Nvidia GeForce RTX 2070 Super vs. AMD Radeon RX 5700 XT: 2020 Update

[Vulcanproject]

I wouldn't quite count on that. 4A came out and said that retooling Metro Exodus for 2.0 from 1.x would take too much time and distract from other projects. And that's for a title that already had a lot of DLSS work done to it.

Control is the *best* case scenario for implementation as they've had NV consultation heavily involved in the title since day one. They're one of the first studios to get something out the door with this, but it's a title that was built almost exclusively with Gameworks tools from go.

It's not like DLSS 2.0 was dropped on the world 4 weeks ago and it was just plug and play for the titles that have it. It's been tested and refined with those titles for months, with those studios, to get it where it is now and still isn't a drop in for title that already have DLSS hooks.

Nvidia has removed the requirement to train the AI network on a per game basis. This was the key stumbling block to industry wide acceptance. DLSS 1.0 basically meant you had to work with Nvidia closely and spend a lot of time tuning it for your title. No longer with DLSS 2.0.

It's not going to be cut and past right now, I didn't suggest it was. The Metro devs obviously spent quite a lot of time working with Nvidia on the first implementation and feel that they have done enough with their game for now. They gotta get paid same as everyone else.

Nvidia are just working towards it, but they need to work alongside developers too. They need continual refinement and technically experienced partners to advance the quality of the algorithms. Not every developer will have a member on the team with the time or background.

It appears that Nvidia are suggesting it to Gameworks partners. They are already giving out a great deal of support to them so it's a logical step to work with those games first on DLSS 2.0.

Beyond that we'll just have to wait and see. In the original implementation Nvidia added a fair few more games after a few months (I think it was up to 25) and clearly had aims to rapidly expand compatibility.

 

[Shirley Dulcey]

The deals can also make a difference.

Typically an RX 5700 XT is a $400 card. But I was able to buy an open box one (full return privileges, warranty, and even the free games from AMD) for $300. At that price level I'd be comparing it to a 2060 KO, which is no contest at all; the store didn't have any comparable deals on NVIDIA cards.

 

[Ravalo]

And you have 0% support for raytracing on the current Navi cards which automatically make them a non starter for anyone looking at gaming beyond this year.

I'll take tech that has a future over dead end good deals today.

But AMD will enable support for ray tracing in some months for navi cards

Right now I don’t care about ray tracing because most games I play dont even have support for ray tracing (Rainbow Six Siege, TF2, CSGO, etc.)

 

[LeroN]

It's interesting. The DLSS support guesses more long life for investments in the middle segment video cards so you won't worry about the driver's optimisation for old generations that is a pain now. Also even some obsolete hardware could bring pretty decent performance or quality. If AMD won't answer to Nvidia soon then it's possible to loose the market for them while 3-5 years. But a matter is if Nvidia keeps DLSS support for the current generation after 3-4 years for example. How the Internet connection speed and quality can impact the performance of DLSS? Still there are some questions and no answers. However the DLSS looks very promising and I like the idea.

 

[Evernessince]

Yeah let me just give you some idea about DLSS, these 3 images all use Nvidia Freestyle IS
Resolution scale 80%
Native
DLSS
DLSS actually look better than native resolution, all the jaggies are gone, and this is with all the settings maxed out. Nvidia DLSS 2.0 white page mention about Nvidia use 16k images to train the AI network, that's why they are called super sampling for a reason.

Yes the jaggies are definitely gone but there's still a clear loss in detail. Look at the dark spots on the metal pane of the windowsill on the bottom left hand corner. The shape of the spot is slightly different than the up-scaled picture and you can see what appears to be Dithering near the edges. There's a faint pattern to the bottom metal pane that's completely erased in the DLSS version as well. You can see the dithers all throughout the DLSS screenshot. This is just a consequence of rendering from a lower resolution. That's why the material color in the DLSS version is slightly different. Fine grain details like that applied over large surface are much easier to spot a difference with.

