• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

The Medium: DLSS vs. FSR Comparison

Joined
Nov 11, 2016
Messages
3,045 (1.13/day)
System Name The de-ploughminator Mk-II
Processor i7 13700KF
Motherboard MSI Z790 Carbon
Cooling ID-Cooling SE-226-XT + Phanteks T30
Memory 2x16GB G.Skill DDR5 7200Cas34
Video Card(s) Asus RTX4090 TUF
Storage Kingston KC3000 2TB NVME
Display(s) LG OLED CX48"
Case Corsair 5000D Air
Power Supply Corsair HX850
Mouse Razor Viper Ultimate
Keyboard Corsair K75
Software win11
DLSS has access to the raw imagery earlier than FSR, so it can 'reconstruct' data that would have been thrown out by the time FSR sees it

Like uhhh.... seeing the data for a fence *before* a motion blur effect is added, instead of after: more to work with, so a slightly better image

Actually DLSS uses something called camera jitter, which Intel copied with their XeSS, to reconstruct pixel-wide objects that even 4K Native resolution would flicker

U3epvuiuNBGP7k58Ygdjkf-970-80.jpg


Here is an example in CP2077
 
Joined
Sep 17, 2014
Messages
20,780 (5.97/day)
Location
The Washing Machine
Processor i7 8700k 4.6Ghz @ 1.24V
Motherboard AsRock Fatal1ty K6 Z370
Cooling beQuiet! Dark Rock Pro 3
Memory 16GB Corsair Vengeance LPX 3200/C16
Video Card(s) ASRock RX7900XT Phantom Gaming
Storage Samsung 850 EVO 1TB + Samsung 830 256GB + Crucial BX100 250GB + Toshiba 1TB HDD
Display(s) Gigabyte G34QWC (3440x1440)
Case Fractal Design Define R5
Audio Device(s) Harman Kardon AVR137 + 2.1
Power Supply EVGA Supernova G2 750W
Mouse XTRFY M42
Keyboard Lenovo Thinkpad Trackpoint II
Software W10 x64
I suppose if you consider native to be the goal, when a reconstructed image has more detail people tend to balk at the idea, how is it possible?. But why should the native resolution image be where it ends? is that the gold standard? clearly we can push beyond. iirc DLSS is attempting to achieve (by way of training and comparison to ultra high res 16k images) a super sampled look and these images plus other great implementations demonstrate that. I'm not saying it's without flaw, but it is being demonstrated time and time again that beyond native res details are achievable.

You're still limited to a single pixel 'output' on the downsampled image which means the pixel can only have one color. Unless Nvidia magically manages to split them up, but I haven't seen Huang shake that out of his leather jacket yet. In motion, what DLSS can achieve is more accuracy in picking its colors. But its still one color per pixel. At lower resolutions, due to the absence of enough pixels, this will make highly detailed areas get that 'blur' or somewhat faded effect, even if the detail is still all there while the native image might retain more natural sharpness (pixel colors are further 'apart', adding contrast to edges). At higher res, the pixel density is so high you won't notice in both cases, but zoomed in the DLSS image does appear more refined. I think the first 4K comparison highlights that nicely. Less contrasted, but more refined drawing. I can see why you'd say its 'better' - but there isn't more detail. I think the bigger win is the performance alongside it. Its a trade off you'll want to make, for sure.

I mean this one; first pic 4K native vs 4K quality DLSS, max zoom. The 4K quality DLSS is certainly more refined. But same number of pixels, and you can see its more smoothed out to their nearest neighbours, while the 4K image is more 'grainy'. However they both display the exact same details up to and including the shades of the net wires.

1631695400288.png


Note that if you apply AA to a native image, you can actually reduce the detail available on it, but you lose aliasing. So if you compare DLSS to a native image with AA, then sure, it can 'seem' more detailed. But not more detailed than the native image - the pixels are limiting you there. You're just seeing different colors, and early implementations of DLSS (I recall Final Fantasy) truly had quite visible color changes.
 
