Thursday, August 19th 2021

Intel's DLSS-rivaling AI-accelerated Supersampling Tech is Named XeSS, Doubles 4K Performance

Aug 19th, 2021 08:14 Discuss (46 Comments)

Intel plans to go full tilt with gaming graphics, with its newly announced Arc line of graphics processors designed for high-performance gaming. The top Arc "Alchemist" part meets all requirements for DirectX 12 Ultimate logo, including real-time raytracing. The company, during the technology's reveal, earlier this week, also said that it's working on an AI-accelerated supersampling technology. The company is calling it XeSS (Xe SuperSampling). It likely went with Xe in the name, as it possibly plans to extend the technology to even its Xe LP-based iGPUs and the entry-level Iris Xe MAX discrete GPU.

Intel claims that XeSS cuts down 4K frame render-times by half. By all accounts, 1440p appears to be the target use case of the top Arc "Alchemist" SKU. XeSS would make 4K possible (i.e., display resolution set at 4K, rendering at a lower resolution, with AI-accelerated supersampling restoring detail). The company revealed that XeSS will use a neural network-based temporal upscaling technology that incorporates motion vectors. In the rendering pipeline, XeSS sits before most post-processing stages, similar to AMD FSR.

While AMD's FSR technology is purely shader based, the Intel algorithm can either use XMX hardware units (new in Intel Xe HPG), or DP4a instructions (available on nearly all modern AMD and NVIDIA GPUs). XMX stands for Xe Matrix Extensions and is basically Intel's version of NVIDIA's Tensor Cores, to speed up matrix math, which is used in many AI-related tasks. The Intel XeSS SDK will be available this month, in open source, using XMX hardware, the DP4a version will be available "later this year".

Source: VideoCardz

Add your own comment

46 Comments on Intel's DLSS-rivaling AI-accelerated Supersampling Tech is Named XeSS, Doubles 4K Performance

#26

Vya Domus

juularFSR is not temporal, it's regular downscaling+upscaling with sharpening filters slapped together, that's why it's so bad, especially in motion.

You do realize this is literal nonsense, right ? It's the temporal solutions that look horrible in motion, because they blend information from past frames into the current frame. Spatial upscaling introduces zero temporal artifacts, because ... it's not temporal.

Can't believe I have to explain this.

TheUn4seenSo, now we have three proprietary upscaling technologies, with one being less proprietary but fairly primitive.

It's not "less proprietary, it's competently open.

TheUn4seen- If the deciding factor will be a financial incentive for the developers, nVidia will win,

This isn't about money, it's about time and effort. DLSS takes a lot of time to implement, I have some insight into the game development world and I can tell you that developers have to work for months to get DLSS even remotely close to working properly and it's always a side project because it's just not that important to them. FSR on the other hand is pretty much a couple of days worth of work. If it wasn't for the sponsorship campaigns of Nvidia, I bet most studios wouldn't even think about using it.

#27

londiste

Vya DomusYou do realize this is literal nonsense, right ? It's the temporal solutions that look horrible in motion, because they blend information from past frames into the current frame. Spatial upscaling introduces zero temporal artifacts, because ... it's not temporal.

At the same time lack of temporal component introduces potential for shimmering which FSR shows in practice. Also, past frames introduce additional data to use for frame reconstruction which helps with the end result.

#28

TheUn4seen

Vya DomusThis isn't about money, it's about time and effort. DLSS takes a lot of time to implement, I have some insight into the game development world and I can tell you that developers have to work for months to get DLSS even remotely close to working properly and it's always a side project because it's just not that important to them. FSR on the other hand is pretty much a couple of days worth of work. If it wasn't for the sponsorship campaigns of Nvidia, I bet most studios wouldn't even think about using it.

So, it's about money. Implementation time is money. DLSS takes time to be done right = it costs money. FSR is easier so it costs less, but not zero amount of money. Hence, if nVidia can offset DLSS cost with financial or marketing incentives they will win. Everything is always about money, especially in the entertainment industry where money is elevated to a level of deity.

#29

Vya Domus

TheUn4seenSR is easier so it costs less, but not zero amount of money.

No, it really is effectively nothing in terms of cost, unless you can point out to something that I am missing here.