Just from how effective the anti-aliasing is, I really wish they'd develop some AA only version of DLSS. The big question is, for those looking to save performance, how does this compare to Nvidia's FreeStyle or AMD's RIS image quality wise? Do you know of anywhere that has the same picture with all 3 technologies? I would just like to know which has the best image quality in comparison to the performance hit it brings.

 

[Evernessince]

But AMD will enable support for ray tracing in some months for navi cards

Right now I don’t care about ray tracing because most games I play dont even have support for ray tracing (Rainbow Six Siege, TF2, CSGO, etc.)

It's possible given AMD is taking the shader core approach but it may not work that well with Navi . AMD added additional hardware functionality to it's existing cores and we are not sure right now whether that is critical to their RT performance.

Not that I think it's a big deal. There isn't a product on the market right now that can ray trace modern games to an acceptable performance level and actually have a significant visual impact. Newer high poly games typically only do a single hardly noticeable effect. Even games like Minecraft, which can run on a potato, restrict the cluster size to 8 from the default 96. This means that you are looking at 8.3% of the render distance and less then half the FPS with RTX on a level that has a significant visual impact. That's before considering that's on an extremely simple game and as GamersNexus noted, there is still clearly noise artifacts.

 

[Puiu]

Do you even have an RTX card to back up your claim lol. DLSS in quality mode can boost performance anywhere from 35-70% depending on the title while giving approximately equal visual to native resolution.
Let just leave the image sharpening subject aside because you can apply that to native resolution and DLSS too, so 80% resolution scaling with IS would objectively look worse than native resolution with IS.
I already post some uncompressed images from Mechwarrior 5 in the post above, you can download and compare between the 80% resolution scaling, native and DLSS images.

I don't need an RT card to know simple facts demonstrated by many reputable places, including techspot. others know how to do testing better than me. I don't care about your 1 example, especially when there are so many ways you can tweak things from resolution to algorithms (you can even use AMD's own sharpening algorithm on Nvidia cards if you so desire)

 

[krizby]

Yes the jaggies are definitely gone but there's still a clear loss in detail. Look at the dark spots on the metal pane of the windowsill on the bottom left hand corner. The shape of the spot is slightly different than the up-scaled picture and you can see what appears to be Dithering near the edges. There's a faint pattern to the bottom metal pane that's completely erased in the DLSS version as well. You can see the dithers all throughout the DLSS screenshot. This is just a consequence of rendering from a lower resolution. That's why the material color in the DLSS version is slightly different. Fine grain details like that applied over large surface are much easier to spot a difference with.

Just from how effective the anti-aliasing is, I really wish they'd develop some AA only version of DLSS. The big question is, for those looking to save performance, how does this compare to Nvidia's FreeStyle or AMD's RIS image quality wise? Do you know of anywhere that has the same picture with all 3 technologies? I would just like to know which has the best image quality in comparison to the performance hit it brings.

You are in luck, Mechwarrior 5 just seems to have every feature from AMD FidelityFX to Nvidia DLSS and RTX.
For your comparison, I took screenshots of Mechwarrior 5 at 75% resolution scale as this provide about the same FPS as DLSS ON
75% with FidelityFX OFF
75% with FidelityFX ON
75% with Nvidia Freestyle
Native resolution with FidelityFX
Native resolution with Freestyle
DLSS ON with Freestyle
DLSS ON RTX ON with Freestyle

Freestyle seems to cost about 4% more performance compare to FidelityFX (or RIS), but the advantages of Freestyle are it's applicable to DX11 games and you can adjust the Sharpening value depend on how you like it, as opposed to fixed value of RIS.
Note: I didn't know that turn on and off DLSS would turn off TAA and FidelityFX so the images from earlier post were with TAA off. These images are will TAA on. TAA make the sparks from the welding machine look horrible though.

I don't need an RT card to know simple facts demonstrated by many reputable places, including techspot. others know how to do testing better than me. I don't care about your 1 example.