Last edited:
Joined
Feb 3, 2017
Messages
3,475 (1.33/day)
Processor R5 5600X
Motherboard ASUS ROG STRIX B550-I GAMING
Cooling Alpenföhn Black Ridge
Memory 2*16GB DDR4-2666 VLP @3800
Video Card(s) EVGA Geforce RTX 3080 XC3
Storage 1TB Samsung 970 Pro, 2TB Intel 660p
Display(s) ASUS PG279Q, Eizo EV2736W
Case Dan Cases A4-SFX
Power Supply Corsair SF600
Mouse Corsair Ironclaw Wireless RGB
Keyboard Corsair K60
VR HMD HTC Vive
DLSS has access to the raw imagery earlier than FSR, so it can 'reconstruct' data that would have been thrown out by the time FSR sees it
Like uhhh.... seeing the data for a fence *before* a motion blur effect is added, instead of after: more to work with, so a slightly better image
AMD's instructions for FSR tell developers to implement it before postprocessing effects and UI.
DLSS lives in the same general stage of pipeline.

Note that if you apply AA to a native image, you can actually reduce the detail available on it, but you lose aliasing. So if you compare DLSS to a native image with AA, then sure, it can 'seem' more detailed. But not more detailed than the native image - the pixels are limiting you there. You're just seeing different colors, and early implementations of DLSS (I recall Final Fantasy) truly had quite visible color changes.
This. DLSS implementation requirements are similar to TAA and it is generally compared to native image with TAA enabled while DLSS does not use TAA but gets own (apparently inherent) antialiasing.
 
Joined
Nov 11, 2016
Messages
3,045 (1.13/day)
System Name The de-ploughminator Mk-II
Processor i7 13700KF
Motherboard MSI Z790 Carbon
Cooling ID-Cooling SE-226-XT + Phanteks T30
Memory 2x16GB G.Skill DDR5 7200Cas34
Video Card(s) Asus RTX4090 TUF
Storage Kingston KC3000 2TB NVME
Display(s) LG OLED CX48"
Case Corsair 5000D Air
Power Supply Corsair HX850
Mouse Razor Viper Ultimate
Keyboard Corsair K75
Software win11
This. DLSS implementation requirements are similar to TAA and it is generally compared to native image with TAA enabled while DLSS does not use TAA but gets own (apparently inherent) antialiasing.

Well, the best general purpose AA is Super Sampling after all, which DLSS is trying to replicate (hence the name Deep Learning Super Sampling).

The Medium 4K Native vs 6K Native (DSR)
 
Joined
Aug 7, 2019
Messages
334 (0.20/day)
I suppose if you consider native to be the goal, when a reconstructed image has more detail people tend to balk at the idea, how is it possible?. But why should the native resolution image be where it ends? is that the gold standard? clearly we can push beyond. iirc DLSS is attempting to achieve (by way of training and comparison to ultra high res 16k images) a super sampled look and these images plus other great implementations demonstrate that. I'm not saying it's without flaw, but it is being demonstrated time and time again that beyond native res details are achievable.

Where are those demonstrations? You can't achieve more detail than native unless the developer did a terrible job in first place adding terrible filters, shaders, etc. And those are usually added to achieve a higher FPS in first place (or utter incompetence).

It is literally impossible to achieve more detail than native over an unfiltered output. Then there is also the problem of altering what the developer intended in first place.
 
Joined
Nov 11, 2016
Messages
3,045 (1.13/day)
System Name The de-ploughminator Mk-II
Processor i7 13700KF
Motherboard MSI Z790 Carbon
Cooling ID-Cooling SE-226-XT + Phanteks T30
Memory 2x16GB G.Skill DDR5 7200Cas34
Video Card(s) Asus RTX4090 TUF
Storage Kingston KC3000 2TB NVME
Display(s) LG OLED CX48"
Case Corsair 5000D Air
Power Supply Corsair HX850
Mouse Razor Viper Ultimate
Keyboard Corsair K75
Software win11
Trying 6K using DSR and the results are pretty interesting

comparison.jpg

The zipper on her jacket looks higher detailed with DLSS Q than even the Native :twitch:. Seems like DLSS works better the higher resolution.
 

