Current Generation Games Analysis Technical Discussion [2020-2021] [XBSX|S, PS5, PC]

the most powerful car can be beaten in corners, but could be the faster on straight lines.
XsX will shine with its exclusives. Each console will in fact. And 3rd party titles will play good enough on both.

 

The reason that teraflops are sometimes called out as not meaningful is because the figure represents the theoretical maximum number of floating point operators that the GPU can perform but in the history of modern GPUs, few GPUs have ever sustained this for more than a few microseconds because being able to get to the GPU to do that is very reliant on the CPU, excellent code, optimum data, excellent cache utilisation and plenty of bandwidth and literally no bottlenecks and these are all things that are really, really difficult to architect. More so for third party games where the balance of these things are all different.

If you take fairly crude comparisons like number of GPUs CU at any given clock speed, against cache (and percentage of cache hits) and available cache and memory bandwidth per CU, you'll find the current crop of console (4 last gen, 3 this gen) are all over the place in terms of 'balance'.

So teraflops is not meaningless, but it's also not a very good indicator of what actual performance you will get from given hardware.

 

The rated theoretical maximum TFLOPs output isn't the only indicator that GPU's actual TFLOPs throughput during a real gaming scenario. The total theoretical bandwidth to the off-chip memory isn't the only indicator on how often and for how long the GPU is stalling / waiting for that memory, much less bus size (width?). Effective memory bandwidth is what matters here.

But the last mistake here (which I confess I also made at some point) is claiming the PS5's and SeriesX's architectures are "mostly the same" just because they seeming share the RDNA2 instruction set and Dual-CU arrangement, despite being quite different everywhere else. The PS5 has a lower number of ALUs per shader engine, and on top of that they run at 22% higher clocks, so a higher ALU occupancy on the PS5 should be expected.
The cache scrubbers should increase the proportion of cache hits within the L2 (otherwise why bother) and we don't know exactly how that affects the console's effective bandwidth. (We do know why AMD didn't want the cache scrubbers for their PC RDNA2 GPUs though, and it's because they're already spending a ton of die area on LLC in there)
The PS5 having a considerably faster I/O could also mean they don't need to cache as much data inside the system RAM, with assets coming in "on-the-fly" just a handful of frames before they're needed, and we again don't know how that will impact effective bandwidth.

And this is all before we start considering the differences between the SeriesX's VRS and Sony's own implementation for foveated rendering, the impact of the custom geometry processor, the PS5's higher pixel fillrate, etc.


So despite the SeriesX having larger big numbers (max theoretical TFLOPs throughput and max theoretical bandwidth), the fact that the consoles' GPUs are actually substantially different in many of the other resources means we can't really think that "10 vs 12 means 12 is 20% faster" and "448 vs 560 means 560 is 25% faster".



The Vega 64 averages above 11 TFLOPs and 480GB/s bandwidth, and these are about the same big numbers as the GTX 1080 Ti. It never reached GTX1080 Ti numbers AFAIK.

 

The PS5's gpu substantially higher clocks probably help it 'outmatch' the XSX the most. By increasing core clocks everything gets faster, instead of increasing CU counts. Since the XSX and PS5 are rather close in TF (1.9TF diff?), the additional TFs are not making up for the much higher clocks the PS5 gpu has. Its just that RDNA2 really likes extreme clocks. Which was abit of a surprise i think.

 

I *seriously* doubt the XSX gets these spurious max frame times because it runs with a lower gpu clock. Some rarely used function is either extremely slow or something spuriously disrupts something which I consider software problems until proven otherwise.

 

Xsx exclusives ? They have to support xss too, and pc most of the time no ?

 

i meant exclusives that won't release on PS5.

 

Yeah but they have still to deal with lesser hardware. That's the strong point of Sony imo. When a dev will decide to drop ps4/pro, they won't have to have in mind hdds, less ram,etc...

 

*tinfoil hat on* people who write software documentation buys out entire PS5 stocks! *tinfoil hat off*

 

Problem is that there maybe isnt pure series x exclusives at all.

If they have to design around slow PCs + slow xss = compromises.

And even if they would Make xsex only games, those too are dragging the burden of windows+xbox one + xss common tools + directx.

I mean if they would have only series x with custom dev tools + hardware that doesnt have to care about PC/slower consoles maybe they could get "95% out of the hardware" but now "85-90%" as it is not so specialized

(Compared to PS5 with more efficent APIs and dev tools made only for PS5)

Numbers from hat as it is speculation

 

The first post should really be a list with all the games which are compared, a date of said comparison, as well as a short summary or a link to the post where the differences are explained.

For now it's easy to remember which console runs games the best because one console 'won' everything, but it might change in the future. So while the list would be all blue now, maybe in a while, there will be some green as well?

 

Depends on the mode used, like some titles the SeriesX performs better in 3 out of 4 modes.

Also, its a technical discussion, not a list war. If you want list wars then go elsewhere.

 

I highly doubt that, but this is beside the point.
Also you don’t understand, it is not meant as a list war, it’s meant as an index to give a summary about the current situation, and to link to highly informative posts like for example the excellent summary which Shortbread gave a few posts above this.

it also easier to do analysis on the results if presented in a list: maybe there is a pattern of open world games performing better on a certain architecture, or high FPS modes favoring slower memory bandwidth.

with the current situation you can have posters like yourself claiming that some titles, which is plural, perform better in 3 out of 4 modes which I am not even going to get into.
With a clear list visitors can immediately see if such claims are truth, or just wishful thinking.

 

Summary: PC wins all.

 

Just posted about fifa in the df thread, so I looked up videos. This one isnt great (no real like for like comparisons -- different camera angles, lighting situations, cutscenes, etc), but it looks like a hard game to capture like for like on. Going by this video, both consoles seem to dip a little below the 30 target when they switch to "cutscenes" -- but the ps5 has more tearing and a harsh, prolonged dip to 20 (in what looks like a hard workload -- has a bunch of closeup players, and there arent any in the xbox series x cutscenes shown). Both effortlessly lock 60 in the gameplay shown. Also the series S is the real star here imo, has greatly toned down settings but looks like it runs pretty great considering.

(in some shots the grass shader distance also looks farther on xbox, but its impossible to tell really without more similar captures and better analysis)

 

Video got deleted. It was "Dirt 5 - The Next Generation Technical Analysis"
TL;IMI PS5's game modes has resolution tweeks mostly, the tweeks are all over the place on XSX