Nvidia GeForce RTX 4000 GPUs may feature 71% more CUDA cores

pcgamesn.com:

Nvidia confirmed that it suffered a data breach at the end of February, and we’re now seeing several GeForce RTX 4000 graphics card details emerging from all corners of the internet. This potentially gives us our clearest look at Lovelace’s specs ahead of its reported release at the end of the year, and what kind of improvement we can expect versus the current generation RTX 3000 series.

Hardware leaker harukaze5719 has neatly summarised the key specifications of Nvidia’s five desktop GPU processors, in conjunction with fellows xinoassassin1 and kopite7kimi. Based on their findings, it appears that the flagship AD102 die will feature 18,432 CUDA cores. This would be a whopping 71% improvement versus the GA102 currently found in Nvidia’s best graphics cards, such as the GeForce RTX 3090.

It seems that Nvidia is also planning to massively increase the amount of L2 cache on its RTX 4000 GPUs, with the least powerful AD107 die boasting 32MB of the stuff, topping out at 96MB with AD102. This means that RTX 4000 cards may have between 6-16 times the amount found in GA102, but we’ll need to see how this impacts real-world performance.

While these kinds of specs multipliers are exciting, you may need a pricey power supply to accommodate any RTX 4000 cards, to the tune of 1,500W. This could indicate that Nvidia will continue to trail AMD when it comes to the power efficiency of its graphics cards, which could give team red’s upcoming RDNA 3 GPUs an advantage when they debut later this year.

AD104 192-bit 48MB AD106 128-bit 32MB AD107 128-bit 32MB

— XinoAssassin (@xinoassassin1) <a href=«https://twitter.com/xinoassassin1/status/1498920248535560193?ref_src=» https: www.pcgamesn.com rel=«noopener»>March 2, 2022

Rumours suggest that