[SOLVED] Nvidia DLSS vs AMD card ? Will Amd also support a similar technique?

david_the_guy

BANNED
May 11, 2019
77
14
35
0
Okay, we all know DLSS has been used on RTX cards, and Nvidia implemented it, and it kind of doesn't look good to be honest. I don't like the DLSS image quality. What about AMD cards ? Can we get a similar upscaling feature, or it's not possible.

thoughts ?
 
YES, its possible. That's a good question though.

AMD Radeon VII, and few other AMD cards support DirectML, which is an alternative to NVIDIA’s Deep Learning Super Sampling (DLSS).

Think of DirectML as the Machine Learning equivalent of DXR (DirectX Raytracing), allowing DX 12 to support advanced features, and utilize AI to improve future games. DirectML is a Machine Learning (ML) extension to DirectX, and Microsoft and NVIDIA/AMD have already showcased it super sampling video-games by using the AI.

This is similar to Deep Learning Super Sampling (DLSS), and there no need to use third party tools for any custom resolution settings. This feature is coming soon.

Now similarly to DXR, and while Radeon VII could support DirectML, it will be up to the developers to implement this feature in their games. Moreover, we don’t know whether AMD GPU would be offering a big of a performance boost similar to the RTX series. Because, contrary to the RTX series, the Radeon VII does not feature any AI-specific hardware (like NVIDIA’s Tensor Cores) so it will be basically running DirectML in “software” mode.

But nonetheless, DirectML is the answer to DLSS.

According to Microsoft, DirectML supports all DirectX 12 compliant hardware and uses DX12 to support advanced features and utilize AI to improve future games. DirectML supports all DirectX 12 compliant hardware, much like DXR, and like DXR, it can also exploit the hardware acceleration capabilities of modern graphics architectures.

Microsoft has already showcased the potential of machine learning in gaming applications (see image below) before, as to what happens when Machine Learning is used to upscale an image to four times its original resolution (basically from 1080p to 4K) to generate a sharper final image and reduced aliasing. The image below is a comparison between ML Super Sampling and bilinear upsampling.

IF successful, an AMD-powered AI upsampling has a lot of potential for use within both next-generation games consoles and PCs users alike. But I guess only TIME will tell. It's a bit early to comment on this feature.



You can see here that the image on the left side is much sharper as well.

 
Last edited:
Reactions: david_the_guy

lux1109

Proper
BANNED
Apr 30, 2019
117
25
110
0
I really don't like DLSS, but the above upscaling looks good. The image on left side is very sharp and clear. Good to know AMD has something up their sleeves. when are we going to see this in games though ?
 
to be honest i don't think AMD will ever make something similar to DLSS. the primary idea behind of DLSS is to train the AI to produce correct image. this is time consuming and can be costly. it is not simply train the image once and be done with it. IMO DLSS can be a big success or big flop to nvidia. if they were a flop then nvidia will just move on onto something new. FXAA, TXAA, MFAA, DSR and now DLSS.
 

lux1109

Proper
BANNED
Apr 30, 2019
117
25
110
0
Yes AMD will make an equivalent for DLSS, like Metal messiah mentioned above, via directML. even in future they might have a dedicated hardware for this tech, next navi gpus.
 
to be honest i don't think AMD will ever make something similar to DLSS. the primary idea behind of DLSS is to train the AI to produce correct image. this is time consuming and can be costly
Nah. What makes you think that ? In a recent interview with 4Gamers, AMD's Adam Kozak confirmed that their upcoming Radeon VII graphics card would support DirectML, a Machine Learning (ML) extension to DirectX.

Also, MS/AMD/Nvidia have already showcased this technique during one of Microsoft's SIGGRAPH 2018 tech talks. That talk, which was entitled "Deep learning for real-time rendering: ""Accelerating GPU inferencing with DirectML and DirectX 12"", showcased hardware upscaling Forza Horizon 3 from 1080p to 4K using DirectML in real time.

DirectML indeed has the potential to improve the graphical fidelity of future console and PC games.
 
Reactions: lux1109
Yes AMD will make an equivalent for DLSS, like Metal messiah mentioned above, via directML. even in future they might have a dedicated hardware for this tech, next navi gpus.
it's not about able to do it or not. but are they willing to do it or not. also nvidia DLSS directly taking advantage of tensor core. AMD made Vega 20 to compete directly with nvidia Tesla V100 but Vega 20 did not have any tensor cores.
 
