Digital Foundry Article Technical Discussion [2022]

Status
Not open for further replies.
The situation for the PS development advantage peaked the day before this generation started. It was a given when Sony had a 2.5:1 install base advantage and the better base hardware (PS4 vs. X1) that PS would almost always be the lead platform for 3rd party games. The further we get from that day, that PS dev advantage is slowly eroding. Now they are only outselling Xbox world-wide by about 35% according to latest numbers and they arguably have the weaker console for next-gen compute heavy engines that use all the RDNA2 features. It's certainly moving towards a more level playing field. It's not a given that PS will be the lead platform going forward.

MS made the long-term investment rather than the short-term regarding hardware. The same thing can be seen in their service/software department, look at GamePass, not that promising at the start but oh boy now when its taking off finally.
It also seems that Sony designed their hardware around BC with the PS4/Pro, and probably design started earlier it being a earlier RDNA design as opposed to the Xbox's. The Ray Tracing made it to the PS5 kits at a later stage for example, and certain certain RDNA2 features never made it to the custom AMD design spec.
Sony has had the advantage of brand-name in PAL regions and the larger initial install base and hardware-wise that was a benefit to them during the crossgen period (narrow/high clocks, PS4-optimization advantages). The Xbox though is catching up fast with the numbers, and its hardware is going to show its true colours the more we go forward into the generation. I wouldn't be comfortably riding on the brand-loyality or previous install base successes if i were Sony (and they probably dont), they have some serious competition now. The move to the PC platform has alot to do with that i think.
 
I agree the xbox will show more of its power advantage over the generation, but double resolution is never going to be explained by hardware or feature set I dont think. Even if vrs was somehow allowing double the top line resolution, why wouldn't you make an almost-as-good software VRS solution and get the ps5 ~80% of the way there? Just a weird outlier.
 
I agree the xbox will show more of its power advantage over the generation, but double resolution is never going to be explained by hardware or feature set I dont think. Even if vrs was somehow allowing double the top line resolution, why wouldn't you make an almost-as-good software VRS solution and get the ps5 ~80% of the way there? Just a weird outlier.
VRS implies a compute bottleneck which I don’t think this is. I think if there was a bottleneck happening here it’s at the bandwidth level. Bandwidth can really be a limiter that can supersede FF units, rops, compute power. No bandwidth no high resolution.

but it’s probably just a coding level issue.
 
I agree the xbox will show more of its power advantage over the generation, but double resolution is never going to be explained by hardware or feature set I dont think. Even if vrs was somehow allowing double the top line resolution, why wouldn't you make an almost-as-good software VRS solution and get the ps5 ~80% of the way there? Just a weird outlier.
It's not just VRS, it's a combination of factors.
XSX will have much more free space for computing due to various kinds of hardware accelerators. At the same time, it has noticeably more bandwidth and stable and always achievable 12 tflops + 3.6GHz

By the way, there is an interesting detail in one of the recent videos.
Horizon FW according to DF at 30 fps running in native 4k, while at 60 fps the resolution drops to checkerboarded 1800p (which is usually 900p).
Has anyone noticed that something is wrong here?
The difference in resolution is almost sixfold.
 
Last edited by a moderator:
It's not just VRS, it's a combination of factors.
XSX will have much more free space for computing due to various kinds of hardware accelerators. At the same time, it has noticeably more bandwidth and stable and always achievable 12 tflops + 3.6GHz

By the way, there is an interesting detail in one of the recent videos.
Horizon FW according to DF at 30 fps running in native 4k, while at 60 fps the resolution drops to checkerboarded 1800p (which is usually 900p).
Has anyone noticed that something is wrong here?
The difference in resolution is almost sixfold.
Checkerboard 1800p is not 900p. Its 1600x1800.
 
Last edited:
VRS should result in different pixels. If the pixel quality is imperceptibly different, a simpler shader would be used in the first place, no? For VRS to have a significant impact on framerates, it should be showing as a sizeable delta off base quality in image comparisons similar to how JPEG differs even if it looks the same - that's the whole point, doing less work per pixel to save on processing. Ergo any situation where VRS is considered a contributing factor should be supportable by identifiable areas of lower-quality shading.
 
It's not just VRS, it's a combination of factors.
XSX will have much more free space for computing due to various kinds of hardware accelerators. At the same time, it has noticeably more bandwidth and stable and always achievable 12 tflops + 3.6GHz
An interesting an alternative perspective to view this is that, if there was no issue with programming (fairly confident it is) and you assume this is a hardware deficiency problem - then the Series consoles have a massive flaw in its design; it can rarely spread it's wings.

