Nvidia is already working on another card from the Quadro RTX series. The GPU is to offer as much as 48 GB of memory on a 384-bit bus with a bandwidth of 768 GB/s.
According to the leaked screenshot above, we now know that the entire Nvidia chip has 10752 CUDA cores. We must bear in mind that the GeForce RTX 3080 has 10 GB of GDDR6X on a 320-bit memory bus with a memory bandwidth of 760 GB/s, and the number of CUDA cores reaches 8704.
In the case of the RTX 3090, the CUDA cores are 10496 with 24GB of GDDR6X memory. This means that the Quadro RTX will have 2% more CUDA cores.
The leaked image we are seeing is the Control Panel of Nvidia which lists several key specifications of the card itself. It is not stated what the card would be called, but according to Kopite7kimi, the card could probably be called the Quadro RTX A6000, which means there will be room for the high-end Quadro RTX A8000 card in the future.
The spec lists 10,752 CUDA cores, which are grouped into 84 SMs. There’s also a full-featured Ampere GA102 GPU, which should be faster than the GeForce RTX 3090 as it has 2% more cores, however, this will mainly depend on the clock speeds the card is running at. The Quadro RTX A6000 is clocked at 1860 MHz, which equates to approximately 40 teraflops of processing power.
The card has about 48GB of GDDR6 memory running through a 384-bit wide bus interface. This should equate to a total bandwidth of 768 GB/s. This is less bandwidth than the GeForce RTX 3090, which should be reaching almost 936 GB/s, as it uses GDDR6X memory.
The hypothetical Quadro A8000 could be using, in this case, GDDR6X memory to increase the bandwidth with its availability in the year 2021.