwen3.8 27B · q820.0 t/s(4 cards)
- Qwen3.8 27B · IQ3_XXSPP 400.0 t/s
- Qwen3.8 27B · NVFP4PP 1300.0 t/s(3 cards, speculative decoding)
- Qwen3.6 27B · IQ4_XS19.0 t/s(speculative decoding, part in system RAM)
- Qwen3.6 27B · IQ4_XS22.0 t/s
Frequently asked
How fast is llama.cpp on the NVIDIA RTX 5060 Ti 16GB?
On the NVIDIA RTX 5060 Ti 16GB, llama.cpp runs Qwen3.8 27B at 3-bit at a median 28.8 tokens per second for generation, from 2 plain runs (4K to 200K context).