Rumors are swirling about the specs of Nvidia’s GeForce RTX 40 graphics cards. The latest concerns the release of this new series, which could take place as early as August.
The graphics card market calendar continues to be talked about. AMD has just officially announced the revival of its Radeon RX 6000 GPUs, refreshed versions with higher operating frequencies for better performance, while Intel is expected to launch its Arc graphics cards soon, which are a little overdue for the moment.
An outing for July/August?
On the Nvidia side, we expected the GeForce RTX 40 from September at best, but it seems that their launch could take place earlier. The leaker @kopite7kimi indeed mentioned on Twitter a release at the beginning of the 3rd quarter, in July therefore. If this can give rise to the hope of getting our hands on these graphics cards in the summer, we have some doubts about a July release, which would be unusual, Nvidia being more accustomed to releases at the end of August, around Gamescom.
Q3 early
— kopite7kimi (@kopite7kimi) May 15, 2022
New features for this RTX 40 series
The latest rumors concerning the RTX 40 indicate that the GPUs will be based on Ada Lovelace architecture (AD102, AD103, etc.), replacing Ampere (GA102, GA103…). This excludes the Hopper architecture, which was unveiled in March for the professional sector with a clear focus on hardware acceleration and AI.
To sum up, we expect much better performance with a number of calculation units (CUDA cores) significantly higher than the previous generation (18,432 for the AD102 against 10,752 for the GA102, i.e. almost double), this which would also translate to 20-70% more SM (Streaming Multiprocessor) blocks depending on the GPU. Finally, another interesting element, the L2 cache could be multiplied by 16, going from 6 MB to 96 MB. On the video memory side, GDDR6 and GDRR6x will however still be on the program.
To ensure such developments and despite the use of fine engraving in 5 nm, Nvidia necessarily had to increase the thermal envelope (TGP) of its GPUs, which could literally fly away on high-end models. Many figures have been put forward in recent weeks, but we can think that the RTX 4090s will exceed 600 or even 800 W (when the 3090 Ti FE tops out at 450 W, which is already huge). Powering these graphics cards will therefore require the use of the new PCIe 5 standard, which offers up to 600W of power through a single connector. It will therefore be necessary for users to equip themselves with a muscular power supply, with the appropriate PCIe 5 connectors. For more mainstream GPUs like the RTX 4060, RTX 4070, and RTX 4080, however, we don’t yet have any information on their TGPs, which will likely be much lower and more “acceptable” for our PCs. desktop.
We therefore only have a few months to wait before being able to test this series, which looks promising. Until then, we will probably know a little more according to leaks and rumors.