Nvidia Cyber ​​Attack: Ada GPUs and GeForce RTX40s technical features leaked

Nvidia, which was hit by a cyber attack a few days ago, is now subject to the release of confidential data. These leaks have specifically revealed some technical details about the GeForce RTX40 and their Ada GPUs.

Earlier rumors surrounding the next generation of Nvidia graphics cards – which we hope will take the name GeForce RTX40 – specified the use of ADA architecture. Apart from this description, it does not tell us much about the capabilities of these graphics cards, except that this generation should point to a better improvement in terms of performance – and still be happy – and the chip engraving for the GeForce RTX 30 GPUs goes through a 5nm (DMSC) process against 8nm (Samsung) .

This is not to mention the cyber attack that Jensen Huang’s company faced in late February. If the GeForce RTX 30’s LHR (Lite Hash Rate) mining limiter is not disabled or its graphics drivers are not provided free of charge, the smugglers threaten to disclose confidential details about the company’s plans and technologies. Since then, data has been growing everywhere. The DLSS source code was thus released, and more interestingly, the first architectural details about the GeForce RTX40s and their Ada GPUs.

Another Twitterian originally posted ‘another AD102 block map’, but as the original creator, I will add more description.https://t.co/7UoSeY3u30
I will continue with the next answer …

—Ftiwvoe (@ftiwvoe) March 2, 2022

The leaked details tell us that Nvidia should provide its own copy of AMD’s Infinity Case. It takes the form of a large secondary (L2) cache, which is shared by the GPU’s GPC modules. This L2 cache can reach a maximum of 96 MB on the AD102 chip – high – which is more than the maximum 6 MB of the GeForce RTX 30 under ampere configuration. Obviously, lower versions of chips do not benefit from more cache, while AD103 and AD104 GPUs are limited to 64 MB when entry level AD106 and AD107 are satisfied with 48 MB and 32 MB, respectively.

Note: GAxxx GPUs are similar to GeForce RTX 30 and ADxxx GPUs are similar to RTX 40.

In addition, the shift in workmanship at 5 nm will allow for a greater number of computing units, with an increase of up to 71% compared to current generation graphics cards. The AD102 chip can hold an area of ​​more than 600mm, just as it already does on the GeForce RTX 3080 and above GA102 chip. It remains to be seen whether power consumption will go on indefinitely and beyond … Recent rumors are not very encouraging in this regard, sometimes triggering over 450 watts.