TheUn4seenHence, if nVidia can offset DLSS cost with financial or marketing incentives they will win.

But this is what you don't understand, studios stand to gain nothing from this. They waste development time for something that can only be used on one platform by one particular sub group of that customer base. When they have to choose between allocating resources between getting DLSS working and getting the game working, you can be sure they'll choose the latter no matter what the marketing incentives are. That's why in the few games that use DLSS, support for it was usually added post launch.

#30

TheUn4seen

Vya DomusNo, it really is effectively nothing in terms of cost, unless you can point out to something that I am missing here.

But this is what you don't understand, studios stand to gain nothing from this. They waste development time for something that can only be used on one platform by one particular sub group of that customer base. When they have to choose between allocating resources between getting DLSS working and getting the game working, you can be sure they'll choose the latter no matter what the marketing incentives are. That's why in the few games that use DLSS, support for it was usually added post launch.

You seem to be somewhat out of touch with economy. Let's say, for the sake of argument, that FSR is incredibly, unreasonably easy, to the point it takes a day to implement into a complex product. One day of work times a hundred developers who either have to do it or at least read the documentation to maintain the code and you have a million dollars of costs. If you tell the management to either spend a million for a feature some people will use or spend five million for another version of that feature, but get some of it returned as cash or marketing, guess what the management will choose? Games aren't made by enthusiasts, they are made by accountants.

#31

HisDivineOrder

This is why we need a third player so badly. Let's just hope that their solution is an actual competitor to DLSS and is as good as they claim because we desperately need one to put pressure on Nvidia.

#32

Vya Domus

TheUn4seenbut get some of it returned as cash or marketing

How does the use of DLSS return as cash or marketing to the game studio ?

#33

auxy

R00kiethis infographic is confusing, if its an upscaler and it improves performance, shouldn't frametime be lower rather than higher with it enabled? :confused::confused::confused:

What are you confused about? The frametime goes down with XeSS enabled compared to native 4K rendering.

ZoneDymoI would like to know what the difference is between this and Nvidia's implementation because in base terms it all sounds the same

Particularly considering they poached the guy who invented DLSS at Nvidia recently.

defaultluserThey even made their own FXAA competitor you can force in the driver (it does almost nothing to edges, and it costs nearly as much performance.)

Because (unlike SMAA and FXAA) we will never be able to insert FSR, game devs are going to have to choose on or two techs to bother with!

1) Are you talking about AMD's MLAA? Because that's actually the predecessor to Nvidia's FXAA.
2) We already can inject FSR using ReShade. There's no reason anyone can't inject FSR, it's very easy and simple to do, since it's just a shader.

VerpalI am aware that NVIDIA support DP4a since Pascal, but is there any documentation on when AMD start supporting DP4a?

RDNA2 is the first AMD architecture with INT32 support.

#34

Trompochi

So it's called "SSeX" but backwards...... I'm so sorry, I had to... :roll::roll::roll::roll:

#35

watzupken

DrediXeSS will be implemented using open standards and should work across all platforms similarly to FSR.

In my opinion, "open standards" is likely used to try to sway developers from bespoke tech like DLSS. But being open here is one part of the equation, the second being how easy is it to implement as compared to DLSS and the more primitive FSR? It is the ease of integration without sacrificing too much details that will ultimately win developers over. Intel can sponsor some games here and there to get the developers to implement XeSS, but it is not a sustainable method. People can argue that FSR is a low bar because it is not "smart", but from the game developers' standpoint, you get performance back without a significant hit to performance, plus, its very easy for them to implement. Imagine when they had to optimise a game to make it run at a performance target on consoles, that part can be taken care of easily by FSR. So less work for them.

#36

HTC

chodaboy19We need some kind of baseline to be able to quantify this "double" the performance... hehe

Yup: for example, if it doubles the performance @ 4K from ... 1 FPS to 2 FPS ... it probably isn't as good as they think ...

#37

Vya Domus

HTCYup: for example, if it doubles the performance @ 4K from ... 1 FPS to 2 FPS ... it probably isn't as good as they think ...