I don't want to argue with someone who has nothing to backup his claim but keep talking nonsense anyways.
If you value Techspot testing, Tim did say DLSS provide equivalent Image Quality to native resolution while offering 30% more performance.
All the nonsense you spew are lowering render resolution with FidelityFX would have the same effect, but you forget that you can use FidelityFX or Freestyle with native resolution or DLSS anyways, making your entire point invalid as lowering render resolution will always look worse than native resolution.
I will demonstrate with 3rd grade mathematics that even you would understand better:
0.75 + RIS < 1 + RIS

Here is the comment from Steve the author of Techspot that you seems to miss

As for the DLSS comment. No, you can't simply use RIS and that makes it all okay. Nvidia also has an image sharpening technology which is arguably better, so that neutralizes RIS. Moreover, Nvidia's image sharpening filter can be used with DLSS.

 

[Evernessince]

You are in luck, Mechwarrior 5 just seems to have every feature from AMD FidelityFX to Nvidia DLSS and RTX.
For your comparison, I took screenshots of Mechwarrior 5 at 75% resolution scale as this provide about the same FPS as DLSS ON
75% with FidelityFX OFF
75% with FidelityFX ON
75% with Nvidia Freestyle
Native resolution with FidelityFX
Native resolution with Freestyle
DLSS ON with Freestyle
DLSS ON RTX ON with Freestyle

Freestyle seems to cost about 4% more performance compare to FidelityFX (or RIS), but the advantages of Freestyle are it's applicable to DX11 games and you can adjust the Sharpening value depend on how you like it, as opposed to fixed value of RIS.
Note: I didn't know that turn on and off DLSS would turn off TAA and FidelityFX so the images from earlier post were with TAA off. These images are will TAA on. TAA make the sparks from the welding machine look horrible though.

I appreciate it.

Freestyle 75 and FidelityFX 75 (or CAS) look extremely similar. I could honestly not tell a difference.

Now comparing as apples to apples as we can get, the 75 CAS picture does have more jaggies then the DLSS + FreeStyle image so it's the winner in that department. The downside of DLSS + FreeStyle is that it looks oversharpened. The contrast between the pixels on some portions of the image don't look right. If you look at image 2, you can see pixels that are ordinarily supposed to form small subtle details being brought forth. It almost looks like a bunch of black dots in a pattern across the texture. This is a classic example of over-sharpening. The interesting part is that it only occurs when combined with DLSS. Freestyle does not have this issue when used alone.

For example:



same here as well


Of course this is only one scene from the game so we can't really draw any definitive conclusions. Just from this small sample I'd draw three observations: 1) DLSS + Freestyle has the least alaising but sharpens details that shouldn't be sharpened. 2) CAS and Freestyle are identical. 3) Graphical quality between these three methods is very similar.


It's not directly comparable as the image was taken at a different part of the game but the earlier DLSS only image you took did appear to have comparable amount of aliasing to freestyle / CAS. Once again though, different scene. A DLSS only sample was not provide for this specific scene. It would take a significant amount of review to make actual conclusions but just from this small sample, it has already interested me.

 

[krizby]

I appreciate it.

Freestyle 75 and FidelityFX 75 (or CAS) look extremely similar. I could honestly not tell a difference.

Now comparing as apples to apples as we can get, the 75 CAS picture does have more jaggies then the DLSS + FreeStyle image so it's the winner in that department. The downside of DLSS + FreeStyle is that it looks oversharpened. The contrast between the pixels on some portions of the image don't look right. If you look at image 2, you can see pixels that are ordinarily supposed to form small subtle details being brought forth. It almost looks like a bunch of black dots in a pattern across the texture. This is a classic example of over-sharpening. The interesting part is that it only occurs when combined with DLSS. Freestyle does not have this issue when used alone.

For example:

View attachment 86455

same here as well
View attachment 86457

Of course this is only one scene from the game so we can't really draw any definitive conclusions. Just from this small sample I'd draw three observations: 1) DLSS + Freestyle has the least alaising but sharpens details that shouldn't be sharpened. 2) CAS and Freestyle are identical. 3) Graphical quality between these three methods is very similar.


It's not directly comparable as the image was taken at a different part of the game but the earlier DLSS only image you took did appear to have comparable amount of aliasing to freestyle / CAS. Once again though, different scene. A DLSS only sample was not provide for this specific scene. It would take a significant amount of review to make actual conclusions but just from this small sample, it has already interested me.

