According to the latest leaks from an already reliable leaker such as @kopite7kimithe next graphics architecture Nvidia Ada Lovelace will be more than just a “refreshed” version of the Nvidia Ampere architecture. Although it was initially estimated 120% performance improvementthis improvement could be higher, because this improvement was related to an old design of the Nvidia AD102 graphics chip, because it indicates that the Nvidia AD102 chip which will finally arrive on the market is not “the original AD102” which he spoke of in his time.
In this leak, it revealed that the competitor’s high-end graphics chip, the AMD Navi31 (RDNA3), would improve performance 150%So now hopefully Nvidia will be able to deliver an equal or greater performance boost after making some changes to the original chip design. This performance boost isn’t bad at all, given that Nvidia will be stuck with a monolithic chip design while AMD will offer a multi-chip design.
I should clarify that the current AD102 is NOT the original AD102. Ada Lovelace is no longer a mere refresh of Ampere, even though it was that way in the beginning.
– kopite7kimi (@kopite7kimi) April 29, 2022
According to the latest rumors, the graphics Nvidia GeForce RTX 40 series would employ a manufacturing process of 5nm from TSMCand there are even rumors that to further improve performance and compensate for high power consumption, we might see high-end GPUs at a 4nm manufacturing processsince the company will seek the path of raw power at all costs, and it is that Nvidia Ada Lovelace will mark the end of the company’s monolithic designs.
On the other hand, the future high-end GPU that the company will offer based on the graphics chip NvidiaAD102 complete, would have an energy consumption of 900W of powerthus necessitating the use of 2 x 16-pin PCIe 5.0 connectors for your food. This model would offer nothing less than 48 GB of GDDR6X memory @ 24 Gbps, so it is already assumed that could be a new Nvidia TITAN RTX GPU or a future GeForce RTX 4090 Ti.
On the other hand, it is indicated that it is not known whether it is a current product, since the test platforms of said silicon have more than 2 x 16-pin PCIe 5.0 slots. Remember that the more aggressive GeForce RTX 3090 Ti models already have two PCIe 5.0 x16 slots, so basically companies have gone to the trouble of preparing designs for the new generation of GPUs because all the work done in those GPUs will be exportable to the RTX 40 series.
Remember that Nvidia GeForce RTX 4090 It would arrive with 18432 CUDA cores as well as 24 GB of GDDR6X @ 21 Gbps memory with a power consumption of 600W; the GeForce RTX 4080 would have 10752 CUDA cores, 16 GB GDDR6X @ 21 Gbps with a consumption of 350W; while the GeForce RTX 4070 7680 CUDA cores would arrive with 12 GB of GDDR6 presenting a power consumption of 300W.
It should be remembered that the Nvidia GeForce RTX 40 series will be announced between September and Octoberit is therefore normal that as we get closer to the date, more information about them becomes known.