Next-generation NVIDIA Quadro RTX (Ampere) to feature 10752 CUDA cores

NVIDIA is quietly preparing a new generation of RTX Quadro series. The company’s new professional line of graphics cards will likely launch next month.

Next-generation Quadro gets full GA102 GPU

More leaks related to the Quadro RTX Ampere series are emerging. Yesterday Moore’s Law is Dead posted a screenshot from NVIDIA Control Panel showcasing yet unreleased configuration of the Ampere graphics card. He later confirmed it is a Quadro model.

The card would allegedly feature 10752 CUDA cores and 48 GB of GDDR6 non-X memory. This would mean that the next-gen Quadro has 256 CUDA cores more than RTX 3090 graphics card and twice the memory. Interestingly the Quadro would not feature GDDR6X memory technology and the speed will be capped at 16.0 Gbps, the highest frequency available for this technology. Micron currently does not produce 16Gbit (2GB) GDDR6X modules so NVIDIA had no choice but to use GDDR6 instead. There are plans for a higher density GDDR6X module production next year.

The card would retain a 384-bit memory bus though, but with the reduced speed, it effectively lowers the bandwidth to 768 GB/s, just 8GB/s more than GeForce RTX 3080.

Next Gen NVIDIA Quadro RTX specifications, Source: Moore’s Law is Dead

NVIDIA Quadro RTX Ampere series has been rumored for a long time. KittyCorgi, who many of you should probably recognize from previous Ampere leaks has been sharing details on the Quadro A8000 (a possible name for the card). He confirmed the card would feature 48GB of memory back in July. The card might also be called Quadro 6000, another leaker, kopite7kimi claims.

It is unclear if the new series will in fact be called Ax000. Quadro series have usually kept rather a simple naming schema by increasing the second number in the product series name (for instance: RTX 8000 -> RTX 8100).

NVIDIA will host a GTC Online event on October 5 where new Quadro RTX graphics cards are likely to be unveiled.

Source: Moore’s Law is Dead (Twitter)