wolf

Performance Enthusiast
Joined
May 7, 2007
Messages
7,726 (1.25/day)
System Name MightyX
Processor Ryzen 5800X3D
Motherboard Gigabyte X570 I Aorus Pro WiFi
Cooling Scythe Fuma 2
Memory 32GB DDR4 3600 CL16
Video Card(s) Asus TUF RTX3080 Deshrouded
Storage WD Black SN850X 2TB
Display(s) LG 42C2 4K OLED
Case Coolermaster NR200P
Audio Device(s) LG SN5Y / Focal Clear
Power Supply Corsair SF750 Platinum
Mouse Corsair Dark Core RBG Pro SE
Keyboard Glorious GMMK Compact w/pudding
VR HMD Meta Quest 3
Software case populated with Artic P12's
Benchmark Scores 4k120 OLED Gsync bliss
You're still limited to a single pixel 'output' on the downsampled image which means the pixel can only have one color. Unless Nvidia magically manages to split them up
I mean this one; first pic 4K native vs 4K quality DLSS, max zoom. The 4K quality DLSS is certainly more refined. But same number of pixels, and you can see its more smoothed out to their nearest neighbours, while the 4K image is more 'grainy'. However they both display the exact same details up to and including the shades of the net wires.
I'm not claiming Nvidia is magically adding pixels, it's quite simply as @nguyen has put it, a supersampling effect, which is a tried and true method of improving image quality, mostly in antialiasing but also percieved quality and detial. If you took a 1080p render on a 1080p screen, then rendered the same scene in 2160p, downsample to 1080p, I can tell you which image is going to appear to be more sharp and detailed and certainly cleaner as an antialiasing method, yet you're only viewing the same amount of pixels, thus beyond native res rendering can look "better" when downsampled to native, in comparison to native, so native only doesn't appear to be the ultimate goal.

I'm not sure we're looking at the same image? the vertical netting wires are much less defined and visible on native, more so zoomed in, but the effect is definitely there when viewed ~pixel for pixel too. Yes, it's the same amount of pixels, I can't deny it, and sure its 'only' colour choice, but it results in a supersampled effect which I'd say is a massive positive, and it seems that you agree calling it more refined. Is our only disagreement about the use of the term 'detail'?
You can't achieve more detail than native
What do you make of supersampling and the example I have above (1080p or 2160p downsampled to 1080p), those images almost universally appear more detailed and clean, even at the same output pixel count.
 
Last edited:
Joined
Sep 17, 2014
Messages
20,780 (5.97/day)
Location
The Washing Machine
Processor i7 8700k 4.6Ghz @ 1.24V
Motherboard AsRock Fatal1ty K6 Z370
Cooling beQuiet! Dark Rock Pro 3
Memory 16GB Corsair Vengeance LPX 3200/C16
Video Card(s) ASRock RX7900XT Phantom Gaming
Storage Samsung 850 EVO 1TB + Samsung 830 256GB + Crucial BX100 250GB + Toshiba 1TB HDD
Display(s) Gigabyte G34QWC (3440x1440)
Case Fractal Design Define R5
Audio Device(s) Harman Kardon AVR137 + 2.1
Power Supply EVGA Supernova G2 750W
Mouse XTRFY M42
Keyboard Lenovo Thinkpad Trackpoint II
Software W10 x64
I'm not claiming Nvidia is magically adding pixels, it's quite simply as @nguyen has put it, a supersampling effect, which is a tried and true method of improving image quality, mostly in antialiasing but also percieved quality and detial. If you took a 1080p render on a 1080p screen, then rendered the same scene in 2160p, downsample to 1080p, I can tell you which image is going to appear to be more sharp and detailed and certainly cleaner as an antialiasing method, yet you're only viewing the same amount of pixels, thus beyond native res rendering can look "better" when downsampled to native, in comparison to native, so native only doesn't appear to be the ultimate goal.

I'm not sure we're looking at the same image? the vertical netting wires are much less defined and visible on native, more so zoomed in, but the effect is definitely there when viewed ~pixel for pixel too. Yes, it's the same amount of pixels, I can't deny it, and sure its 'only' colour choice, but it results in a supersampled effect which I'd say is a massive positive, and it seems that you agree calling it more refined. Is our only disagreement about the use of the term 'detail'?

What do you make of supersampling and the example I have above (1080p or 2160p downsampled to 1080p), those images almost universally appear more detailed and clean, even at the same output pixel count.

It depends on the scene. The changes in color information arent always an improvement in fidelity. Sometimes you need the added contrast to discern shapes (more contrast on edges = perceived sharpness, without oversharpen effect, which is what native has) Those are the instances where the image may look a bit more washed out.

I never disagreed on the image with DLSS being improved compared to FSR- but after zooming in, zoom out and place it next to the original at same res. Is there any noticeable effect? I beg to differ - pixel density already fixes things for our mind.

The bigger win in both cases (FSR/DLSS) is the performance uplift, I think. If the native res is high enough.

