GPUs & Macs › NVIDIA RTX PRO 6000 Blackwell

NVIDIA RTX PRO 6000 Blackwell: local LLM speed

Llama 2 7B (Q4_0) generates 281 tokens/s on the NVIDIA RTX PRO 6000 Blackwell (CUDA), and processes prompts at 16,619 tokens/s.

Measured results

ModelGenerate
tok/s		Prompt
tok/s		QuantBackendTestSource
Llama 2 7B28116,619Q4_0CUDA FA ontg128llama.cpp #15013 details
Llama 2 7B27414,855Q4_0CUDA FA offtg128llama.cpp #15013 details
gpt-oss 20B28711,522MXFP4CUDA FA ontg128llama.cpp #15396 details
gpt-oss 120B1965,518MXFP4CUDA FA ontg128llama.cpp #15396 details
Llama 3.2 1B Instruct19120,440Q4_K_Mllamafile LocalScore avgLocalScore details
Llama 3.1 8B Instruct85.86,346Q4_K_Mllamafile LocalScore avgLocalScore details
Qwen2.5 14B Instruct64.04,416Q4_K_Mllamafile LocalScore avgLocalScore details

What fits on the NVIDIA RTX PRO 6000 Blackwell

Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 281 tok/s (CUDA, source). Fit assumes the whole model in GPU memory with ~8K context.

ModelBest quant that fitsFileExpected tok/s
Qwen3.8-27BQ8_029.0 GB37.0 estimate
gemma-4-12B-itQ8_012.7 GB84.8 estimate
Qwen3.5-9BQ8_09.5 GB113 estimate
Qwen3.5-4BQ8_04.5 GB240 estimate
Qwen3.6-35B-A3BQ8_036.9 GB170 estimate
Qwen3.8-Flash-NextUD-IQ3_XXS82.0 GB13.1 estimate
gemma-4-26B-A4B-itQ8_026.9 GB130 estimate
MiMo-V2.6-Distill-Qwen-9BQ8_09.5 GB113 estimate
gemma-4-E4B-itQ8_08.2 GB131 estimate
Muse-Glimmer-30BQ8_029.6 GB36.3 estimate
Qwen3-4BQ8_04.3 GB251 estimate
gemma-4-31B-itQ8_032.6 GB32.9 estimate
gemma-4-26B-A4B-it-qat-q4_0-unquantizedUD-Q4_K_XL14.2 GB245 estimate
gemma-4-12B-it-qat-q4_0-unquantizedUD-Q4_K_XL6.7 GB160 estimate

Open in the picker

Similar hardware

NVIDIA RTX PRO 6000 Blackwell Max-Q · NVIDIA RTX 6000 Ada · NVIDIA RTX A6000 · NVIDIA Quadro RTX 8000 · NVIDIA RTX PRO 4500 Blackwell · NVIDIA RTX PRO 4000 Blackwell · NVIDIA RTX A5000 · NVIDIA Quadro RTX 6000