Nvidia RTX 4000 series GPU architecture specifications leaked
So many leaks, so much power
Share this article...
WePC is reader-supported. When you buy through links on our site, we may earn an affiliate commission. Prices subject to change. Learn more
Nvidia’s recent cyberattack has left them with lots of leaks coming from the hacking group and the information they gathered. Videocardz reports that one piece of information is the RTX 4000 series GPU architecture and specifications. These are only rumored and might not be the final ones but highly likely with multiple sources sending the same information it is likely they are the final architecture details.
This leaked information is just another piece of information coming from the infiltrated internal information from Nvidia. Following the released source code for DLSS, these are supposedly only a handful of files from terabytes of data that the hackers managed to retrieve from Nvidia’s internal servers.
What are the RTX 4000 series architecture specifications?
The information released is the CUDA count of the different dies coming in the RTX 4000 series architecture. They contain five desktop graphic processors, the same count as the Ampere (RTX 30) lineup. Here’s the reported CUDA count of those dies.
| ADA GPU | SMs | CUDAs | Memory Bus Width | L2 Cache |
|---|---|---|---|---|
| AD102 | 144 | 18432 | 384-bit | 96 MB |
| AD103 | 84 | 10752 | 256-bit | 64 MB |
| AD104 | 60 | 7680 | 256-bit | 64 MB |
| AD106 | 36 | 4608 | 192-bit | 48 MB |
| AD107 | 24 | 3072 | 128-bit | 32 MB |
| ADA GPU | SMs | CUDAs | Memory Bus Width | L2 Cache |
|---|---|---|---|---|
| GA102 | 84 | 10752 | 384-bit | 6 MB |
| GA103 | 60 | 10752 | 256-bit | 4 MB |
| GA104 | 48 | 6144 | 256-bit | 4 MB |
| GA106 | 30 | 3840 | 192-bit | 3 MB |
| GA107 | 20 | 2560 | 128-bit | 2 MB |
As of writing, even more information has been revealed by Videocardz and leakers. This is concerning the L2 cache of the GPU and the significant increase in the amount of cache available on the GPU. Even the lowest Lovelace GPU has 32MB, nearly 30 MB more than a GA102 die, which can be found in the RTX 3080s and 3090s. The overall leap in cache increases the throughput of the cache from 512 KB per 32-bit to a staggering 16 MB of cache per 54-bit memory bus.
Overall the significant increase in the RTX 4000 series architecture should produce significant performance increases over the current generation of cards. Which is also expected to come with the AMD Radeon lineup, giving us the most powerful consumer cards yet. An exciting prospect to look forward to later on in the year.