I vividly remember viewing supersampled content downsampled to 1080p. DSR... and it was either just as aliased or you added smoothing to create a washed out effect. It was different, but more detailed it was not. Again... you need the higher pixel density on screen at which point diminishing returns also kick in at a normal viewing distance. For much the same reasons, we also disable any AA at high res. Simply doesnt add anything except performance hit.
 
Last edited:
Joined
Aug 7, 2019
Messages
334 (0.20/day)
IWhat do you make of supersampling and the example I have above (1080p or 2160p downsampled to 1080p), those images almost universally appear more detailed and clean, even at the same output pixel count.
Detailed? The leaves in the DLSS image lost a lot of definition in the nerves for example. There is aliasing in the floor. Her hair looks dreadful.

Clean? Do you realize that cleanliness is the product of removing stuff, right?
 

wolf

Performance Enthusiast
Joined
May 7, 2007
Messages
7,726 (1.25/day)
System Name MightyX
Processor Ryzen 5800X3D
Motherboard Gigabyte X570 I Aorus Pro WiFi
Cooling Scythe Fuma 2
Memory 32GB DDR4 3600 CL16
Video Card(s) Asus TUF RTX3080 Deshrouded
Storage WD Black SN850X 2TB
Display(s) LG 42C2 4K OLED
Case Coolermaster NR200P
Audio Device(s) LG SN5Y / Focal Clear
Power Supply Corsair SF750 Platinum
Mouse Corsair Dark Core RBG Pro SE
Keyboard Glorious GMMK Compact w/pudding
VR HMD Meta Quest 3
Software case populated with Artic P12's
Benchmark Scores 4k120 OLED Gsync bliss
It depends on the scene. The changes in color information arent always an improvement in fidelity. Sometimes you need the added contrast to discern shapes (more contrast on edges = perceived sharpness, without oversharpen effect, which is what native has) Those are the instances where the image may look a bit more washed out.
If we're talking DLSS I'd agree, it absolutely won't always be an improvement, if we're talking a 2x each axis (4x res) supersample and downscale with no smoothing, I think I've yet to see that not be an improvement overall. Sure there may be some parts of an image, depending on the game and scene that supersampling just can't really do much with, but overall it's sharper, appears more detailed and is very effectiveantialiased. If I had an unlimited rendering budget, I'd be running all my games at 4x resolution downscaled to native I think, unless for whatever reason there were unforeseen bugs or issues.
I never disagreed on the image with DLSS being improved compared to FSR- but after zooming in, zoom out and place it next to the original at same res. Is there any noticeable effect? I beg to differ - pixel density already fixes things for our mind.
Yeah this is where I disagree, viewed at a normal to me viewing distance, the effect is noticeable to me, perhaps it's just me, where my eyes are drawn to, how good/bad my vision is, what I perceive, my monitor etc
The bigger win in both cases (FSR/DLSS) is the performance uplift, I think. If the native res is high enough.
Can't disagree with that. If the entire image is even generally comparable but comes with more frames, it's a win.
I vividly remember viewing supersampled content downsampled to 1080p. DSR... and it was either just as aliased or you added smoothing to create a washed out effect. It was different, but more detailed it was not. Again... you need the higher pixel density on screen at which point diminishing returns also kick in at a normal viewing distance. For much the same reasons, we also disable any AA at high res. Simply doesnt add anything except performance hit.
Smoothing iirc is meant for values less than 'perfect' scaling options like doubling each axis, if you want to do a quad supersample smoothing should be off entirely. So like I've said for me I've yet to see it look worse (supersampling, not DLSS), and 100% the performance hit is real and generally not worth it, but if the rendering budget is there its a great way to make newer (or just crazy powerful) hardware breathe life into older titles, and it's what DLSS is trying to achieve, just with lower resolution inputs + motion vectors + historical frames + training against 16k images, instead of rendering higher sample counts.
Detailed? The leaves in the DLSS image lost a lot of definition in the nerves for example. There is aliasing in the floor. Her hair looks dreadful.
Yeah, some parts are better, I didn't say all parts, but this is what DLSS is trying to do, it's trying to supersample. And given generally new titles with DLSS tend to look ~comparable to native, with some stronger and some weaker aspects, given how well it antialiases the image, AND this comes with a performance boost, yeah it's pretty decent and definitely is capable of producing parts of the image better than native rendering, especially on titles where inbuilt AA cannot be disabled.
 
