RTX PRO 6000 Blackwell leaks with 24,064 cuda and 96GB GDDR7 centers with 600W TGP

Nvidia is preparing to take its Blackwell architecture to Workstations with two new GPUS, according to recent leaks. The company allegedly intends to offer a 6000 RTX with 24,064 CUDA and 96GB capacity centers, as well as a 6000 x RTX model, with the same memory and possibly even more nuclei.

The information was found by the Leadtek website and the famous Harukaze “Leak”. Gathering data extraction from these two sources, Tom’s hardware has organized what is known about possible Blackwell -based RTX Pro 6000 specifications. Again, we emphasize that the information below is not yet official.

RTX Pro 6000 specifications:

• 24.064 NÚCLEOS CUDA
• 188 SMs
• 97.92% of DIE qualified
• Chip GB202-???
• 188 Nuiscleos RT
• 752 Tensor Cores
• 96GB of memory GDDR7
• 512-bit in the memory bus
• TGP (ou TBP) de 600W

For the RTX Pro 6000 x version we don’t have so much information, just appearing the same memory capacity. In this model, however, the GPU full name leaked as GB202-870.

96GB of Memory for RTX Pro 6000 Mira in IA?

It is interesting to compare the leakage of these specifications with the currently available RTX 6000, based on ADA architecture. For starters, it is curious that Nvidia will maintain such a similar name among GPU generations, but its Workstation nomenclature is often less intuitive than in desktop models.

The supposed RTX Pro 6000 brings almost six thousand cuddha nuclei more, and is twice the TGP, going from 300W to 600W. But one of the most striking parts is the impressive memory capacity, bordering on the 100GB. It is also double what the RTX 6000 offers, and from GDDR6 generation to GDDR7.

Related News:

Of course, it is speculated that the considerable increase in memory was made to promote the new plates as a solution to AI. Hype with technology has been Nvidia’s main source of income, and famously the feature is quite demanding memory in its processing.