NVIDIA GeForce RTX 40 Series to Feature Huge Increase in L2 Cache, 90 MB More than Ampere RTX GPUs

[Tsing]

Image: NVIDIA



NVIDIA GeForce RTX 40 Series GPUs are getting a massive upgrade in the L2 cache department, according to the latest data derived from the cyber attack on green team.



The AD102 GPU, which will presumably power next-generation flagship graphics cards such as the GeForce RTX 4080/4090, will feature up to 96 MB of L2 cache, a huge increase over the GA102’s (e.g., GeForce RTX 3080/3090) 6 MB of L2 cache.



Some believe this could be NVIDIA’s own version of AMD’s Infinity Cache, which red team implemented for its Radeon RX 6000 Series graphics cards as an effective way of boosting memory bandwidth while keeping to a smaller memory bus.



Ada will have huge L2 @VideoCardz pic.twitter.com/sjMGXttX0Y— XinoAssassin (@xinoassassin1)...

Continue reading...

 

[Brian_B]

Those are going to be some huge dies

 

[MadMummy76]

I didn't even know GPUs had any cache. It is not a good sign that the stat worth highlighting is one nobody even mentioned in relation to GPUs until now.

 

[Grimlakin]

I didn't even know GPUs had any cache. It is not a good sign that the stat worth highlighting is one nobody even mentioned in relation to GPUs until now.

You don't recall all the hubbub around "infinity fabric" and 128mb of l2 gpu cache around the Radeon cards? This is Nvidia catching up to that.