Joined
Aug 7, 2019
Messages
334 (0.20/day)
Yeah, some parts are better, I didn't say all parts, but this is what DLSS is trying to do, it's trying to supersample. And given generally new titles with DLSS tend to look ~comparable to native, with some stronger and some weaker aspects, given how well it antialiases the image, AND this comes with a performance boost, yeah it's pretty decent and definitely is capable of producing parts of the image better than native rendering, especially on titles where inbuilt AA cannot be disabled.
No, they are not. Some parts are worse and some parts are different. If you think they are better or have more detail it is just your opinion.

This isn't IRL photography. When the game fetches a texture it isn't some fuzzy process with physical limitations. It is impossible for DLSS to have more detail than a image rendered at native resolution because all the information is already there.
 

wolf

Performance Enthusiast
Joined
May 7, 2007
Messages
7,726 (1.25/day)
System Name MightyX
Processor Ryzen 5800X3D
Motherboard Gigabyte X570 I Aorus Pro WiFi
Cooling Scythe Fuma 2
Memory 32GB DDR4 3600 CL16
Video Card(s) Asus TUF RTX3080 Deshrouded
Storage WD Black SN850X 2TB
Display(s) LG 42C2 4K OLED
Case Coolermaster NR200P
Audio Device(s) LG SN5Y / Focal Clear
Power Supply Corsair SF750 Platinum
Mouse Corsair Dark Core RBG Pro SE
Keyboard Glorious GMMK Compact w/pudding
VR HMD Meta Quest 3
Software case populated with Artic P12's
Benchmark Scores 4k120 OLED Gsync bliss
If you think they are better or have more detail it is just your opinion.
Well, If if you think it's not, that's just your opinion, mine is shared by this reviewer and many others. I'm not saying it applies unilaterally to every game or even every part of the image, but, better than native is possible. Believe what you like.

TPU, maxus24
DLSS, on the other hand, does a great job reconstructing small objects; it even results in more detail and better image quality than the native
Digital Foundry, Alex Battaglia
Nvidia's DLSS - which delivers image quality better than native resolution rendering.
Venturebeat, Geoff Grubb
But DLSS, meanwhile, actually has even more detail than the native 4K.
DSO, Gaming
As we can clearly see, the DLSS image looks sharper with less aliasing. You can clearly notice these image improvements in the screenshots that have the Helicarrier. The distant objects look more refined with fewer jaggies in the DLSS screenshots. Native 4K comes in second place
Hardware Unboxed, Tim
the final results are as good as or in some circumstances better than the native 4K image.

And of the total ~80k respondents here, 19% is ~15k people

Screenshot_20210917-082542_YouTube.jpg
 
Joined
Aug 7, 2019
Messages
334 (0.20/day)
@wolf
Where are those demonstrations? You can't achieve more detail than native unless the developer did a terrible job in first place adding terrible filters, shaders, etc. And those are usually added to achieve a higher FPS in first place (or utter incompetence).

It is literally impossible to achieve more detail than native over an unfiltered output. Then there is also the problem of altering what the developer intended in first place.

If TAA is giving you trouble because it is shit maybe you should try again without it and I guess it is the problem in those reviews 100% of the time.

DLSS isn't magic and can't achieve higher detail than native because it is physically impossible.
 
Last edited by a moderator:
D

Deleted member 190774

Guest
@wolf


If TAA is giving you trouble because it is shit maybe you should try again without it and I guess it is the problem in those reviews 100% of the time.

DLSS isn't magic and can't achieve higher detail than native because it is physically impossible. Stop being a moron.
I'm not a fan of the ongoing debate between DLSS and FSR, but perhaps you should watch this video:


I do honestly think that DLSS isn't as technologically amazing as Nvidia would have us believe, and the onbaord hardware is probably completely overkill for what it does.
 

wolf

Performance Enthusiast
Joined
May 7, 2007
Messages
7,726 (1.25/day)
System Name MightyX
Processor Ryzen 5800X3D
Motherboard Gigabyte X570 I Aorus Pro WiFi
Cooling Scythe Fuma 2
Memory 32GB DDR4 3600 CL16
Video Card(s) Asus TUF RTX3080 Deshrouded
Storage WD Black SN850X 2TB
Display(s) LG 42C2 4K OLED
Case Coolermaster NR200P
Audio Device(s) LG SN5Y / Focal Clear
Power Supply Corsair SF750 Platinum
Mouse Corsair Dark Core RBG Pro SE
Keyboard Glorious GMMK Compact w/pudding
VR HMD Meta Quest 3
Software case populated with Artic P12's
Benchmark Scores 4k120 OLED Gsync bliss
the problem in those reviews 100% of the time.
Right, everyone else is wrong, got it.
DLSS isn't magic and can't achieve higher detail than native because it is physically impossible
The premise is very similar to a traditional supersample, but instead of rendering at a higher sample count, jitter, the history buffer and 16k image training are employed.

