PS5: Leaked Design And Technology [RUMOUR]

[Pemalite]

KBG29 said:

EricHiggin said:

I was also thinking about AMD's GPU roadmap and how Navi was said to have next gen memory (GDDR6) and scalablility. While that could mean different things I wondered if it could potentially hint towards multiple GPU chiplets like the Ryzen CPU chiplets.

There are other rumors at the moment that Navi 10 is still partially GCN and partially RDNA, and that Navi 20 next year will apparently become fully RDNA. With AMD mentioning that PS came into the picture after Navi had been in development, maybe the reason for the hybrid GCN/RDNA was for them? Since PS can add future advancements to their planned past/present GPU architecture using AMD's semi custom branch, maybe it's possible that PS5 could use smaller multiple APU chiplets based on Navi 10 that incorporate a Navi 20 advanced IF chiplet design?

The advanced IF could cost more but likely shouldn't be more costly than fabricating an 8 core, 72 CU, monolithic APU, and it's losses due to yields. If it works for CPU and can have costs as low as they do, then surely it could work for GPU as well once the die get's large enough that yields are a problem. Could it work for smaller separate APU's though?

Did you see Apple at WWDC?

Radeon Pro Vega II Duo, two Radeon Vega II's connected via Infinity Fabric!

People have underestimated AMD so much, but the road maps have ben telling us this stuff is comming all along. I brought up Scalability meaning they were going to make a huge push towards Portable and Mobile devices with Zen and Navi, and now look at the Laptop space, and the deal Samsung just signed for their Mobile products. 

I believe more and more that PS5 is going to be a much more scalable platform than we saw with PS4 and PS4 Pro. I would not be surprised to see Sony take advantage of Infinity fabric for a future Premium tier PS5 revisions. Ryzen and RDNA are already perfect for a Mobile PS5, Sony has to have been thinking about this in the development of PS5, especially with the state of the console market in Japan. Don't be surprised if the rumors FoxyGameUK of a Portable PlayStation come to fruition. A stand alone PlayStation VR device like Oculus Quest running on Ryzen and RDNA is another real possability. A low power Ryzen/RDNA Mobile chip would be a perfect replacement for the garbage MediaTek SOC's in their Bravia TVs as well. 

Sony and AMD might finally be in position to achieve what Ken Kutaragi was aiming for with Cell and PlayStation. Exciting times lay ahead. Sony, Microsoft, and AMD are ready to rock the boat hard, and greatly realign the tech industry over the next decade. 

If you think two Vega 2 Pro Duo GPU's are relying on a 84GB/s bandwidth link for all communication... Then respectfully I need to tell you that you are greatly misinformed.

AMD is using the infinity fabric for the GPU's to communicate directly... To assist in keeping the workload in sync, AMD has always taken this approach even with the ATI Rage Fury Maxx back almost 20 years ago... Prior to using Infinity Fabric with dual Vega chips AMD would leverage something like PCI-E for their cross-chip communication... Like on the 4870 X2.
https://www.cnet.com/news/amd-to-nvidia-two-chips-are-better-than-one/

Think of the infinity fabric as like a modern day crossfire bridge built onto the card itself and then you might be starting to understand it's purpose.

84GB/s is just not a replacement for the 512GB/s or more of bandwidth a modern GPU needs at the memory level to keep things fed.

That doesn't mean AMD can't take a chiplet approach to GPU's, but infinity fabric isn't fast enough to be the only interconnect, the chips need insanely fast access to off-chip caches and DRAM... Which are much faster than what infinity fabric can possibly provide at this stage.

For CPU's that isn't much of an issue, the CPU's only need to communicate with themselves and the I/O, the I/O chip can handle communication to DRAM and so on... And for that 100GB/s is more than sufficient as DDR4 typically tops out at around 50GB/s in dual channel configurations anyway.

EricHiggin said:

Pemalite said:

Pretty much! I mean, if the APU chiplets were significantly smaller with like 8 CU's or less, then it would be entirely possible as 100GB/s tends not to be the bottleneck but the 25-50GB/s of bandwidth to DDR4 is.

I think with a next-gen infinity fabric, allot more things will become possible on this front.

I would assume a single core APU wouldn't make all that much sense, so the smallest would be a dual core APU? With PS5 having 8 cores, that would mean 4 APU chiplets, with 32 GPU CU's total. With Pro having 36 CU's, even with Navi arch gains, I can't believe this would happen because the GPU probably wouldn't have the grunt necessary to make a worthy next gen console.

Now if each APU only had 1 CPU core, then that would be a 64 CU GPU total. Which is 8 chiplets, just like Epyc.

A single CCX is what makes the most sense, otherwise AMD would have to redesign some of their core layout and that could get expensive... Then again, they might have done just that... The consoles are using semi-custom chips, so some extra design work has gone into them anyway.

Two die-harvested 4-core CCX is a possibility though... But pretty unlikely.
CU counts is where things can get really muddy, just not enough information to base anything on. But 40-50 CU's is a good ballpark with aggressive clocks.