One of the reasons why a processor, of any type whatsoever, is broken down into several chiplets, in order to thus exceed the limit of the grid that the wafer allows. This allows more complex configurations than with a monolithic chip. This was executed by AMD in its processors from its Ryzen 3000 on the desktop, however, in graphics cards we will see it from the Ryzen 7000 with the RDNA 3 architecture.
The new design of the Radeon Technology Group will divide the GPU into two different types of chips. The first one called GCD will have what is the GPU itself and the caches up to the second level, a configuration can have several GCDs, but from the second it will be subordinate to the first. On the other side will be the MCD, which includes the Northbridge of the GPU and, therefore, its interface to VRAM, LLC like Infinity Caché, as well as accelerators such as DMA units, display controller, codec. video. etc.
AMD will reach 3 GHz with its 3x Navi GPUs
First of all, we must remember that the information we are going to comment on is not official from AMD, but comes from what has been leaked as well as a series of speculations. So the information you see in the tweet above is a mixture of information and speculation.
The most powerful GPU under RDNA 3 architecture that we will see will be Navi 31, which will have a configuration of 60 WGP and, therefore, 120 compute units. 50% more than the current RX 6900 XT. The RX 7900 XT will consist of two GCDs and one MCD with 256 MB of Infinity Cache, which would be installed in V-Cache mode. The three chips will be on top of a common interposer and with a configuration of 16 GB of GDDR6 with a 256-bit bus. As for his clock speed is 2.5 GHz.
Secondly, we have Navi 32, which we are supposed to see in the RX 7700 XT, its configuration is 40 WGP and, therefore, of 80 calculation units, the same configuration as the current Navi 21. However, the similarities end there, since it will be a configuration of one GCD and one MCD, when the latter has a 192 MB Infinity Cache. As for the bus with memory, it will be 192 bits, so it will have a configuration of 12 GB of memory. The Boost of these GPUs could reach 2.6 and 2.8 GHz.
Finally, the third GPU will be Navi 33, with a configuration of 16 WGP and, therefore, 32 compute units, the same as the current Navi 23. The difference would be that he would have a 64 MB Infinity Cache. This AMD GPU will go to 3 GHz maximum speed, being the first graphics card in history to achieve it. Your VRAM configuration? 8 GB of GDDR6 a bus from 128 bitss.
Navi 33 could be the first to appear
Apart from the leaked rumors, the information that has emerged in recent weeks tells us that Navi 33 could be the first GPU under RDNA 3 architecture to also appear as a monolithic chip under the 6nm TSMC node, while Navi 31 and Navi 32 will use TSMC’s 5nm node. This makes us think that Navi 33 could be a rebranding with more cache.
In any case, what does not suit us is the configuration of 60 WGP with 2 GCD and 40 WGP for 1 GCD, since we should see configurations of 80 WGP with 2 GCD and 30 WGP with only one. There is still time for AMD to release information, although we are awaiting the first details of RDNA 3 at CES in preview.