Very good observation, for CAS and Freestyle to look the same I increased the sharpen filter to 60%. However DLSS does have sharpen effect of its own, making DLSS + Freestyle 60% look over-sharpened as you mention, but it's easily fixed though.
If you compare the level of details, DLSS easily outdoes the native resolution, as showed here.

75% CAS - 100% CAS - DLSS Freestyle

Jup, this is with maximum magnification, I don't know if anyone can notice while playing game. However it is the solid truth that DLSS Image Quality is the same (or better) than native res.

 

[meric]

We are comparing these effects looking at still photos but would it be possible to even discern the quality difference (other than the fps) in game?

 

[krizby]

We are comparing these effects looking at still photos but would it be possible to even discern the quality difference (other than the fps) in game?

Everything from Image Quality to FPS are entirely subjective, nothing stop you from enjoying games even if you have a potato PC. Though for the sake of benchmarking GPU, either Image Quality or FPS must be normalized. Every reviewer these days choose to normalized IQ and look for FPS difference while HardOCP (gone now, the editor went to work at Intel) normalized playable FPS and look for Image Quality difference (which GPU can afford higher IQ settings).

So yeah, normalized IQ is the keyword here, you can't just reduce IQ by 5% for a 20% uplift in FPS and call it a day, since competing GPU can just do the same. That's also the reason DLSS 1.0 results were omitted.

 

[Evernessince]

Very good observation, for CAS and Freestyle to look the same I increased the sharpen filter to 60%. However DLSS does have sharpen effect of its own, making DLSS + Freestyle 60% look over-sharpened as you mention, but it's easily fixed though.
If you compare the level of details, DLSS easily outdoes the native resolution, as showed here.

75% CAS - 100% CAS - DLSS Freestyle

Jup, this is with maximum magnification, I don't know if anyone can notice while playing game. However it is the solid truth that DLSS Image Quality is the same (or better) than native res.

DLSS + Freestyle appears over-sharpened again. The small details on the welder arm that aren't supposed to be prominent are. This is the same method that produced the black speckle all over example 2 above. You can see more details with DLSS + FreeStyle but the downside is that many of those details were never meant to play a prominent role to begin with and you get overshaprened artifcats.

 

[treetops]

Yeah let me just give you some idea about DLSS, these 3 images all use Nvidia Freestyle IS
Resolution scale 80%
Native
DLSS
DLSS actually look better than native resolution, all the jaggies are gone, and this is with all the settings maxed out. Nvidia DLSS 2.0 white page mention about Nvidia use 16k images to train the AI network, that's why they are called super sampling for a reason.

It depends on how many titles implement it and how well it works in those titles. With a few outlier results where DLSS will give better FPS, I think that just using Nvidia's own image sharpening and an 80% resolution scaling should give you similar results. The same goes for AMD cards too.

You can use Trixx boost and image sharpening on any game. Or Reshade on Nvidia\AMD and turn down the resolution. Your example lacks Image Sharpening.

Nvidia Freestyle vs. Radeon Image Sharpening vs. Reshade

After testing AMD's new Radeon Image Sharpening feature for Navi GPUs, we've gone back for even more testing. We're now comparing RIS with other sharpening options including...

www.techspot.com

 

[Rdmetz]

All that really matters to me in all of this is the fact that I can get a 30% uplift in performance and not lose image quality compared to native anything else really doesn't matter if Im playing and "seeing" the same quality yet my game "feels" better (because of the higher fps) then that's the way I'm going to want to play the game.

I'm not bringing RIS or freestyle or anything else into it that's IN ADDITION to what DLSS can offer me and is much more subjective.

AMD has no answer to DLSS like technology and until they do I can't see them being a choice I'd take over the extra performance on Nvidia.

 

[Evernessince]

All that really matters to me in all of this is the fact that I can get a 30% uplift in performance and not lose image quality compared to native anything else really doesn't matter if Im playing and "seeing" the same quality yet my game "feels" better (because of the higher fps) then that's the way I'm going to want to play the game.

I'm not bringing RIS or freestyle or anything else into it that's IN ADDITION to what DLSS can offer me and is much more subjective.