Since multiple subpixels per pixel can be sampled, polygonal details smaller than one pixel that might have been missed can be captured and made a part of the final rendered image if enough samples are taken.

They can't add pixels to the final image, but that image at the same pixel count can be better.
 
Joined
Mar 21, 2016
Messages
2,195 (0.75/day)
It depends on the scene. The changes in color information arent always an improvement in fidelity. Sometimes you need the added contrast to discern shapes (more contrast on edges = perceived sharpness, without oversharpen effect, which is what native has) Those are the instances where the image may look a bit more washed out.

I never disagreed on the image with DLSS being improved compared to FSR- but after zooming in, zoom out and place it next to the original at same res. Is there any noticeable effect? I beg to differ - pixel density already fixes things for our mind.

The bigger win in both cases (FSR/DLSS) is the performance uplift, I think. If the native res is high enough.

I vividly remember viewing supersampled content downsampled to 1080p. DSR... and it was either just as aliased or you added smoothing to create a washed out effect. It was different, but more detailed it was not. Again... you need the higher pixel density on screen at which point diminishing returns also kick in at a normal viewing distance. For much the same reasons, we also disable any AA at high res. Simply doesnt add anything except performance hit.
If resolution is high enough post process is fine even zoomed in or up to a more tolerable point if it's too low it's easy to identify and point out differences because it didn't have enough sample feedback data to draw from in the first place with enough inference methodologies to reconstruct the missing information. Also the higher the resolution the higher the post process overhead and the lower the resolution the lower the post process overhead. You basically have a scale with post process from low overhead low feedback reconstruction to high overhead high feedback reconstruction with post process in relation to resolution.

Where post process will improve is masking the reconstruction techniques themself while maintaining lower overhead or variably faking the data between multiple intermittent regions and/or frames convincingly enough to fool the eye like a slight of hand magic trick.
 
Joined
Nov 11, 2016
Messages
3,045 (1.13/day)
System Name The de-ploughminator Mk-II
Processor i7 13700KF
Motherboard MSI Z790 Carbon
Cooling ID-Cooling SE-226-XT + Phanteks T30
Memory 2x16GB G.Skill DDR5 7200Cas34
Video Card(s) Asus RTX4090 TUF
Storage Kingston KC3000 2TB NVME
Display(s) LG OLED CX48"
Case Corsair 5000D Air
Power Supply Corsair HX850
Mouse Razor Viper Ultimate
Keyboard Corsair K75
Software win11
I'm not a fan of the ongoing debate between DLSS and FSR, but perhaps you should watch this video:


I do honestly think that DLSS isn't as technologically amazing as Nvidia would have us believe, and the onbaord hardware is probably completely overkill for what it does.

Well Intel engineers certainly didn't think so
tensor.png


Intel thought so highly of DLSS that they copied the whole concept of DLSS, down to the 64 samples per pixel super-sampling part, but make it so that XeSS run natively only on Intel hardware.
"Imitation is the sincerest form of flattery"
 

wolf

Performance Enthusiast
Joined
May 7, 2007
Messages
7,726 (1.25/day)
System Name MightyX
Processor Ryzen 5800X3D
Motherboard Gigabyte X570 I Aorus Pro WiFi
Cooling Scythe Fuma 2
Memory 32GB DDR4 3600 CL16
Video Card(s) Asus TUF RTX3080 Deshrouded
Storage WD Black SN850X 2TB
Display(s) LG 42C2 4K OLED
Case Coolermaster NR200P
Audio Device(s) LG SN5Y / Focal Clear
Power Supply Corsair SF750 Platinum
Mouse Corsair Dark Core RBG Pro SE
Keyboard Glorious GMMK Compact w/pudding
VR HMD Meta Quest 3
Software case populated with Artic P12's
Benchmark Scores 4k120 OLED Gsync bliss
Well Intel engineers certainly didn't think so
It's forward-looking architecture, these things are clearly not going away, they will only become more and more prevalent.
 
