llama.cpp on the NVIDIA RTX Pro 6000 Blackwell
On the NVIDIA RTX Pro 6000 Blackwell, llama.cpp runs Qwen3.8 125B · 6B active at 4-bit at a median 105.5 tokens per second for generation, from 2 plain runs (2K to 240K context).
Speed by model on the NVIDIA RTX Pro 6000 Blackwell
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 125B · 6B active 4-bit | 105.5 t/smedian of 2, 2K to 240K context | ik_llama.cpp 40.0 t/sone report, 200K context NInfer 172.0 t/sone report, 512 context |
| GLM-5.3 320B · 18B active 2-bit | 69.9 t/sone report, context not stated | no other engine reported |
11 llama.cpp reports on the NVIDIA RTX Pro 6000 Blackwell
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.
- GLM-5.3 320B · 18B active · IQ2_XXS69.9 t/s
- DeepSeek V4 Flash 284B · 13B active · MXFP436.1 t/s(2 cards)
- Qwen3.6 35B · 3B active · Q4_K_MPP 13816.2 t/s(3 cards)
- Qwen3.6 27B · Q8_0118.0 t/s(speculative decoding)
- Qwen3.8 125B · 6B active · IQ4_XS101.9 t/s
- DeepSeek V4 Flash 284B · 13B active · Q3_K_XL27.2 t/s(2 cards)
- DeepSeek V4 Flash 284B · 13B active · Q4_K_XL31.2 t/s(speculative decoding, part in system RAM)
- Qwen3.8 27B · Q8_099.6 t/s(speculative decoding)
- Qwen3.8 27B · Q4_K_M153.9 t/s(speculative decoding)
- Qwen3.8 125B · 6B active · UD-IQ4_XS109.1 t/s
- Tencent-HY3 295B · 21B active · Q6_K57.4 t/s(8 cards)
Frequently asked
How fast is llama.cpp on the NVIDIA RTX Pro 6000 Blackwell?
On the NVIDIA RTX Pro 6000 Blackwell, llama.cpp runs Qwen3.8 125B · 6B active at 4-bit at a median 105.5 tokens per second for generation, from 2 plain runs (2K to 240K context).
Is llama.cpp faster than other engines on the NVIDIA RTX Pro 6000 Blackwell?
Same card, same model size and bit class, plain runs only: Qwen3.8 125B · 6B active at 4-bit gives llama.cpp 105.5 t/s (2 runs) and ik_llama.cpp 40.0 t/s (1), NInfer 172.0 t/s (1). Context length and engine build differ between the runs, so treat it as a lead, not a benchmark.