It's not really impressive anyway, of course it doubles the performance if the internal resolution is much lower than native. It's not a miracle that this happens.

auxyRDNA2 is the first AMD architecture with INT32 support.

?

All GPUs have support for INT32, otherwise they wouldn't be able to work and RDNA1 had support for mixed INT32 execution.

#38

R00kie

auxyWhat are you confused about? The frametime goes down with XeSS enabled compared to native 4K rendering.

it would have been less confusing if they put framerate instead of frametime on the X axis, otherwise it just looks like it decreases performance

#39

Vayra86

TheUn4seenSo, it's about money. Implementation time is money. DLSS takes time to be done right = it costs money. FSR is easier so it costs less, but not zero amount of money. Hence, if nVidia can offset DLSS cost with financial or marketing incentives they will win. Everything is always about money, especially in the entertainment industry where money is elevated to a level of deity.

They wont win because they are only offsetting that cost in a small selection of games. That is the whole point. You cant keep up with all those games getting released but going forward in time, the need for DLSS to get decent perf will only get higher.

Look at PhysX. Nuff said. Where it was implemented, it worked admirably. But it was never everywhere and still at odds with other physics engines even if they worked less good - devs wont be happy to support just half the market with a different experience.

The end result is ALWAYS that its effectively just being used for marketing, appearing in high profile eye catchers. Look at DLSS support history for perfect proof of that. And RTX is more of the same. , but with an even higher dev investment.

Nothing is free and time to market is money too. The bill cant ever get paid in full, and believing it will is just setting yourself up for another deception.

#40

phanbuey

TheUn4seenSo, now we have three proprietary upscaling technologies, with one being less proprietary but fairly primitive. I don't think this kind of segmentation will last for long since developers don't want to limit their target audience and certainly don't want to spend money on implementing three separate technologies to achieve a single goal. So now:
- If the deciding factor will be a financial incentive for the developers, nVidia will win,
- If ease of implementation, FSR will win.
- If performance, Intel has the upper hand, if their first party benchmarking is to be believed. But they will have to really make it into a polished product since they have zero market share and brand recognition as far as dedicated GPUs go.

Personally I think if AMD can create FSR 2.0 with improved quality and performance AND make it hardware agnostic, it will be a clear winner for developers and consumers. As much as I admire the complexity and elegance of AI, the mass market works on the KISS principle.

FSR will win in terms of broad adoptability - DLSS 2.0/3.0 will win for ultimate quality -- I think it will be same outcome as the current adaptive sync situation. You will still have "GSYNC Ulitmate" which is pretty sweet but most displays will use the Freesync and it will be awesome - in this case FSR will be a must for most games, with nvidia studios ones pushing DLSS as well. But for sure FSR 2.0 is the more promising tech due to the openness.

#41

defaultluser

auxy1) Are you talking about AMD's MLAA? Because that's actually the predecessor to Nvidia's FXAA.

No, I'm talking about Intel's own knockoff of the tech:

software.intel.com/content/www/us/en/develop/blogs/conservative-morphological-anti-aliasing-cmaa.html

They say it has nothing to do with MLAA, but what who knows?

This is the only option you have in your control panel to insert full-scene post-process (so, if thi is the level of support you can expecrt on Xe, you'd better get used to bone-dry game override options, and maybe q handful of "only invented here" techs.

#42

medi01

Xe written in Cyrillic reads like "he" as if Raja is hinting us at something, hehehe.

btarunr, the Intel algorithm can either use XMX hardware units (new in Intel Xe HPG), or DP4a instructions (available on nearly all modern AMD and NVIDIA GPUs). XMX stands for Xe Matrix Extensions and is basically Intel's version of NVIDIA's Tensor Cores

If you were wondering whether DLSS not running on 10xx series was nVidia being into money milking to much: it was.

PunkenjoyI am the only one that dislike that we don't have standard AI/Temporal upscaling tech supported on all hardware having the required function instead of each vendor having their own receipe ?

Uh, I'm one of the few in "temporal: no thank you" camp, I guess.
And stop showing that DLSS 2 (the TAA derivative) stuff showing good results on scenes that barely move.

juularthat's why it's so bad, especially in motion.

Amazing post.
You should stop watching videos by "3080 is 2 times faster than 2080" and "8k gaming with 3090" (totally not shills) folks.