D

Deleted member 190774

Guest
@nguyen Yes, we can all read, but do YOU understand what it all means?

Intel also wrote in that article:
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.

There's no point in discussing any of this really, unless you have sufficient experience in utilising this hardware with your own code; otherwise it's just numbers on a page.
 
Joined
Feb 3, 2017
Messages
3,475 (1.33/day)
Processor R5 5600X
Motherboard ASUS ROG STRIX B550-I GAMING
Cooling Alpenföhn Black Ridge
Memory 2*16GB DDR4-2666 VLP @3800
Video Card(s) EVGA Geforce RTX 3080 XC3
Storage 1TB Samsung 970 Pro, 2TB Intel 660p
Display(s) ASUS PG279Q, Eizo EV2736W
Case Dan Cases A4-SFX
Power Supply Corsair SF600
Mouse Corsair Ironclaw Wireless RGB
Keyboard Corsair K60
VR HMD HTC Vive
I'm not a fan of the ongoing debate between DLSS and FSR, but perhaps you should watch this video:

...

I do honestly think that DLSS isn't as technologically amazing as Nvidia would have us believe, and the onbaord hardware is probably completely overkill for what it does.
How long does it take for TecoGAN to enhance these images and how does it compare to DLSS's 1.5ms (at 2160p)?
Especially with video and the temporally coherent in the name, comparing to FSR would not be fair.

@nguyen Yes, we can all read, but do YOU understand what it all means?

Intel also wrote in that article:
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.

There's no point in discussing any of this really, unless you have sufficient experience in utilising this hardware with your own code; otherwise it's just numbers on a page.
The question here is not about capability, it is about performance.
 
Joined
Mar 21, 2016
Messages
2,195 (0.75/day)
How strong will Intel ARC GPU's will be for rasterization is what I'm wondering looking at those specs on it looks stout on the TMUs and ROPs plus tons of tensor cores for inference image scaling. Raja may actually have pulled a rabbit out of the hat despite all the harsh criticism.
 
Joined
Nov 11, 2016
Messages
3,045 (1.13/day)
System Name The de-ploughminator Mk-II
Processor i7 13700KF
Motherboard MSI Z790 Carbon
Cooling ID-Cooling SE-226-XT + Phanteks T30
Memory 2x16GB G.Skill DDR5 7200Cas34
Video Card(s) Asus RTX4090 TUF
Storage Kingston KC3000 2TB NVME
Display(s) LG OLED CX48"
Case Corsair 5000D Air
Power Supply Corsair HX850
Mouse Razor Viper Ultimate
Keyboard Corsair K75
Software win11
It's forward-looking architecture, these things are clearly not going away, they will only become more and more prevalent.

Yeah, I'm used to the look of clean, stable images of DLSS that it's annoying to play without DLSS, particularly in the new Deathloop game, why doesn't it have DLSS yet arghhhhh.....


How strong will Intel ARC GPU's will be for rasterization is what I'm wondering looking at those specs on it looks stout on the TMUs and ROPs plus tons of tensor cores for inference image scaling. Raja may actually have pulled a rabbit out of the hat despite all the harsh criticism.

Yeah I think that Intel is banking on XeSS being the most preferable form of rendering, meaning players would default to 4K XeSS Q for all intent and purposes and would forgive the lackluster rasterization performance.
I mean if 4K XeSS look better than 4K Native in all aspects, comparing 4K Rasterization for apple to apple comparison is just pointless.
 
Last edited:
Joined
May 31, 2016
Messages
4,323 (1.51/day)
Location
Currently Norway
System Name Bro2
Processor Ryzen 5800X
Motherboard Gigabyte X570 Aorus Elite
Cooling Corsair h115i pro rgb
Memory 16GB G.Skill Flare X 3200 CL14 @3800Mhz CL16
Video Card(s) Powercolor 6900 XT Red Devil 1.1v@2400Mhz
Storage M.2 Samsung 970 Evo Plus 500MB/ Samsung 860 Evo 1TB
Display(s) LG 27UD69 UHD / LG 27GN950
Case Fractal Design G
Audio Device(s) Realtec 5.1
Power Supply Seasonic 750W GOLD
Mouse Logitech G402
Keyboard Logitech slim
Software Windows 10 64 bit
First glance at the comparison of the two techniques with native, and i can say that DLSS give the impression of smoothing the image while FSR is sharpening it.
In both cases pros and cons.
FSR has a slight problem with the lines like the net but not necessarily if you ask me. It looks like the light is coming differently on those and these look different than DLSS. If you zoom in you see the net is kinda incomplete on the FSR.
But no doubt I can say that DLSS is blurrier than Native and far blurrier than FSR. You can clearly see this in the video comparison side by side. Like the scene where the main character is climbing the wall. How the wall and the surroundings look. It does look better on FSR. Or with the junk car at the street where the street crack is. FSR make is look more natural.
I think it is a matter of balancing the smoothness and sharpness.
 
