GeForce RTX 4000: Nvidia wants to know it again classic with Lovelace


Not only about AMD's RDNA 3 with Navi 31, the rumor mill of these days knows how to share potentially accurate bits of information, there is also news about Nvidia's next generation “Lovelace” from Twitter users who are known for correct predictions to report. It should stay with monolithic GPUs again.

Nvidia should stay with one chip

It has been assumed since 2020 that Nvidia will once again stick to the monolithic approach at Lovelace, i.e. graphics cards will only ever use one chip with all the necessary functional units. AMD, on the other hand, is said to have a multi-chip approach for RDNA 3, at least for the high-end variants based on Navi-31, and AMD's approach to CPUs, the infinity cache approach and the CDNA2 graphics card MI200 speak with two chips strong for it.

In the case of Nvidia, the last known for its accurate rumors, kopite7kimi had already spoken of possibly up to 12 graphics processing clusters (GA102 offers 7) for the consumer top chip AD102 at the end of 2020, which with an unadjusted internal structure up to 144 streaming multiprocessors and 18,432 Shader units would mean – 71 percent more than in the full expansion of the GA102 with 84 SM and 10,752 FP32-ALUs.

Lovelace should set to 5 nm

Twitter user greymon55 announced these days that Nvidia has finally adopted Lovelace as the next gaming GPU platform for GeForce RTX 4000. The statement was a reaction to a rumor that had also been spread that Nvidia might have shelved the 1-chip approach and opted for a multi-chip approach. In May, kopite7kimi had launched this option, at least for the top chip, but according to the latest information, Nvidia will field the AD102 as a single-chip solution against Navi 31. Nvidia's professional amp successor, Hopper, will in turn use modules. It was therefore already speculated that in an emergency Hopper could replace the largest Lovelace chip in a duel with Navi 31.

The production of the Lovelace GPUs should take place at TSMC in 5 nm , also announced greymon55 these days, although no information about the expansion stage (N5 or N5P) was available yet.

Consumption should increase again

A tweet from kopite7kimi, who cites 400 watts of power consumption as insufficient, suggests that the move to more modern production with an almost doubled number of execution units at the same time does not have to mean that consumption is reduced. Currently, the GeForce RTX 3090 (test) and the GeForce RTX 3080 Ti (test) end at 350 watts, with the GeForce RTX 2000 the Ti was ahead with 250 watts (all information relates to the reference).

< blockquote class = "text-asset text-width twitter-tweet" data-dnt = "true">

400 is not enough

& mdash; kopite7kimi (@ kopite7kimi) July 29, 2021

Once again increasing consumption, which must also be assumed for AMD with Navi 31, should not just be another problem for the manufacturers at the end of 2022 when it comes to cooling, it is also becoming increasingly difficult to sell – even if the performance and thus again the efficiency measured in FPS per watt increases.

Officially, Nvidia only has a period of time

Nvidia has so far does not even confirm that the architecture was named after the British mathematician Ada Lovelace. On a roadmap presented in April 2021, the group only spoke of “Ampere Next” and indicated that it would be published in 2022.

Nvidia's x86 and ARM roadmap (Image: Nvidia)