PS5: Leaked Design And Technology [RUMOUR]

[Pemalite]

EricHiggin said:

Based on the rumors that PS5 is using either a monolithic APU or dedicated chiplets, what if PS and AMD combined the APU and chiplet idea?

Basically the rumors stated that anything can happen? You don't have much choice than those two options.

EricHiggin said:

Instead of having one big APU or many smaller dedicated chiplets linked with infinity fabric, what if they had two smaller APU 'chiplets' linked?

That is one possibility, but won't happen.

You will likely have the I/O, CPU chiplets, GPU chiplet... That way AMD can optimize the process for the chip designs better.

EricHiggin said:

Since we know it has Ryzen and Navi, each APU 'chiplet' could have 4 Ryzen CPU cores, and maybe 36 Navi GPU CU's let's say (the same amount as the PS4 Pro GPU). Take both APU 'chiplets' and link them together with infinity fabric, and use an I/O die if necessary. This way you would end up with 8 CPU cores and 72 GPU CU's, and could do so much more cheaply because your yields should be remarkably higher due to having much smaller chiplet APU's instead of one big monolithic APU.

Maybe this isn't technically possible or just wouldn't make sense, as a monolithic APU or dedicated chiplet design would be better overall possibly?

The infinity fabric likely doesn't have the appropriate bandwidth for GPU chiplets of 36+ NAVI CU's to work together.
100GB/s isn't enough... Especially as said GPU's will be accessing bandwidth multiples more than that to local RAM.

It is a good idea in theory though. But CPU's tend to be less bandwidth intensive than GPU's.

Lafiel said:

Yea, large monolithic chips hopefully can be replaced by clusters of chiplets in the future as that would mean much less wasted silicon waver area and should result in better prices. The problem is the bandwith of Infinity Fabric (IF), which afaik isn't quite high enough to seemlessly link several GPU chiplets into a big GPU. Even if AMD has a much more advanced IF rdy to go it could be prohibitively costly, as they don't seem to use it in their upcoming Navi GPUs (the one Lisa showed on stage at Computex was monlithic).

There are engineering ways around it though.
AMD could set up a direct high-bandwidth link from the GPU chiplets that avoids the infinity fabric direct to DRAM or another piece of logic and have them work together there... And use the Infinity fabric for local chip communication that tend to be smaller transfers.
But that approach would get complicated pretty quickly.

Either way, for the GPU... The monolithic die is here to stay for the immediate future for GPU's, AMD simply hasn't shown they have solved the limitations yet.

EricHiggin said:

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?

AMD is trying to distance itself from GCN. RDNA is very much built on the foundations of GCN... Just like VLIW4 was built on the foundations of VLIW5... Remember AMD was championing how Vega was one of the largest deviations of GCN at one point as well.