Seems similar limitation to DLSS. Taau and DLSS can be applied globally as long as the game supports taa in the first place.
Digital Foundry Article Technical Discussion [2022]
- Thread starter BRiT
- Start date
- Status
Seems similar limitation to DLSS. Taau and DLSS can be applied globally as long as the game supports taa in the first place.
Intel XeSS upscaling requires DPA4 / int8, so for at least that form of upscaling there seems to be an advantage to that feature.
MLAA such as DLSS and XeSS use temporal elements, and it [edit] should [/edit] be possible to make this work after an engine's own TAAU has taken place, or incorporate them into it.
MLAA such as DLSS and XeSS use temporal elements, and it [edit] should [/edit] be possible to make this work after an engine's own TAAU has taken place, or incorporate them into it.
Because they have the dedicated silicon to brute force it. There are now very good reconstruction techs on console so DLSS is not needed. What is needed on PS5 is VRR.
Sure, VRR is needed.
But something like DLSS would be good, too. E.g. the first few forms of useable DLSS (the first version was just awful) did just use GPU-compute. So it should be possible to port DLSS over to consoles.
The extra support of lower precision (double rate) modes might help to reduce this compute-task. E.g. the PS5 silicon will at least support double rate FP16, so this shouldn't be to compute intensive after all.
Indeed! It's an easy fix for games with sloppy performance. Enable vrr, and bam, no more issues. At least on my Xbox.
The 40fps mode is a good compromise tho.
snc
Veteran
Good work, as df said is like low vs ultra settings, on pro still looks really good, on basic dynamic 1080p with some vrs or cb probably not that good on big tv
I don't think it is scaling from "low" to high. This game is even on PS4 not low detailed. I've seen much lower detailed current games with higher requirements for the GPU.
It does indeed look and feel like a game made for the PS4 on the PS4 and made for the PS5 on the PS5. But the bigger problem in those videos is like always the youtube compression. Even at 4k much detail is just lost due to the compression. This means that e.g. games like horizon might look very similar on youtube. This might get a big problem for the marketing as they can't really show the graphical quality of the game. People might just not see why they might "need" a next gen console because of that.
Nah. This didn't stop people from purchasing cross-generational title Spider-Man: Miles Morales on PS5 in droves (third most sold 1st-party PlayStation title), nor the PS5 system itself. The same applies to Forza Horizon 5 across Xbox platforms. As such, the PS5/XBSX are constantly sold out, and the aforementioned cross-generational games on these platforms are selling well. So, I see no 'worries' or 'concerns' about gamers not purchasing Horizon Forbidden West based on poor YouTube video quality (such as not picking up high frequency details when comparing cross-generational titles).
Last edited:
Its one of the DF commenting its akin to low/ultra settings on pc games. Its just that GG has done a good job scaling the settings between the platforms, some pc games scale better then others do. This is great news for a later pc port.
Using what we know of XeSS.
The general implication uses Dp4a INT 8 I believe.
So not taking into account things like data localisation, doesn't dp4a also have additional instructions.
So PS5 could easily be twice or maybe even four times slower. I'm just guessing 4, so let's just go with 2.5 - 3, someone with more knowledge about dp4a and inference will have to chime in.
The point is that just being able to convert it to RPM 16 it may still use a lot more compute than you would think.
I'm not saying PS5 can't use XeSS, just that right now I would put it bellow XSS.
Reason being if it works like DLSS it's based on output resolution.
So PS5 2160p, XSS 1080/1440p using dp4a.
And XSS may not have enough headroom to do it at 60fps?
Can't wait to start to get decent benchmarks from XeSS etc.
Edit:
Found this blurb from an intel pdf
Looks like marketing material but gives some indication that will likely be more than twiice as fast.
Last edited:
The primary implementation uses XMX and the DP4A is the general fallback. I haven't found any actual timings either, just the same general representative image showing relative quality and timing.
From an interview @ WCCFTech [ https://wccftech.com/intel-xess-interview-karthik-vaidyanathan/ ]:
Usman: And I actually heard yesterday that you guys are planning to support XeSS on older models as well. So can you elaborate on how that would work. To my understanding, the reason FSR is not machine learning based is because AMD wanted to cast a wider net on older models as well and we already know DLSS requires some form of inference capability in the hardware. So, how, how would you plan on rolling this out to older models, if you are planning for those?
Karthik: Yes, we require inference capabilities but matrix acceleration is not the only form of inference capability that is available on GPUs. If you go all the way back to- I think Skylake- we had dot product acceleration, which is DP 4.4 – there’s various names for it. Nvidia has had this I think, since Turing and AMD has this now on RDNA2. So even without Matrix acceleration you can go quite far. It might not be as fast as matrix acceleration, but certainly meets the objective. And as I said, the objective is to maintain the fidelity of your render and achieve smooth frame rates. So, when it comes to older models, on older internal GPUs, we've had dot product acceleration (DP4a) for a while now.
Microsoft has enabled this through Shader Model 6.4 and above and on all these platforms XeSS will work.
Karthik: Yes, we require inference capabilities but matrix acceleration is not the only form of inference capability that is available on GPUs. If you go all the way back to- I think Skylake- we had dot product acceleration, which is DP 4.4 – there’s various names for it. Nvidia has had this I think, since Turing and AMD has this now on RDNA2. So even without Matrix acceleration you can go quite far. It might not be as fast as matrix acceleration, but certainly meets the objective. And as I said, the objective is to maintain the fidelity of your render and achieve smooth frame rates. So, when it comes to older models, on older internal GPUs, we've had dot product acceleration (DP4a) for a while now.
Microsoft has enabled this through Shader Model 6.4 and above and on all these platforms XeSS will work.
There's some specific benefits for XSS.
The lower the native resolution the harder it is to get a good output one.
Even though TAAU with UE5 being a good example of a really good implementation, going from <=720p to 1080p probably won't look native.
ML-U I expect to generally give a better final image.
Other inherent benefits that are forgotten, are things like the higher relative fillrate at lower resolution that it may struggle with otherwise
When I said general, I meant non Intel ARC specific implementation.
Dp4a is more than just simply INT8 support.
As in just using INT16 it more than likely will be more than twice as slow
I'm curious how much difference you can generally expect from dp4a compared to RPM for inference.
And XeSS dp4a implementation compared to XMX, bit more detailed than their bar graph, although useful
XeSS will work by default on XS consoles for any game that uses it.
Be interesting to see it used, although ARC GPUs aren't anywhere to be seen at the moment.
That depends what the input data is and where in the pipeline the upscaling occurs. If you're doing the upscale way down close to post processing, yes. If you can get the upscaler to do the heavy lifting on the compute intensive items (like lighting/ GI etc) then perhaps you can still come out with a relatively good image. Not saying it's easy, but we're only a few years into realtime AI upscaling, and I've seen some crazy stuff for non-realtime.
Last edited:
Not if it's too slow on PS5 it won't be better than nothing.
But then I think that TAAU will be more than good enough.
Even if it never looked 100% as good, and had to also have settings tweeked lower.
90% of gamers won't care, only small section of the Internet.
We here generally care from a technical discussion.
I just don't think the difference will be big enough to matter to most people.
Hence other things are more important in that regards.
- Status
Similar threads
