We have a very very delicious reputation that makes the rounds today, and it comes from a very credible source. @ kopite7kimi, the Twitter leak responsible for virtually all Ampere leaks, revealed a treat about the upcoming NVIDIA Ada Lovelace architecture (probably called NVIDIA ADA). Our colleagues at 3DCenter extrapolated a lot of information from the shape size, which Kopite seems to have more or less confirmed. Although the source is very reliable, we still mark this post as a rumor due to the size of the leak.
NVIDIA ADA GPU Leaked: Monster 64 TFLOPs GPU with 18432 CUDA Cores and 5nm process architecture
The Ada Lovelace architecture – which will probably only be referred to as NVIDIA ADA – was recently leaked by Kopite (and confirmed by Videocardz), and we already seem to have the preliminary specifications for NVIDIA’s upcoming GPU. As we mentioned in the original Ada article, it seems that Hopper has been delayed for now (and with it, NVIDIA’s MCM ambitions). Fortunately, it seems that NVIDIA has kept the pedal against the metal, and the Ada architecture, championed by the AD102 GPU, will be an absolute beast. Below is the original shape size leak:
GA102 has a “7 * 6” structure.
Maybe the AD102 will get a “12 * 6” structure.
– kopite7kimi (@ kopite7kimi) December 28, 2020
The people at 3DCenter quickly extrapolated lots of details (we’ve revised their TFLOP numbers to be a little more conservative with a 1.75 GHz clock), as Kopite confirmed:
And a larger cache. It looks that way.
– kopite7kimi (@ kopite7kimi) December 28, 2020
For those who want all the information in one place, there is a table that summarizes everything:
NVIDIA Lovelace AD102 Reputable GPU Specifications
|Process||5pm||Samsung 8nm||TSMC 12nm NFF|
|Graphics Processing Clusters (GPC)||12||7||6|
|Texture Processing Clusters (TPC)||72||42||36|
|Streaming Multiprocessors (SM)||144||84||72|
|Release||Mar 2017||September 19th||2022 (TBC)|
NVIDIA AD102 “ADA GPU” appears to have 18432 CUDA cores based on the information provided by Kopite. This is almost twice the number of cores present in Ampere, which was already a big step up from Turing. The only way this is even possible is because NVIDIA apparently builds this on the 5nm process which has significant die area and power reduction. Interestingly, if you assume a clock speed of 1.75 GHz, you can also get top performance with a precision of ADA 102 GPU: 64 TFLOPs.
According to Kopite, the ADA architecture will contain a much larger L2 cache (both Turing and Ampere have 6 MB cache), which means that this can be a larger architectural revision (as Turing was for Pascal and Pascal was for Fermi / Kepler) in instead of just the usual process shrinks. It is also unclear at this time whether NVIDIA will use Samsung’s 5nm process or TSMC. While the company has previously experienced poor returns at Samsung, the fact that TSMC is all suffocated and will not have access to the factory for quite some time now, and NVIDIA may be more willing to tap Samsung and take a shot at “unlimited” production .
Summary of NVIDIA ADA GPU architecture
In many ways, Ada Lovelace can be considered the world’s first computer enthusiast. She is the first person to understand that the analytical engine proposed by Charles Babbage had applications beyond mere calculation and also published what is believed to be the first algorithm (becoming the first computer programming program) intended to be carried by such machine. This was almost half a century before Alan Turing wanted to complete his work and invent the general computer during World War II.
NVIDIA has been known for basing its architecture on prominent physicists, mathematicians and scientists, and Ada Lovelace is no different. Videocardz actually managed to find a big hint in NVIDIA’s own department store that seems to confirm this rumor that the Lovelace architecture is the next generation of GPUs from the company. If you look at the heroes presented under the GTC’s main theme for 2018, you will find not only Ada Lovelace, but what are potentially all future architectural codenames from NVIDIA. Jensen may have been fooled by the entire future roadmap (as far as code names go) in the GTC’18 main key.
There are now several rumors that it seems that the Lovelace architecture will be based on a 5 nm process. Since NVIDIA has switched to Samsung’s foundry, it is unclear whether 5nm refers to a TSMC process or Samsung’s. Keep in mind, however, that a recent report from Korea had also confirmed a 6nm order from NVIDIA – meaning that it is either a new generation from NVIDIA before Lovelace or that the 6nm process was for the update lineup.
5nm is right
– kopite7kimi (@ kopite7kimi) December 21, 2020