We are facing one of those rumors that border on fake on the one hand and that on the other have part of sense, enough to launch this article in reference to what we could expect from the new top-of-the-range GPU based on AMD based on RDNA2. The rumor spread earlier in the week and was rated by some people as fake, but the greater number of data contributed today gives it an interesting credibility step, will RDNA2 be 50% faster than RTX 2080 Ti?
AMD has become truly watertight with its new products. Not surprisingly, the information comes to us with drops and any detail, silly as it may seem, is practically news of a mandatory presence on the Internet. But today we have one of the so-called «Big Leak» and it is nothing more than the supposed (I imply in supposed) technical characteristics of the new high-end GPU from AMD, is it as good as it seems?
Take it with an entire rock of salt, since what has been said from now on is as optimistic as possibly erroneous, but at the same time it has certain frames that we have been listening to some time ago.
AMD's new RDNA2-based GPU would come with a fairly compact 427mm2 array, and while we've calculated between 450 and 460mm2 for 80 CUs, some rumors about adding units for Ray Tracing claimed that AMD could save space on the matrix with the new architecture.
Part of this reduction could effectively be in the announced 72 CUs in total for the top-of-the-range GPU, which could be divided into two clusters of 36 CUs each, that is, the same CUs as the RX 5700 but divided into two sections. independent for the same physical chip. These numbers would give us a whopping 4608 Shaders.
These 4608 Shaders would be attached to a 384 bit bus for its GDDR6, so AMD would rule out using HBM2E for its high-end, something that we do not see from the 7000 series. If this were correct, we are talking about 12 memory controllers that would suppose configurations of 12 GB of VRAM.
As for its speeds, the sources claim that it would come with a 2.05 GHz Game Clock and a 2.15 GHz Boost Clock, something that in Navi 10 as a general rule they end up using a frequency closer to the Base Clock than simply the Game Clock. If this has changed in RDNA2 we would have a performance in FP32 of 18.89 TFLOPS for the first and 19.81 TFLOPS for the second, more than double the current RX 5700 XT if we tend to compare performance in the simplest way.
Consumptions are around 285 watts and 310 watts, with an average of 300 watts finally, which would be 33% more than the RX 5700 XT. Last but not least controversial, sources indicate that this new RDNA2-based graphics card could outperform an RTX 2080 Ti by 40% to 50%.
Considering that currently and on average an RX 5700 XT is over 34% below the NVIDIA card, the performance jump would therefore be between 74% and 84% respectively for the RDNA2-based card. These numbers are feasible for these features, the problem is that they would really fall short, since this new GPU gets 80% more Shaders, with various performance improvements for pure architecture, and as for TFLOPS we talk about + 103% vs. an RX 5700 XT.
Therefore and in theory, these numbers would affirm two things: either the leaker places the difference between the NVIDIA card and the RX 5700 XT with a greater distance between them in terms of performance, or a GAP of approximately 16% is missing and the Difference should go to 56 or 66% compared to the RTX 2080 Ti. We will see in future leaks if any of the data shown is confirmed, although for the moment it is better to be cautious and not take anything for granted.