A new leak reveals something unprecedented and truly surprising about what Sony is preparing with its new console, or at least the mid-life review of the current PS5. And it is that the new PS5 Pro will not be a leap forward in performance, but possibly two, at least, if we take into account the amount of data that has been leaked. And it is that this PS5 Pro would have one APU that would hold almost the power of a Ryzen 7700 linked to the performance of the hypothetical RX 7800 capada, or RX 7700 limitedwould also include ddr5.
AMD and Sony prepare a punch on the table that would correspond almost more to a console generation change than to an upgrade as such. At least, if the figures are close to what we are going to see next, which, as usual, we will try to elucidate based on our own speculations based on the leaks.
PS5 Pro will be a beast according to new leaks
It is difficult to believe, so let’s go with the arguments that are exposed from RGT. First of all, AMD would have created a new APU based on a lithographic process that could be 5nm or 4nm, in theory, the latter as we saw at the time, which would explain the rest of the following argument.
With this in mind, we once again confirm the existence of 8 cores under Zen 4 and one iGPU with 60 CUs (30 WGP) based on “RDNA 3.5”. In other words, it’s a customized Strix Halo APU with reduced performance and consumption, but this brings many advantages.
The first is the frequency, since on this occasion it is specified that it will exceed the 3.6 GHz for CPU cores and, surprisingly, it would go as far as 2.7 GHz on the iGPU.
The specifications on the latter also detail that it would have 96 ROPS, which represents an increase of 50% in this very key section. In addition, and to finish the section on leaks, it is stated that it will have 16GB of GDDR6 VRAM exclusive for the iGPUwhile getting DDR5 (LPDDR5?) for the OSwithout counting improvements in the SSD, which we understand would be in its capacity or speed.
That being said, we are going to deal with two different hypotheses from now on about what the PS5 Pro and its SoCs could be.
Hypothesis 1: Your APU would be a mix of a Ryzen 7700 + RX 7800
With this on the table, what we can launch as speculation is that the APU, based on Strix Halo, will be a kind of Ryzen 7700 gelded with a limited RX 7800. In the CPU section, the differences should be minimal and everything should be reduced to the filtering of frequencies due to pure consumption and heat.
The problem, and the dilemma we have here, is the iGPU. In theory, be careful with the salt, that here we jump directly into the well, is that we have an RX 7800 limited to 3,840 Shadersbut in 6 Arrayswhich maintain the proportion of the original PS5 in 10 CUs and 5 WGPs for each of them, respecting the L1 assigned to them.
Therefore, it would go from 4 SA to 6 SA with 60 CU, where also by including 2 more SA it goes from 64 ROPs to 96 ROPs. Interestingly, pure speculation on our part, no more PHY will be added to the equation and will be respected 8 existing 32-bit (256 bits in simplified IMC), keeping the 16 GB of VRAM, except that now the iGPU has all the bandwidth exclusively for it. Increasing ROPs by 50% and jumping from 36 CU to 60 CU would require it, keeping costs down here.
That is why we talk about DDR5 now, because the CPU is going to have its own exclusive IMC/PHY with this type of memory, which will be minimal in capacity, since to work with Sony’s OS it will hardly require speed as such, nor too many GB, at least, on paper.
Performance of this play by AMD
Having said the numbers, let’s go with the theory and then we will return to our speculations based on what has been said. If everything is fulfilled, and for now it seems that it is on the right track, what we would have in terms of performance is just over 40 TFLOPS in FP32which is almost 4 times more than the original PS5, but where’s the catch?
Well, in two key aspects. How is AMD going to respect the proportions TOTALS (important, and that’s why it’s capitalized) in terms of number of units on PS5 and PS5 Pro according to AS, ROPs and L1, given the jump of cores / Shaders in total, It seems really difficult for us to reach 2.7 GHz without triggering consumption and temperaturesvery close to the limit already on PS5, and TSMC’s 4 nm doesn’t work miracles either (ask NVIDIA).
Therefore, we believe that, finally, in this hypothesis 1, frequencies would be lower. My personal theory is that they will be, curiously, above 2.3GHz, giving a performance of 35 TFLOPS, that in the best of cases. But a server does not rule out that AMD respects the frequencies of 2.233 MHz of PS5, Because? Well, because it would be the easiest way to respect the performance per Array given the jump of L1 and ROP for each AS.
Hypothesis 2: what if they repeat the play of PS5 capando a CU by AS to give a PS5 Pro with a limited RX 7700?
That being said, and last lucubration here. Everything said within the CPU section would be respected, without news, but the iGPU section has more “crumb”. Of those 3,840 Shaders in 60 CUs, AMD can do what it did with the original PS5, and that’s disable a CU for each SA. To review what was done and we see in said PS5, of 40 CUs that it integrates, really only 36 “work” (are enabled) giving the 2,304 skillful Shaders and 64 ROPs distributed among the 4 SA.
And if AMD repeats play? Well, out of 60 CU in 6 AS we would have 54 capable, functional, which would give a result of 3,456 Shaderswith 96 ROPs and one ratio per AS exactly the same as PS5that is to say, 576 Shaders per Array.
Therefore, AMD would keep the same proportions for Array, exactly the samewith certain improvements in the architecture, such as adding the AI Accelerator of RDNA 3, the 64 Dual Issue SP or the 1.5x VGPR that we already saw in the aforementioned article that we encourage you to return to if you have any doubts.
All this would boost performance clock by clock, not much as we could see in the comparison at the same frequency between RDNA 2 and RDNA 3 Lite, but it would improve performance per watt. Here we return to the previous section on frequencies. And it is that if of those 3,840 Shaders there are only enabled 3,456then the frequencies could go up, although 2.7 GHz still seems a lot to us of the filtration.
Conclusions: the PS5 Pro would have a limited RX 7700
The two scenarios are on the table, where we bet on the latter, since we believe that the PS5 Pro will have a kind of RX 7700 with 3,456 Shaders, 96 ROPs in 6 Arrays, respecting L1 (32 KB for L1D and L1I), where it will double its size for the console of sony. The question now is, which version of RDNA 3 will AMD use here? Let’s remember that there is RDNA 3 and RDNA 3 Lite, the first in Navi 31 and Navi 32, and the second in Navi 33.
As we already addressed the subject at the time, we will only remember that the problems came from the reduction of the vector register file by SIMD, which was reduced in size from 192 KB to 128 KB, causing the bottleneck that we saw with the RX 7600, highly commented. Therefore, in this case, we would be more before a somewhat shrunken RX 7700 than a limited RX 7800.
If AMD and Sony decide to use RDNA 3 Lite as “RDNA 3.5” the performance will be lower, closer to 30 TFLOPS than the commented 35 TFLOPS. So there are still many doubts in the air to deal with, but at least we already have two hypotheses on the table.