...makes video editing perfect!The Different Editions of the
NVIDIA Pro 6000 Blackwell GPU
Where exactly are the differences?
And what exactly
can they do?
Berlin, August 2025
Read the English version of this text here.
Common Base Components
-
All variants use the same GB202 chip with nearly full specifications: 24,064 CUDA cores, 752 Tensor cores, 188 RT cores, 96 GB GDDR7 ECC, 512-bit memory bus, approx. 1792 GB/s memory bandwidth
-
5th Generation Tensor Units: Maximum AI performance with FP4 and DLSS 4
New Streaming Multiprocessors: Optimized for neural RTX shaders
4th Generation Raytracing Units
-
Workstation Edition
-
Offers the highest clock speed (Tom’s Hardware example boost clock ~2.6 GHz), delivering maximum FP32 performance (~125 TFLOPS), AI (~4000 TOPS) and RT (~380 TFLOPS)
-
TGP is about 600 W
Max-Q Workstation Edition
-
Power consumption halved (~300 W), with only about 10% performance loss compared to the workstation variant
-
Cooling system with blower-style, better suited for multi-GPU configurations (exhaust to the back)
Server Edition
-
Typically passive design, built for server airflow (no own fan)
-
Configurable TDP up to 600 W, adaptable to data center requirements
-
Also optimized for server RAIDs, AI factories, data-center workloads
-
| Feature | Workstation Edition | Max-Q Workstation Edition | Server Edition |
| GPU Architecture | Blackwell (GB202) | Blackwell (GB202) | Blackwell (GB202) |
| Memory | 96 GB GDDR7 ECC | 96 GB GDDR7 ECC | 96 GB GDDR7 ECC |
| Memory Bus | 512-bit, approx. 1792 GB/s | 512-bit, approx. 1792 GB/s | 512-bit, approx. 1792 GB/s |
| CUDA/Tensor/RT cores | 24,064 CUDA, 752 Tensor, 188 RT | 24,064 CUDA, 752 Tensor, 188 RT | 24,064 CUDA, 752 Tensor, 188 RT |
| AI Performance | ~4000 FP4 TOPS | Up to ~3600 FP4 TOPS | Up to ~4000 FP4 TOPS (configurable) |
| FP32 (Single Precision) Performance | ~125 TFLOPS | ~110 TFLOPS | ~117 TFLOPS (configurable) |
| Ray Tracing | ~380 TFLOPS | ~340 TFLOPS | ~380 TFLOPS (configurable) |
| TDP / Power Draw | ~600 W | ~300 W | ~600 W (configurable) |
| Cooling / Form Factor | Active cooling, pass-through design | Active cooling, pass-through design | Passive (server fans handle cooling) |
Key Differences Summarized:
Workstation Edition: Optimal for single-GPU desktop scenarios. Maximum performance, higher power consumption.
Max-Q Edition: Efficient alternative with significantly reduced power consumption and nearly comparable performance – ideal for systems with multiple GPUs and limited cooling.
Server Edition: Specifically for data centers and GPU clusters. Passively cooled, configurable performance, excellent for dense GPU installations.