NVIDIA RTX PRO 4000 Blackwell
GPUBlackwell; 24GB VRAM, 672 GB/s bandwidth, 145W TDP, CUDA cores 8,960. Release date: 2025-03-18. Source: LCZ hardware catalog, with missing fields supplemented from NVIDIA official materials.
4
Run records
1
Models covered
2
Frameworks covered
L0 Self-reported × 4
Evidence mix
Typical measured performance
Published benchmark averages per model × framework × quantization on this hardware.
- SGLang · IQ2_XS Decode 2.1 · Prefill 15.8 ·
- SGLang · IQ3_S Decode 1.8 · Prefill 13.5 ·
- vLLM · IQ2_XS Decode 2.2 · Prefill 16.5 ·
- vLLM · IQ3_S Decode 1.8 · Prefill 13.5 ·
Key specs
- Vendor
- NVIDIA
- Architecture
- Blackwell
- VRAM
- 24 GB
- Memory bandwidth
- 672 GB/s
- TDP
- 145 W
- Release date
- 2025-03-18
Run records
4 configurations
| Model | Framework | Quantization | Decode | Gen time | Samples |
|---|---|---|---|---|---|
| DeepSeek-V4.1-Flash | SGLang | IQ2_XS | 2.1 tok/s | — | 1 |
| DeepSeek-V4.1-Flash | SGLang | IQ3_S | 1.8 tok/s | — | 1 |
| DeepSeek-V4.1-Flash | vLLM | IQ2_XS | 2.2 tok/s | — | 1 |
| DeepSeek-V4.1-Flash | vLLM | IQ3_S | 1.8 tok/s | — | 1 |