4 mores years of dev experience, software development and hardware development in ML/AI & RT/PT (just think how far we've come on these in the last 5yrs in the PC space).
No widespread increase in target display resolution (4K) which means half the frame budget will no longer getting thrown away just on pixels, in fact with advanced, mature ML upscaling/frame-gen we'll probably be rendering less pixels and getting better IQ.
Engines which over the next 4yrs will better integrate Mesh Shaders and/or Virtualised Microgeometry, the aforementioned ML/AI & RT/PT; as well as better utilise the SSD-I/O functionality.
Cheap stackable cache and additional cache dies, GDDR7 chips that'll be much more mature than GDDR6 was for PS5, N2/N2P/A16 will be 18-24mths old by mid 2028 and backside power delivery (BPDN) may be a prospect (preliminary tests show 10% higher clocks and 30% more power efficient as a baseline) and finally multi-GCD chiplet SoCs; which will likely be much more common by then and could cut cost by 30% or so. Frankly I'm not sure if it's even possible to make an APU with an adequate boost even taking into account all the prospective efficiency gains available. I think Chiplets may be a necessity.
Chiplets would be a big part of the solution simply for the scalability they offer. For example if the PS6-based portable rumors end up being true, it'd be much smarter to have a scalable chiplet design that could fit both the regular PS6 and a PS6-based portable, if it has great granularity.
Why dedicated separate wafers for two wholly different products (APU size-wise) when you can streamline shared wafers for as many components as possible and save the separate production pipeline stuff for later in the product assembly stage?
There's a lot of areas to cut costs too. RAM likely won't need to go up that much and GDDR7 will offer a lot more flexible configuration options to dial in the sweetspot. The SSD could just do a straight doubling to 2TB @ 11-13GB/s via 8Ch on PCIe5 which will be old tech by then and be cheap. I doubt more than 11-13GB/s RAW will be of any use. The optical drive can be offloaded to a readily available $50 add-on for all SKUs and the console could put the extra into better core hardware. They can settle for Wi-Fi6e rather than 7, Bluetooth 5.2 will be perfectly adequate, USB4 will be mature and HDMI2.1b @ 48Gbps will likely suffice.
Yeah, things like the SSD I don't see getting a massive boost in performance in terms of raw bandwidth for 10th-gen, there is no real reason to. So a drive that's internal m.2-based with maybe 16 channels (I think more parallel channels would probably be beneficial for I/O performance) at 12 GB/s could do just fine. It's data decompression that would see the bigger improvements, not just for the SSD but also on the GPU for graphics data.
That way like you said, there wouldn't be as much a need for a big increase in RAM capacity or even bandwidth, particularly if they could move processing closer to the processors with cache & near-processing buffers (like PNM). 32 GB @ 768 GB/s would be a good target. Maybe a bit more bandwidth if they pushed for 26 Gbps chips, but my assumption is that they'd want to stick to 256-bit and would probably also like to keep the number of modules to a minimum while increasing capacity.
For those reasons I could see them using GDDR6W instead of GDDR7, and just pushing the clock for the GDDR6W memory a bit higher to hit say 832 GB/s (which would almost 2x the PS5's bandwidth)
Infinity Fanout Links mentioned a few times previously could allow for two small GPU GCDs on an advanced node like A16 with BPDN, then savings could be made on a 16C Zen 7C made on N2P with a once-again cut down FPU, SMT-logic deleted, a relatively small L3 cache on-die with more stacked on top, clocks in the 4.2-4.5GHz range; likely providing 2.5-3x the base PS5's Zen2 capability per core. Old node N5/N4 MCDs could be used for the SoC/GPU L3 caches and for offloading memory buses, perhaps better facilitating different memory bus setups alongside G7 memory chips with 2GB, 3GB or 4GB capacities and I'm pretty sure at least 4GB will be dedicated to the OS, so they could use low bin, slow chips for that much.
A lot of things line up pretty well for 2028. Plus I expect as always there'll be a few interesting curveballs we didn't think of.
I just hope SIE have plans to leverage that power in meaningful ways from early on, and without relying on other platforms (even PS5) to do so. Hardware performance isn't necessarily my concern with PS6; it's the possibility of very few games genuinely leveraging the technology in transformative ways, even among SIE's internally developed titles.
6-7 high-tier AAA games over the course of a generation isn't going to be enough and I say that because I don't think the GAAS titles would necessarily prioritize targeting the hardware the way non-GAAS would (especially if they're also targeting PC Day 1 for those games). The possibility of fewer 3P exclusives means even more of the onus would fall on internal 1P teams to show the capabilities of the hardware.
I think these type of things would heavily inform what in fact the hardware specs target. Is there much a point in developing an uber-graphically pushing & focused machine if you're going to get experiences like TLOU2 or HFW you can count on one hand over the course of 7-8 years? Or do you prioritize the hardware to support what would be the majority of your software output? The latter seems more sensible to me and I need to see what software SIE actually prioritize in the longer-term going forward (and in what ways) to feel assured they take PS6 in one direction vs. another specs-wise.
I mean I'm already down about the likelihood VR will not be a focus at all for them next-gen in any serious capacity; I do question how much simply having yet a more powerful console is going to appeal to younger customers/gamers as a priority. SIE are probably considering the same things.
I think we'll see 38-40TF Single-Issue FP32 but a significant bolstering of RT/PT throughput, very capable ML/AI hardware and the rest is going to be all efficiencies. 24GB dedicated just to GAME will probs be enough and a 12GB/s SSD could fill it out in 2s. But after that it's gonna be all about system bandwidth/throughput. ~1TB/s on the RAM and a well balanced distribution of L0/L1/L2/L3 cache to further reduce the load.
This gen has been a massive half measure with terrible timing; all this cool stuff likely gets to fly next gen.
Most of that sounds good & reasonable. 24 GB RAM for the framebuffer might be adequate if GPU decompression is better, and if I/O decompression is also boosted. Data optimization will always be improving anyhow; CPU/audio RAM would probably benefit still sticking with the same memory type though so 320-bit bus could be possible.
PSSR will bring the resolution upscaling benefits freeing up processing for other things like RT/PT, but SIE will probably have additional tech in the system specifically targeting that stuff as well.
maybe for ps7
