thicc_girls_are_teh_best
Member
Indeed spreading load needs to happen with purpose built co-processors that do things better than a single processor. But if you're talking like for like i.e. more cpu cores vs less but faster cpu cores than simple is probably still preferred where available. I'm not technical enough, just looking at it from a common sense perspective. Besides Cerney himself said it in the presentation, unless he's some kind of hack or spinning PR.
No Cerny is definitely not a hack, that much is safe to say xD. But you also have to understand the position he is in, they will try to embellish a few things here and there where they can. What he says isn't inherently false, but it doesn't take into consideration that there will also be devs who will of course work with saturating a wider and slower setup as well, especially if they need to.
I think it's also worth considering that both systems actually use both approaches in selective areas, when you break it down.
-CPU: Same 8C/16T setup for both but XSX favors faster speed and can go "narrow and fast" when disabling SMT
-GPU: PS5 favors narrow and fast, XSX wide and slow, but it remains to be seen how this will really play out. AKA PS5 favors speed but XSX favors bandwidth
-RAM: PS5 actually takes the wide & slow approach here while, for graphics-orientated tasks XSX favors narrow & fast approach. This is probably MS's way of making sure the GPU is fed with enough data at fast enough speeds since the GPU is slower than PS5's.
When you combine the GPU & RAM together PS5 seems to favor faster rate (2.23 GHz GPU) of smaller set of unique calculations (36 CUs) on a wider body of unique data assets (14 or so GB GDDR6 minus OS requirement) more slowly (448 GB/s memory bandwidth), while XSX favors slower rate (1.825 GHz GPU) of larger set of unique calculations (52 CUs) on a narrower body of unique data assets (10 GB GDDR6, high priority VRAM partition) more quickly (560 GB/s memory bandwidth)
-SSD: PS5 favors narrow and fast, XSX favors wide and slow. Since both systems can memory-map partitions of their SSDs as a v-cache, and the GPUs will be addressing that partition directly, that factors into the GPU clocks. In other words, PS5's GPU is probably higher clocked because it also needs to keep up with the faster SSD speeds reading texture data from memory-mapped v-cache, but if any of that texture data has to be altered or transformed it likely still has to go into RAM because the data from NAND is stored compressed in block format and can't be byte-alterable or bit-alterable on the NAND directly.
Same limitation applies to XSX, but MS mentioned something about real-time texture alteration with ML hardware. Not known yet if PS5 has this or something similar (the decompression they mentioned isn't the same thing, but it's basically like MS's decompression yet more powerful, again because of necessity due to SSD speed)
@dark10x btw looking at your PS5 spec page I see that the die size is missing. Is that not provided to you? Or is it part of the second round of reveals or something. As I'm suspecting although PS5 36CU < SeX 56CU their die size could actually be other way around with all the PS5's 62% enlarged CUs and custom made on-die I/O chips and all.
If true that means Sony not only put 20CU worth of custom hw there but also their 36 CUs are somehow supercharged and not exactly base RDNA2 CUs at all, like the ones MS is using for their brute force approach.
MS is not using a "brute force" approach; both MS and Sony are using pretty customized and optimized approaches. They emphasize a few different things in certain areas and each have a few notable advantages over the other in given areas, but somehow this myth that MS has gone brute force has spread wildly when it's been very misleading.
A brute force approach would've seen a system well more than 12TF and with even faster CPU cores (and probably more of them, too). That hasn't happened. Cerny's mention of larger CU cores was in relation to PS4's CU cores, and more reflective of the general size increase in CU cores for Navi over GCN.