Nvidia may use its own “Infinity Cache” on RTX 40 series graphics cards

Following a recent cyberattack on Nvidia, there have been a lot of leaks, ranging from DLSS source code to upcoming GPU specs. As dataminers continue to dig, we're learning new information, with the latest findings pointing to Nvidia utilising its own ‘Infinity Cache'-like tech for RTX 40 series graphics cards. 

The discovery was made by XinoAssassin, who found out that Nvidia RTX 40 series cards would come with 16MB of L2 cache per 64bit memory bus. Combining this information with a recent report from kopite7kimi regarding the memory buses of the Ada GPUs, we've made some simple multiplications to determine the amount of L2 cache each GPU will feature.

Ada will have huge L2 @VideoCardz pic.twitter.com/sjMGXttX0Y

— XinoAssassin (@xinoassassin1) March 2, 2022

In the table below, you'll find the current expected specs for RTX 40 GPUs based on recent leaks:

GPU	SMs count	CUDA cores count	Memory Bus Width	L2 Cache
AD102	144	18,432	384-bit	96MB
AD103	84	10,752	256-bit	64MB
AD104	60	7,680	192-bit	48MB
AD106	36	4,608	128-bit	32MB
AD107	24	3,072	128-bit	32MB

Based on the information currently available, the RTX 40 series cards will feature between 32MB and 96MB of L2 cache. This doesn't quite match up to the maximum cache available on RX 6000 series cards, and AMD may be able to increase its Infinity Cache with RDNA 3 coming later this year.

Discuss on our Facebook page, HERE.

KitGuru says: If these figures are indeed correct, we will see an entry-level RTX 40 series card with more L2 cache than the RTX 30 series flagship. That doesn't necessarily mean comparable performance but it is interesting nonetheless. 

Become a Patron!