PNY NVIDIA RTX PRO 6000 Blackwell Server Edition 96GB GDDR7 workstation

The RTX PRO 6000 Blackwell is the undisputed king of workstation AI, with 96GB of VRAM that makes everything else look underfed. Just make sure your power supply and your wallet are ready for the hit.

The AI numbers are stupid good. 100th percentile for AI workloads and VRAM, 99th for ray tracing. That 96GB frame buffer means you can load models that would choke anything else in a single card. The 27,523 PassMark G3D score is solid, but raster performance sits at 78th percentile, which is a bit of a letdown for a card this expensive. The real surprise is the efficiency. 16th percentile. This thing drinks power like it's going out of style, and the passive cooling means you better have serious airflow in your chassis.

Against the ASUS TUF RTX 5090, the PRO 6000 wins on VRAM and AI muscle but loses on raster bang for the buck. The 5090 is a gaming card that happens to do AI well. This is an AI card that happens to do graphics. The Gigabyte AERO 5080 and MSI Ventus 5070 Ti aren't even in the same conversation for memory-hungry workloads. If you need 96GB, there's no real competition here. If you don't, buy a 5090 and pocket the difference.

There's no way to spin this as a value buy. The affordability score is 0th percentile. Zero. If you're asking 'is it worth it,' you're probably not the target audience. For the right workload, it pays for itself in time saved. For everyone else, it's a flex.

The RTX PRO 6000 Blackwell Server Edition is a monster. 96GB of GDDR7 on a 512-bit bus, 24,064 CUDA cores, and a 600W TDP that says 'I don't care about your power bill.' This is the card you buy when a 5090 feels like a toy and you need to run massive LLMs or render scenes that would make lesser GPUs cry. The one thing to know: it's the absolute best AI card you can slot into a workstation right now, and it's priced like it.

If you're a gamer or a casual creator, this isn't your card. Go grab an RTX 5090 or even a 5080 and spend the leftover cash on literally anything else. If you're doing light AI work, a 4090 or 5090 with 32GB will handle most hobbyist models fine.

Buy it if you're running production AI workloads, massive simulations, or rendering pipelines where 96GB of ECC VRAM is the difference between shipping and stalling. For everyone else, this is overkill with a capital O. It's the best at what it does, but what it does is very specific.