The RDNA 2 architecture has undoubtedly been a major step forward for AMD and as such, it has managed to catch up with NVIDIA in almost every aspect. But according to what is being rumored, the Lisa Su’s could give a blow to the table as never seen before, since its new RDNA 3 architecture and its Navi 31 chip could mean a performance increase so big that it would destroy the gaming market as we know it today.
We have been reporting for some time about AMD’s plans, its patents, and new technologies that are going to be implemented by and for RDNA 3. As we already know, the red firm’s next flagship chip will be Navi 31, but what we could not expect is that its performance would skyrocket to the point of changing the current paradigm of GPU evolution.
Everything is going to hinge on the well-known multi-chip MCM configuration that we have been talking about for almost a year now. Speculations are now almost a confirmation and as such, AMD’s high-end GPU would implement 2 chiplets, mainly regarding Navi 31 and Navi 32, being in total size by logic much bigger in terms of physical space on the PCB.
It is rumored that the total area for such chips will be almost double that of the current Navi 21, which is surprisingly large considering that they are expected at 5 nm under TSMC. Such rumors point and spin finer: it would be two chiplets with 80 CUs each and an IO die as if it were a Ryzen processor.
On the other hand, Navi 33 as a midrange and upper-midrange chip would have an impressive monolithic die with 80 CUs. Early performance estimates for the biggest of the siblings stipulated an impressive 2.5X performance increase, but just upon release, this information was further refined: it could be almost 3X higher than what was seen in Navi 21 and the RX 6900.
More than 10000 cores, 15% higher speed, and lower power consumption.
The leap from TSMC’s 7 nm to the 5 nm that will be implemented in these 3X series chips will mean a series of radical improvements. Starting with the possibility of launching a monolithic chip with 80 CUs with lower power consumption and higher speed, but above all, launching two chiplets without the area being a problem.
The 160 CUs in total of Navi 31 would mean 10,240 Shaders/cores, which added to the 15% speed improvement for TSMC’s N5P node and the 30% power consumption reduction could mean a performance jump like the one we have named.
In addition, and as a last rumor, there is speculation that AMD would be at NVIDIA’s level in a section where it is lagging behind: Ray Tracing. The problem is that Navi 31 would perform like the current GP102 of its rival, so it is to be expected that, although the jump in performance is remarkable, they are still behind.
The biggest problem is that both NVIDIA’s and AMD’s new GPUs are expected by the end of next year or even early 2023, probably because the shortage of chips is an added problem to the technology and developments themselves.