Joined
Jan 26, 2020
Messages
416 (0.27/day)
Location
Minbar
System Name Da Bisst
Processor Ryzen 5800X
Motherboard GigabyteB550 AORUS PRO
Cooling 2x280mm + 1x120 radiators, 4xArctic P14 PWM, 2xNoctua P12, TechN CPU, Alphacool Eisblock Auror GPU
Memory Corsair Vengeance RGB 32 GB DDR4 3800 MHz C16 tuned
Video Card(s) AMD PowerColor 6800XT
Storage Samsung 970 Evo Plus 512GB
Display(s) BenQ EX3501R
Case SilentiumPC Signum SG7V EVO TG ARGB
Audio Device(s) Onboard
Power Supply ChiefTec Proton Series 1000W (BDF-1000C)
Mouse Mionix Castor
First glance at the comparison of the two techniques with native, and i can say that DLSS give the impression of smoothing the image while FSR is sharpening it.
In both cases pros and cons.
FSR has a slight problem with the lines like the net but not necessarily if you ask me. It looks like the light is coming differently on those and these look different than DLSS. If you zoom in you see the net is kinda incomplete on the FSR.
But no doubt I can say that DLSS is blurrier than Native and far blurrier than FSR. You can clearly see this in the video comparison side by side. Like the scene where the main character is climbing the wall. How the wall and the surroundings look. It does look better on FSR. Or with the junk car at the street where the street crack is. FSR make is look more natural.
I think it is a matter of balancing the smoothness and sharpness.
I have seen the same in other games, DLSS smudges textures for some reason where FSR keeps them closer to native. On the other hand, DLSS is a bit better with very slim lines.
Overall both give similar result quality wise, and that is impressive for both, but I would say more impressive for FSR since it does not need dedicated hardware.
 

Mussels

Freshwater Moderator
Staff member
Joined
Oct 6, 2004
Messages
58,413 (8.21/day)
Location
Oystralia
System Name Rainbow Sparkles (Power efficient, <350W gaming load)
Processor Ryzen R7 5800x3D (Undervolted, 4.45GHz all core)
Motherboard Asus x570-F (BIOS Modded)
Cooling Alphacool Apex UV - Alphacool Eisblock XPX Aurora + EK Quantum ARGB 3090 w/ active backplate
Memory 2x32GB DDR4 3600 Corsair Vengeance RGB @3866 C18-22-22-22-42 TRFC704 (1.4V Hynix MJR - SoC 1.15V)
Video Card(s) Galax RTX 3090 SG 24GB: Underclocked to 1700Mhz 0.750v (375W down to 250W))
Storage 2TB WD SN850 NVME + 1TB Sasmsung 970 Pro NVME + 1TB Intel 6000P NVME USB 3.2
Display(s) Phillips 32 32M1N5800A (4k144), LG 32" (4K60) | Gigabyte G32QC (2k165) | Phillips 328m6fjrmb (2K144)
Case Fractal Design R6
Audio Device(s) Logitech G560 | Corsair Void pro RGB |Blue Yeti mic
Power Supply Fractal Ion+ 2 860W (Platinum) (This thing is God-tier. Silent and TINY)
Mouse Logitech G Pro wireless + Steelseries Prisma XL
Keyboard Razer Huntsman TE ( Sexy white keycaps)
VR HMD Oculus Rift S + Quest 2
Software Windows 11 pro x64 (Yes, it's genuinely a good OS) OpenRGB - ditch the branded bloatware!
Benchmark Scores Nyooom.
We may need to see comparisons of video for this

I wonder how to achieve that without uploads butchering quality....

@W1zzard you're the filehosting guy, can something like a single second of 60FPS recording be captured in some lossless format? or would the filesizes be immense (or too small to be useful?)
 
Top