The main reason for using 36 instead of something bigger is to reduce cost. However here are some of the reasons they think the reduced GPU hardware cost is worth it, and how it's doable:
1. They are implementing variable frequency and pushing GPU frequency really high, with the claim that, at the same teraflop capability, a smaller GPU with a higher clock will give better overall real world performance than a bigger GPU with a lower clock.
2. They are implementing variable frequency in a non-typical way. Instead of basing it on thermals, they're basing it on power usage. That way performance will be consistent across units since it won't be affected by the environment they're in, and from a dev prospective, situations are repeatable and consistent.
3. In situations where the CPU is being stressed to the max but the GPU isn't or vise versa, power split can be shifted to one part more than the other in order for it to sustain the highest clock possible, where the other part gets downclocked without much impact to performance since it wasn't being fully stressed, freeing up unneeded power.
4. In the worst case scenario where both GPU and CPU are being fully stressed and a downclock is required, a small downclock results in a relatively big reduction in power usage, so the downclock should be minimal.
5. The more CUs you have, the harder it gets to use them very efficiently.
6. In terms of thermals and noise management, they've commented on how current gen cooling wasn't adequate and that it would be better this time around, with the details on cooling being revealed at the teardown. So, at the very least it should be better than last time.
In theory, this sounds very smart. However, real world performance is the real judge and it will be very interesting to find out. If this does work out well then yes, a 52CU Xbox that uses this same idea would perform better than a 52CU Xbox that doesn't, i.e. it would give more bang for buck.
CPU performance comparison is a lot more straightforward between the two consoles. (with multithreading enabled) The difference is less than 3%, that is beyond insignificant. However in the worst case scenario, the difference could grow to say 10 or 12 percent. Xbox also has a 3.8 CPU clock without multithreading, which could be useful during the transition period to next gen.
Now all this is purely hardware, and theoretical from our prospective, software and software features will play just as significant of a role in real world performance and percieved quality, Sony hasn't talked about VRS or using AI for instance, where Microsoft has. Xbox could also possibly have a significant advantage in hardware ray tracing capability. Again, given the difference in approach, real world performance is going to be the real judge.
Now going back to the beginning, the reason is to reduce GPU cost. However, this doesn't necessarily mean that the PS5 itself is going to be cheaper than Xbox, as Sony have invested much more in storage capability and Audio, and possibly other things we don't yet know about, like controller features or VR related hardware being built in...etc.
So basically, in terms of hardware, seems like, with the above considered, Sony thinks a much faster storage system and innovation in audio will be a bigger game changer (and it very possibly could be), where Microsoft thinks a bigger GPU, possibly significant advantage in ray tracing performance, and slightly faster CPU, with locked clocks is the way to go.
Personally, from a hardware prospective and being mainly PC, I'm currently finding the PS5 more interesting than Xbox. Reason being that the Xbox is more straightforward and just seems like a worse version of the PC I'm building later this year, in every way. Where PS5 might actually provide hardware capability in some regard that I won't find on PC for at least a while, like the audio engine or the redicioulsly fast storage system, maybe the controller features and whatever else.
As a side note, I'm finding the difference in approach between the two this time very interesting, however I wonder if the significant difference will result in neither of the two being fully utilized.
Alright, I'm done taking a shit. Time to end this.