Vayra86Now that there are three, it gets a whole lot harder to defend that idea. Three inventors of a wheel where two are destined to fail is quite a bit more risk than the odd 50% adjusted for market(ing) share. Devs won't go about supporting three technologies either. They want them fed to them or they're not happening.

Uh, I suspect you have missed the two elephants in the room:

1) Effort matters. Is it low? Well, heck, devs can slap a bunch of various implementations in, no prob.
2) There is NO competition between "some crap that runs only on one manufacturer's HW" and stuff that runs on everything. Like at all. The former is in survival "do I even still make sense to exit" mode. The latter can be inferior, but will still do fine, all it needs is being better than standard upscaling solutions. (effort does still matter though, but with FSR we have seen "hard to distinguish from true 4k" and super low effort to implement too, so, hell, good luck with that)

#43

TheUn4seen

phanbueyFSR will win in terms of broad adoptability - DLSS 2.0/3.0 will win for ultimate quality -- I think it will be same outcome as the current adaptive sync situation. You will still have "GSYNC Ulitmate" which is pretty sweet but most displays will use the Freesync and it will be awesome - in this case FSR will be a must for most games, with nvidia studios ones pushing DLSS as well. But for sure FSR 2.0 is the more promising tech due to the openness.

It's true. One might become the "open for everyone" option with the better proprietary technology being the "high end" option for some. Makes me wonder how Intel can push their technology with zero market share and zero recognition. They do have a lot of money and leverage on the market though.
Finally something interesting in the PC world after years of "slightly better numbers and more blinking lights" from the same companies over and over again.

#44

Anymal

On desktop they will have 99 problems, in laptops they should prevail as already with CPUs.

#45

Anymal

TheUn4seenYou seem to be somewhat out of touch with economy. Let's say, for the sake of argument, that FSR is incredibly, unreasonably easy, to the point it takes a day to implement into a complex product. One day of work times a hundred developers who either have to do it or at least read the documentation to maintain the code and you have a million dollars of costs. If you tell the management to either spend a million for a feature some people will use or spend five million for another version of that feature, but get some of it returned as cash or marketing, guess what the management will choose? Games aren't made by enthusiasts, they are made by accountants.

1 vs. 5 million?

#46

Punkenjoy

medi01Uh, I'm one of the few in "temporal: no thank you" camp, I guess.
And stop showing that DLSS 2 (the TAA derivative) stuff showing good results on scenes that barely move.

I am on the same boat as you as for the temporal artefact. Although it's not all game that have that, the implementation of DLSS is very unequal and they seems to always use the best examples to make DLSS shine.

That also let me think that the AI portion in DLSS isn't that much AI. There is simply not enough time to do real AI processing there. And cut me with the pre-learned craps, they probably just figured what would be the best algorithm. They probably use the tensor core and do some calculation in int8 or other AI data format but that doesn't means much.

AI seems to be the new Nanotechnology of the past where everything you stamp with the word get trending and get financing. Don't get me wrong, there was good stuff coming out of this, but a lot of things were just not that at all.

To me, a real deep learning up sampling would be able to eliminate completely those ghosting artefact. Also, it could also leverage less the temporal data and more the AI portion to resolve details. But this take a lot of calculation power and you probably can't run that in realtime. So what we have is a glorified proprietary temporal upsampling.

But also, don't get me wrong, Shimmering that FSR can bring are as annoying if not more than temporal upsampling.

Add your own comment

Intel's DLSS-rivaling AI-accelerated Supersampling Tech is Named XeSS, Doubles 4K Performance

46 Comments on Intel's DLSS-rivaling AI-accelerated Supersampling Tech is Named XeSS, Doubles 4K Performance

Latest GPU Drivers

New Forum Posts

Popular Reviews

Controversial News Posts

Intel's DLSS-rivaling AI-accelerated Supersampling Tech is Named XeSS, Doubles 4K Performance

Related News

46 Comments on Intel's DLSS-rivaling AI-accelerated Supersampling Tech is Named XeSS, Doubles 4K Performance

Latest GPU Drivers

New Forum Posts

Popular Reviews

Controversial News Posts