This one is tricky. Cache/Wider bus, both stand to increase the size of the chip. An alternative would be to just use the same-sized bus but with faster RAM. I expect they increase the cache in the APU vs. the current PS5, but not by anything that defines the architecture of the system. Eg. The PS5 already has a unified cache of 8MB for its CPU and 4MB for the GPU. So I see them making the CPU 16MB and GPU 6 -12MB. I am not sure they can make both the GPU and CPU feed from the same cache.
True, I think the first thing they would attempt, if bandwidth is a need, is to go with faster RAM. It is the simplest solution to take; if a threshold's reached where that solution isn't applicable, they'll consider increasing the cache. If that doesn't do, the last resort is to increase the bus size. Each option prioritizes one thing over the other, even if each one is a successive escalation of sorts:
-Faster RAM (22 GT/s): Mitigates bus width increase (256-bit), mitigates cache size increase (no semi-large 3D V-cache/Infinity Cache)
-Cache increase (3D V-cache/Infinity Cache): Mitigates faster RAM (18 GT/s), mitigates bus width increase (256-bit)
-Larger Bus (320-bit): Mitigates faster RAM (18 GT/s), mitigates cache increase (no semi-large 3D V-cache/Infinity Cache)
And then they leave the APU bus at 265bit. Just don't see them spending die real estate on a larger bus, they kinda already showed their hand with the cache scrubbers, which suggests they would always aim for the more efficient getting more with less approach. And More cache improves the system across the board, both CPU and GPU perf boosts. Basically, whatever the PS5pro is, it won't just be a direct 2+2=4 affair, it would be more like a 2+1=4 type thing.
Ideally, I think the PS5 Pro goes for 256-bit bus @ 576 GB/s, same 16 GB capacity but maybe using GDDR6W (doubles capacity, peak 22 GT/s bandwidth per chip, 64-bit chips; would save on PCB footprint/real estate of RAM modules and power consumption, could be worth it if price is right). CPU is Zen 3 (or 4) with unified 16 MB L3$ (Zen 2 has non-unified L3$ split between CCXs, this could be a reason to use Zen 3 or 4 over Zen 2), around 4 GHz. GPU around 7800 XT (maybe 7900?) with 24 MB or 32 MB Infinity Cache.
Aso, I think people tend to forget that this is a console, at best, it's designed to hit 30-60fps at quality mode and 60-120fps at performance mode while dropping quality. It doesn't need the kinda bandwidth we would typically see in a PC where the hardware is designed to be able to push upwards of 300fps+ at all sorts of resolutions where applicable.
Agree with you here. The IPC gains alone going from Zen 2 to 4 is already going to be enough. Couple that with an up clock to say 4Ghz and that would make for a meaningful CPU boost over the current PS5 even if the core count is the same.
Yeah, and another thing which I forget sometimes is, Zen 3 and 4 have a truly unified L3$; Zen 2 doesn't. I don't know if AMD could customize Zen 2 to have a fully unified L3$ but I imagine the cost in doing that would not be worth it, versus just going with a Zen 3 or Zen 4 design. It's extra work on AMD's part when they already have newer CPUs with the feature, and other performance benefits not present in Zen 2.
Even if that results in a PS5 Pro CPU that is say just closer to 4 GHz instead of 4.5 GHz, at the end of the day, I think Zen 3 or 4 @ 4 GHz is going to give you a better performance than Zen 2 @ 4.5 GHz, in large part due to having a fully unified L3$.
3.5Ghz? Naaaa.... the good thing is that we get to see the `PS5 GPU` on the market for over a year before the console is released. When we see what clocks they can hit on the PC version of the card, we then know where the PS5 can realistically land. My guess is that the PS5 would be somewhere between the boost and base clock of the 7700xtx if it's on a 5nm process. If on 4nm (which is a more advanced node than what the 7700xt is using) then it would either be at the boost clock of the GPU or a little bit higher.
Well TBH I'm saying maybe the PlayStation 6's GPU can reach around that level. But it'd depend on a mixture of things, like future RDNA design features, use of more chipletization, 3D packaging technologies, and switching to more power-efficient memories like HBM2 or HBM3 while pushing even further into localized processing (PNM/Processing-Near-Memory) that's as close to memory as possible (something which in part we've already seen the PS5 do with its SSD I/O, WRT NAND memory).
A lot of that is beyond what can be done with PS5, so no way PS5 Pro's GPU gets anywhere near 3.5 GHz. At best it'll probably hit between 2.65 GHz - 2.8 GHz, but I think that depends a lot on how many CUs are left active. If all 60 are active, I'd assume they're using RDNA4, and could hit 20 TF with a lower clock. But they may only want that if they can adjust the clock for backend logic, in addition to the GPU allowing separate frontend and shader clocks like RDNA3 currently does.
If it's 54 CUs, they are prob using RDNA3 or some RDNA3.5 which doesn't implement separate backend logic clock rates (assuming, again, RDNA4 actually features this) and will go with a faster clock, at least for shader logic, as that would also affect backend logic clock rate. Maybe a bit faster of clock than 2.8 GHz, but not that much faster. It'd peg PS5 Pro just shy of 20 TF, but you'd be getting a punchier increase in pixel and texture fillrate as well as BVH calculations, plus potentially whatever hardware Sony/AMD are using for accelerating RT and ML, or super-resolution, as those would likely be tied to the shader & backend clock rate.
I'm seeing a lot of assumptions it'll be zen 4 why is that? Wouldn't that raise the b.o.m significantly? I think it'll be zen 2 with higher clocks (4.5Ghz) so they can go crazy with the gpu for any raytracing customizations I don't think this machine will be a penny over 499.99 they have to cut somewhere plus I just want to see what Cerny and team can do with a bigger gpu budget
Yeah, I can see them doing this. Zen 2 on more efficient node process should allow for a higher clock, at least to within range of what Zen 2 is designed to handle at its peak.
If, that upper range is still not enough for what the PS5 Pro would need to keep its GPU fed (and still handle all other typical game logic and OS background tasks), then they will use a Zen 3 variant. They might still go Zen 3 or 4 anyway, though, for CPU architecture and feature improvements beneficial for game performance, that Zen 2 would lack. They may also be financially incentivized by AMD to go with Zen 3 or 4 over sticking to Zen 2.
There's also the fact that Zen 2 doesn't have a fully unified L3$; it's split between the two CCXs. Though Sony could maybe get a custom Zen 2 with Zen 3 & 4 fully unified L3$, it would cost them, and not be worth the cost. Considering AMD have two newer CPU designs with the feature already, and additional features for performance gains Zen 2 would lack, Sony would have a more affordable route simply picking a Zen 3 or Zen 4 CPU design for the PS5 Pro.
I personally doubt the Pro will break 4+ GHz on the CPU unless it's like the the base PS5 with dynamic power use. I'm sure the Pro will have the As Series X CPU frequency.
It has to get a clock increase, though, on the CPU in order to feed the GPU its frame data. Otherwise the CPU will be the bottleneck again, similar to how it was for the PS4 Pro.
I can agree with Crispy Gamer that they could potentially stick with Zen 2, but if they do, it's going to get a clock bump to around at least 4 GHz. The improved GPU will need that especially if core and thread counts don't get a bump up.
OTOH, if they go with Zen 3 or Zen 4, it is possible that the CPU in fact doesn't break 4 GHz, or just meets 4 GHz but goes no further. It probably wouldn't need to, due to architecture improvements and, notably, fully unified L3$. Meanwhile you'd still get a big bump in draw call data to feed the GPU for those improved frame rates.
