It has become clear that AMD is there to compete with NVIDIA in GPUs, and although the fight is fierce, the truth is that it has not been ahead in general terms of performance by the minimum. But that could change radically with RDNA 3 since it seems that AMD would have their graphics cards with MCM design ready, which means that they could double the power in a generation, all thanks to the new chiplet-based Navi 31 chip.
The Navi 31 chip is not a novelty as such, since it was already leaked on previous occasions, but the truth is that, although it has been linked from the first moment towards the RDNA 3 architecture, it has not been until now when more light has been provided to this matter.
AMD RDNA 3 and Navi 31, more efficient and with MCM design?
Nothing I can confirm 100% now, but from what I know Navi 31 is a 80 CU chiplet and top SKU has 2 of them.
— Kepler (@Kepler_L2) January 1, 2021
We all know that Intel is going to debut this year with its new graphics cards for gaming and HPC hand in hand with its Xe architecture, which is based on Tiles or, in other words, on chiplets. This seems to be the paradigm shift of the GPUs of the future, which will be divided into several of them to increase power while lowering the total cost and maximizing the chips per wafer.
This is a fact, as was stated by the executive vice president of AMD at the time, where he already affirmed that the performance commitment per watt was the same in RDNA 2 as in RDNA 3. What is the news then? A tweet from Kepler_L2 where he comments that RDNA 3, presumably Navi 31 has a design based on chiplets with 80 CUs for each of them, where the RDNA 3 TOP GPU would have two of them integrated.
In other words, we're talking about a whopping 160 CUs in a package and GPU that would give a total count of no less than 10,240 actual Shaders.
Navi 31 and RDNA 3 would bring an improvement in ray tracing to match NVIDIA
Patent: Computer resource scheduling using generative adversarial networks - AMDExascale resource scheduler for EHPMore details: https://t.co/QpVUg22b3n pic.twitter.com/AQYwoXbHvq— Underfox (@Underfox3) January 22, 2021
The second rumor is much more unfounded since it comes from an AMD patent, in which new technology is patented to synchronize the workload between the different GPUs that are based on an MCM design.
As if this weren't enough, it appears AMD will have a new command processor for the next-generation ray tracing pipeline. This, together with the creation of a new technology that competes with NVIDIA's DLSS 2.0, as well as the units to work this technique of super sampling by AI, should give a qualitative leap to AMD in order to compete with the supposed TOP GPU of your rival.
At the moment, the company has not released any statement in this regard, since it is carrying everything with great secrecy, but what seems clear is that it will be late for the war against Intel and perhaps at the same time as NVIDIA, because these are not expected. new GPUs until 2022, although the announcement is made this year.
That is, they would be presented at the end of the year, but they would not officially reach the market until 2022, where the company seems to have already booked production of the new TSMC EUV node, which it is manufacturing at 5 nm.
0 Comments