New rumors regarding NVIDIA’s next flagship GeForce GPU have arrived courtesy of a leaker on China’s Chiphell forums. According to panzerlied, the GeForce RTX 5090, which is rumored to leverage NVIDIA’s new Blackwell architecture, will feature a range of improvements that include a 50% increase in scale, 52% increase in memory bandwidth, 78% increase in cache, and a 15% increase in frequency versus the GeForce RTX 4090. Those figures would imply that the GeForce RTX 5090 features over 20,000 CUDA cores, a 2.9 GHz Boost Clock, 1.5 TB/s of memory bandwidth, and 128 MB of L2 cache, delivering what the leaker claims is a 1.7x in performance improvement. Other rumors suggest that NVIDIA will be introducing a multi-chip module (MCM) design for some of its Blackwell GPUs.
The scale is increased by 50%, the memory bandwidth is increased by 52%, the cache is increased by 78%, and there is a 15% frequency increase and a
1.7X improvement.
This core architecture can be “broken” without major changes for another generation.
This is for comparison with 4090, not for comparison with AD102
After the dramas of GA100 and GH100, it seems that GB100 is finally going to use MCM.
— kopite7kimi (@kopite7kimi) September 18, 2023
Although the number of units(like GPCs or TPCs) in Blackwell will not increase significantly, there are significant changes in its unit structure.
— kopite7kimi (@kopite7kimi) September 18, 2023