I agree with the latter, now for my thoughts on the bolded:
Functionally, it's when priorities are re-balanced, right?
GDDR's strength is that it offers cheap bandwidth.
GDDR's weaknesses are worse latency and significantly lower density/capacity. Also an energy per bit efficiency that's in between DDR and LPDDR, but I dunno, is that more 'neutral' than 'weakness'? Anyway...
LPDDR would flip those things around; latency that's historically in between DDR and GDDR (but things get murkier in the memory-controller-runs-at-fraction-of-ram-speed era...), win in energy efficiency, and higher density/capacity while losing in raw bandwidth.
For a system like PS5/Series, at the end of the day, bandwidth needs just have that much higher priority over latency/energy/density.
Future-wise... fast forward to PS6.
First, I'll pull up this chart from
here:
It is likely the case that by the time we're talking about a PS6, the options are DDR6/LPDDR6/GDDR7.
Densities and speeds are expected to double. So, GDDR7 would probably look like 2 or 4 GB per 32-bit chip, with transfer rates ranging from 28,000 to 36,000 MT/s. LPDDR6 should probably go up to 12,800 MT/s. Density-wise, if we're seeing a floor of 4 GB per 64-bit with LPDDR5, maybe we're looking at a minimum of 8 GB per 64-bit with LPDDR6.
Energy efficiency-wise, the expectation is that due to the change in signaling type, GDDR7 should reduce usage by 25%, iso-process. So, relative to GDDR6, 2x bandwidth for 1.5x the power draw, iso-process.
As for system priorities...
I think that with the PS6, bandwidth needs will climb again to the point that it's still by far the top priority. So I expect GDDR7; maybe still 256-bit bus width and Sony opting for 4 GB chips, thus living with 32 GB in total.
Fast forward
again to the following generation is when I think priorities can start re-balancing.
1. I'm not sure that the Playstation-style form factor can adequately handle the energy/heat aspects that would come along with a theoretical doubling of GDDR7?
2. I'm willing to bet that further improvements in being more efficient with how much data really needs to be shuffled back and forth (more refined reconstruction techniques, plus whatever other neat ideas*) will lead to the scenario where a theoretical LPDDR7 or LPDDR7X's bandwidth is workable...
...which in turn sees a rise in importance in raw quantity of ram.
*heyyy, just remembered that according to the IEEE roadmap, we're supposed to be seeing stacked SRAM by the end of this decade. Probably not gonna make it in time for a PS6, but by the mid-late 2030's? Fat stacks of cache might be doable. That's one way to help sidestep bandwidth needs.