AMD has no answer to DLSS like technology and until they do I can't see them being a choice I'd take over the extra performance on Nvidia.

Well the thing is, from the images above, FreeStyle / CAS are providing as good images as DLSS visually with around the same performance boost. They also work on a much wider swath of games as well. FYI I don't think FreeStyle + DLSS go well together at all.

I'm not discussing CAS / Freestyle as an addition to DLSS, I'm discussing it as direct competition. Both in image quality and performance, they appear to be very competitive when compared to DLSS.

If all you care about is a performance boost without loosing image quality, any one of those 3 technologies appears to be a good choice.

 

[krizby]

You can use Trixx boost and image sharpening on any game. Or Reshade on Nvidia\AMD and turn down the resolution. Your example lacks Image Sharpening.

Nvidia Freestyle vs. Radeon Image Sharpening vs. Reshade

After testing AMD's new Radeon Image Sharpening feature for Navi GPUs, we've gone back for even more testing. We're now comparing RIS with other sharpening options including...

www.techspot.com

Is it that hard to understand that Lowering render resolution and Image sharpening are available to both vendors at any resolution, therefore they cancel each other out.
Here is a 3rd grade equation:
0.75 + RIS < 1 + RIS (Image Quality, 0.75 is native resolution multiplier)

You can't just use a lower Image Quality on one card then compare the FPS result to the other, however minor Image Quality reduction that is. You might as well test a 5700XT at High settings then say it gives higher FPS than 2070 Super at Ultra settings and they both look similar enough.

The outlier here is DLSS technology, which is exclusive to RTX cards only; Techspot has done extensive testing and concluded that it give equal Image Quality to Native Resolution, therefore can be brought into the performance testing.

DLSS also has 3 performance modes, Quality - Balanced - Performance, which lower the internal rendering resolution. If you want to benchmark at 80% resolution scale, DLSS in Balanced Mode would also give higher FPS while giving the same Image Quality.

It goes like this, if you normalized for Image Quality, DLSS win in FPS. If you normalized for FPS, DLSS win in Image Quality, it's that simple.

Ps: it's easy to create a custom resolution if a game doesn't support resolution scaling, you don't need the crash-prone Trixx Boost for that.

 

[treetops]

Is it that hard to understand that Lowering render resolution and Image sharpening are available to both vendors at any resolution, therefore they cancel each other out.
Here is a 3rd grade equation:
0.75 + RIS < 1 + RIS (Image Quality, 0.75 is native resolution multiplier)

You can't just use a lower Image Quality on one card then compare the FPS result to the other, however minor Image Quality reduction that is. You might as well test a 5700XT at High settings then say it gives higher FPS than 2070 Super at Ultra settings and they both look similar enough.

The outlier here is DLSS technology, which is exclusive to RTX cards only; Techspot has done extensive testing and concluded that it give equal Image Quality to Native Resolution, therefore can be brought into the performance testing.

DLSS also has 3 performance modes, Quality - Balanced - Performance, which lower the internal rendering resolution. If you want to benchmark at 80% resolution scale, DLSS in Balanced Mode would also give higher FPS while giving the same Image Quality.

It goes like this, if you normalized for Image Quality, DLSS win in FPS. If you normalized for FPS, DLSS win in Image Quality, it's that simple.

Ps: it's easy to create a custom resolution if a game doesn't support resolution scaling, you don't need the crash-prone Trixx Boost for that.

Yeah they said the same thing about lowering res and using RIS. "It looks the same". Idk why you need ai to lower resolution and turn on image sharpening.

 

[krizby]

Yeah they said the same thing about lowering res and using RIS. "It looks the same". Idk why you need ai to lower resolution and turn on image sharpening.

The same as what ? Native resolution ? Does RIS somehow work better at lower internal resolution than at native resolution ? that would probably blow anyone mind right now.
Anyways AI is all about filling in missing details which is lost when you use a lower internal resolution. RIS is only a filter that is available for all parties.
Image Quality Wise:
DLSS = 1
0.75 + RIS = 1 (big IF here but whatever)
But
0.75 + RIS < 1 + RIS
Therefore
0.75 + RIS < DLSS + RIS (or Freestyle, they are the same anyways).