The issue cannot be compute, there's no such thing as programming things for wider compute arrays, in fact scaling is well below perfect for non synthetic benchmarks; less CUs will always have better saturation than more CUs. More clockspeed is better than less. In any event, given how close the two are in terms of final TF performance, if compute was the issue it would be a 20% resolution differential approximately.
So for performance to fall off a cliff (4x more pixels) PS5 must run into a serious bottleneck.

Once again, looking at all the hardware available, PS5 has more of it in the fixed function areas, so there's actually nothing Series consoles can gain here.

So all that is really left is bandwidth as being this critical bottleneck that can just hamper all areas of the GPU. And we saw this behavior with PS4 Pro. It was unable to keep resolution up with 1X mainly because it largely lacked the bandwidth for that higher resolution despite having all that additional compute. Along this line of thinking bandwidth is the inherent requirement for Series X to outperform PS5 when bandwidth is the bottleneck; no bottlenecks anywhere would result in XSX only having a 20% performance advantage in compute over PS5 assuming it also didn't run into any bottlenecks.

For a game like lego that has very little memory footprint (texture wise, and model complexity) that the Series consoles can store everything into it's rendering pool of memory (10GB). Thereby enforcing that the console is using all 560GB/s of memory and PS5 being limited to 448GB/s in this case.

So if the wall is indeed bandwidth, in all other games where the two consoles are performing around each other, XSX is likely having significantly less bandwidth than the 560GB/s and that's why it can't get away from PS5.
That's a huge oversight if this has any merit; all this power is unused because they split the pools.
 
For a game like lego that has very little memory footprint (texture wise, and model complexity) that the Series consoles can store everything into it's rendering pool of memory (10GB). Thereby enforcing that the console is using all 560GB/s of memory and PS5 being limited to 448GB/s in this case.

Maybe also something to the rumors that games only have access to 12 GB total on PS5, while it's 13.5 GB on Series X?

If so, then memory breakdowns would be 10 GB ram it's 560 GB/s vs 448 GB/s, then 2 GB ram it's 336 GB/s vs 448 GB/s, and finally it's an extra bonus of 1.5 GB ram at 336 GB/s for Series X.
 
And we saw this behavior with PS4 Pro. It was unable to keep resolution up with 1X mainly because it largely lacked the bandwidth for that higher resolution despite having all that additional compute.

I would argue Xbox One-X having 50% more RAM for those higher resolution frame buffers was more of an advantage than memory bandwidth.
 
Maybe also something to the rumors that games only have access to 12 GB total on PS5, while it's 13.5 GB on Series X?

There's also no evidence to suggest PS5 only gives developers 12Gb, especially as that's something we would have seen cause downgrades in loads of other games but we haven't.
 
I would argue Xbox One-X having 50% more RAM for those higher resolution frame buffers was more of an advantage than memory bandwidth.
Reasonable as well. A great deal of many titles
Should have fit in the 8GB requirement, but after OS, 6.5GB may have been too tight.
 
We already saw the Touryst pushing ~90% more pixels on PS5. But here I'd say the PS4 game could have trouble running on PS5 at full clocks properly. We know some PS4 games running on PS5 have actually a profile, some of them running with lower GPU clocks in order to improve compatibility.

the touryst isn’t a good example. The game was ported for Xbox but rebuilt from the ground up for PS5.
 
There's no evidence that isn't the situation. There are some posts out there that does imply they only have 12 GB.
Maybe this is a needed sacrifice to get this mutiapp-feature the PS5 has to work. The Xbox one also had such a feature to share the screen with another app and therefore sacrificed resources so this was always possible (damn, I miss this feature, was great for games like Diablo 3 while watching Netflix). As this feature was removed a few years after launch, more resources were available for games.
4k does not make such a feature less resource intensive.

the touryst isn’t a good example. The game was ported for Xbox but rebuilt from the ground up for PS5.
I never understood why the touryst needed so much resources. It is a great game but visually not very complex. But it was an example back than how easy it is to port a game to the next gen consoles and to even improve them (not that it is as easy to increase the settings like in the PC version....).
But the game has no textures to speak off so almost only the render target might be memory intensive, but even at 10k I don't see a problem why it shouldn't fit into much less memory. But maybe it is just a problem with the engine.
 
Last edited:
Wow I love the footage on this clip especially the highway car traffic.
youtube version

Truly amazing, the most spectacular Matrix demo so far. Is it the blurring doing it? Adds that cinematic vibe to it, perhaps its the camera work, no idea. He doesnt specify what system specs was being used. Most likely 6900XT/3090 of some sort. There i come with a puny 2080ti :p
 
Status
Not open for further replies.
Back
Top