Lmao only 24GB, I hope to God it's more than that. That's like them going from 512mb(total) with the PS3 to 768mb with the PS4, or 8GB with the PS4 to 12GB with the PS5. 1.5x more RAM is weak sauce, even the Switch 2 brought 3x more ram.
I think there's little need to push beyond 24GB for a launch PS6, assuming the 24GB is fully dedicated to games (they can have a separate small/slow pool of memory dedicated to the OS) and assuming they implement L3 cache for the APU/GPU.
A full 24GB of GDDR7 would be nearly twice that of the base PS5's available RAM for games while also offering ~1TB/s of bandwidth and then paired with a decent L3 GPU Cache the effective overall throughput in well-optimised games could be more akin to 1.3-1.4TB/s, have significantly lower latency, reduce contention and reduce power consumption. If the SSD sees a simple bump to PCIe5 and with it a 2x speed increase, then after the framebuffer and other more persistent resident data, the entirety of the remaining space could be filled in well under 2 seconds (factor in on the fly decompression that's even better than what we have now plus a half-decent streaming system and it's effectively instant).
I think 24GB GDDR7 (8x24Gb) over a 256-Bit Bus with 960-1024GB/s of bandwidth is the smart play (alongside 4GB DDR5 for OS), then put every penny possible into making the main chip/s as capable as possible, incl. getting as big a chunk of L3/Infinity Cache as they can.
I believe the L3 cache in this scenario can provide 1.5-2.7TB/s for 32-128MB of data with a latency about 1/30th to 1/10th that of GDDR7 RAM and significantly lower power use (about 1/10th to 1/3rd). RT/ML workloads will favour this and resources may be better spent on overall throughput and efficiency rather than larger chunks of RAM. If they go with 32GB, they'd likely forgo the extra cheap RAM and use ~4GB of that for the OS and the real extra amount usable would only be ~4GB; with the extra cost they may be more inclined to go with slower RAM chips and less or no L3. I think faster RAM chips and a big chunk of L3 would be a better place to allocate those pennies. AMD's L3 Infinity Cache is a major efficiency gain and would in my eyes be an absolute no-brainer for any console going forward. Frankly I think the best thing Sony could do is turn this thing into a Cache Monster as the last console with a good level of bandwidth relative to its GPU was the base PS4 and the only viable way I see us getting a similar relative throughput at this stage is with cache rather than more or faster RAM.
Also, depending on timelines, cost and availability, 32Gb/4GB chips may not be plausible for PS6's release either, so they might have to go for 16x16Gb/2GB in a clamshell config to hit 32GB (as opposed to 8x24Gb/3GB to hit 24GB), which would not only be a high volume of chips but itself results in less efficient memory access, more complex routing, potentially-slightly higher latency and more thermal complexity. Sony have done this before for PS4, but it's preferable not to.
