The imminent Nvidia GeForce RTX 2080 Ti will come with 11GB of GDDR6 memory and offer 4,352 CUDA cores. Based on the Turing architecture that Nvidia has already used on its Quadro graphics cards for architects, engineers and creative pros, it may also be capable of handing ray tracing.
The latest leaks come courtesy of VideoCardz. It claims that the GDDR6 memory will clock at 14Gbps, increasing the theoretical maximum bandwidth to 616 Gigabits per second (Gbps) on its 352-bit memory bus.
The non-Ti GeForce RTX 2080, meanwhile, will come loaded with 8GB. It will be equipped with 2,944 CUDA cores - not the 3,072 that had been previously reported - and offer memory bandwidth of 448Gbps over a 256-bit memory bus.
Both will be launched today, according to VideoCardz at Gamescon - a break with recent convention by Nvidia whereby it normally launches the Ti edition graphics card some months after.
Other leaks, also courtesy of VideoCardz, indicate that Nvidia will be ditching the SLI connector for the RTX series of graphics cards. Instead, they will feature the new Type-C VirtualLink connector and NVLink, a multi-lane near-range communication link developed by Nvidia.
However, one of the few as-yet unanswered questions is, how much will Nvidia's new cards cost?
That should be revealed within hours…
Double legal trouble for Musk as he also faces civil lawsuit over renewed British pot-holer 'paedo' claims
Battery development could help boost performance of smartphones
Topological photonic chips promise a more robust option for scalable quantum computers
In quantum physics both the chicken and the egg can come first, claim University of Queensland researchers
Cause-and-effect is not always straightforward in quantum physics