NVIDIA today launched the RTX PRO 6000 Blackwell series of professional graphics cards. These cards are based on the latest GeForce "Blackwell" graphics architecture, and the three chips the company already launched on it. Leading the pack, is the RTX PRO 6000, a card that completely maxes out the massive "GB202" silicon, featuring more shaders than even the GeForce RTX 5090, albeit at lower clock speeds. The idea behind this product is to give pro-vis users more shader power, driving a large amount of GDDR7 ECC memory. Specifically, the card comes with 24,064 CUDA cores across all 192 SM physically present on the silicon, besides 768 Tensor cores, 192 RT cores, 768 TMUs, and 192 ROPs. The card gets a humungous 96 GB of ECC GDDR7 memory across the chip's 512-bit wide memory interface, probably using 48 Gbit density memory chips. The card has a TGP of 600 W, making out the 12V2x6 power input. It comes with a board design resembling the RTX 5090.

Next up, is the RTX PRO 6000 Blackwell Max-Q. This card has essentially the same core-configuration as the RTX PRO 6000, but with a reduced TGP, and a simpler 2-slot board design that uses a lateral-blower. This card is meant for machines with multiple such cards installed, though something that isn't quite a rendering server. Lastly, there's the RTX PRO 6000 Server Edition. This card, again, has identical core-config to the others in the lineup, but with a board design optimized for rackmount servers and large rendering farms. The cooler relies on the rack's airflow for cooling.