So yeah as Steve has said, you can take RIS out of the equation and it doesn't make any difference.

Well the thing is, from the images above, FreeStyle / CAS are providing as good images as DLSS visually with around the same performance boost. They also work on a much wider swath of games as well. FYI I don't think FreeStyle + DLSS go well together at all.

I'm not discussing CAS / Freestyle as an addition to DLSS, I'm discussing it as direct competition. Both in image quality and performance, they appear to be very competitive when compared to DLSS.

If all you care about is a performance boost without loosing image quality, any one of those 3 technologies appears to be a good choice.

Here is a blind test for you, can you pick out which one look the best and how comparable they are to each other ?
1 2 3

 

[yeeeeman]

Technically FreeStyle can be used with DLSS 2.0 but that's if the game is on both the FreeStyle white list and has DLSS 2.0 support. You guys said in your RIS vs FreeStyle article:

"But there is a whitelist of games that it works with, so titles like Hitman 2 and Resident Evil 2 aren’t supported, for example."

If you are looking to save performance, which approach is best?

You guys did an article with Nvidia FreeStyle vs RIS: https://www.techspot.com/review/1884-amd-ris-vs-nvidia-freestyle-vs-reshade/

and one with DLSS 2.0: https://www.techspot.com/article/1992-nvidia-dlss-2020/

but nothing comparing the new DLSS 2.0 to any of the existing technologies in image quality and performance. I would not mind seeing DLSS 2.0 + FreeStyle vs RIS, DLSS 2.0 vs FreeStyle. I realize that image quality is somewhat subjective but given high resolution images, we should be able to see material benefit.

RIS is image sharpening. It doesn't obtain the same results from a lower resolution.
DLSS starts from a lower res, hence more FPS and gives you higher res results. So sure, we can compare ris, his, miss, whatever you like with dlss, but it just doesn't make sense because ris doesn't give you anything other than a few more fps compared to taa, whereas dlss can double your fps in certain cases for same resolution.

 

[Evernessince]

RIS is image sharpening. It doesn't obtain the same results from a lower resolution.
DLSS starts from a lower res, hence more FPS and gives you higher res results. So sure, we can compare ris, his, miss, whatever you like with dlss, but it just doesn't make sense because ris doesn't give you anything other than a few more fps compared to taa, whereas dlss can double your fps in certain cases for same resolution.

You have not been following this thread. Read above for answers, they've been said before.

 

[john 93]

For me still, AMD is the best and will remain so.

 

[christiangeli]

Nvidia video card price are too high, driver very poor compared to radeon adrenaline, gaming optimization nonexistent. Nvidia GPU are like Apple product, overpriced overall is the only benefit.
Never mind DLSS, I prefer play 1080p at fixed resolution on a good 1080p monitor , than to play at 4k and introducing latency. You need dlss because your gpu can't handle 4k perfectly and your 4k monitor is stretching pixel at 1440p and 1080p. The fact that dlss are not reducing the game visual is that game texture are not optimized for 4k.

 

[Sam Gupta]

The 5700 XT is a good value. I just bought a 2070 Super, but probably would have gotten a 5700 XT if I didn't have a G Sync monitor. Also, I was able to find a 2070 Super on sale for $460. I don't think I would have paid $500.

5700 XT @ $400 > 2070 Super @ $500+.

And what if I am getting RTX 2070 super ₹2000 cheaper than 5700 XT?
Should I buy it?

 

[Puiu]

And what if I am getting RTX 2070 super ₹2000 cheaper than 5700 XT?
Should I buy it?

If the price is similar or lower then it is a no brainer to go with the 2070 Super. But good luck finding such a good deal.

Is the 5700 XT is very expensive where you live? The Nvidia cards are very expensive in Romania. The cheapest 2070s is more than $550 without VAT (+$660, 19% VAT included) and the average AIB card is 60-100$ above that price. The cheapest 5700 XT is more than $460 (+$550, VAT included) with several AIBs hovering this price point and some $50 above it.