Nah. What makes you think that ?
cost. no doubt AMD can do it via directML but those images need to be trained. for nvidia DLSS they have said it they will do it free for game developer. all they need to do is provide them with the game and they will trained the image on their own super computer (which quite fast itself for being in the world's top 20 when they were first coming out in 2017). if you have been following AMD and nvidia for years you should be aware how AMD roll. i can safely said there is no way AMD wiling to eat the cost. if anything they will try to push the cost onto someone else. just look what happen with Bullet, AMD stereoscopic 3D implementation and even Freesync. good thing with Freesync it was eventually being sort out in the long run.
 
They are willing. It has already been showcased as well. You don't seem to fully understand how all this works.
what they showcase is RIS (Radeon Image Sharpening) vs DLSS. but it did something totally different than what nvidia been doing with DLSS. though what AMD intend to showcase is how what they have been doing with RIS is increasing image quality while DLSS will decrease it. there is no image being trained by AI at all with RIS. and when they do this instead of using machine learning to create their own version of DLSS you already know AMD is not wiling to do something similar to nvidia DLSS.

nvidia have non public event being held around the E3 which attended by several journalist including WCCFTECH. they said they still can't disclosed what it is all about (most likely related to nvidia Super) but they did ask nvidia what they think of AMD new RIS and Anti Lag.

https://wccftech.com/nvidia-weve-had-sharpening-nvidia-freestyle/
 

david_the_guy

BANNED
May 11, 2019
77
14
35
0
YES, its possible. That's a good question though.

AMD Radeon VII, and few other AMD cards support DirectML, which is an alternative to NVIDIA’s Deep Learning Super Sampling (DLSS).

Think of DirectML as the Machine Learning equivalent of DXR (DirectX Raytracing), allowing DX 12 to support advanced features, and utilize AI to improve future games. DirectML is a Machine Learning (ML) extension to DirectX, and Microsoft and NVIDIA/AMD have already showcased it super sampling video-games by using the AI.

This is similar to Deep Learning Super Sampling (DLSS), and there no need to use third party tools for any custom resolution settings. This feature is coming soon.

Now similarly to DXR, and while Radeon VII could support DirectML, it will be up to the developers to implement this feature in their games. Moreover, we don’t know whether AMD GPU would be offering a big of a performance boost similar to the RTX series. Because, contrary to the RTX series, the Radeon VII does not feature any AI-specific hardware (like NVIDIA’s Tensor Cores) so it will be basically running DirectML in “software” mode.

But nonetheless, DirectML is the answer to DLSS.

According to Microsoft, DirectML supports all DirectX 12 compliant hardware and uses DX12 to support advanced features and utilize AI to improve future games. DirectML supports all DirectX 12 compliant hardware, much like DXR, and like DXR, it can also exploit the hardware acceleration capabilities of modern graphics architectures.

Microsoft has already showcased the potential of machine learning in gaming applications (see image below) before, as to what happens when Machine Learning is used to upscale an image to four times its original resolution (basically from 1080p to 4K) to generate a sharper final image and reduced aliasing. The image below is a comparison between ML Super Sampling and bilinear upsampling.

IF successful, an AMD-powered AI upsampling has a lot of potential for use within both next-generation games consoles and PCs users alike. But I guess only TIME will tell. It's a bit early to comment on this feature.



You can see here that the image on the left side is much sharper as well.


Thanks for the detailed information. That was helpful. Didn't know about DirectML before. Good to know AMD is working on this as well. But yes, you are absolutely correct, I have verified this with few other sources as well, and according to them AMD will indeed use this as an alternative to DLSS.

:D
 
Reactions: Metal Messiah.
What about this year's siggraph expo 2019 ? Is it happening again, or we need to wait for some announcement ?
2019's SIGGRAPH begins on Sunday, 28 July and it will end on Thursday, 1st August. Earlier the date was tentative, but now it's once again confirmed and finalized.

cost. no doubt AMD can do it via directML but those images need to be trained.
I honestly don't think cost would be a huge detrimental factor for AMD. This isn't hardware emulation, but more like Software. And they have already planned to use this in future. I agree with some of your points though, but I guess we need to wait and see what AMD has to offer. With the high-end flagship NAVI GPUs, they are going to implement Ray tracing on the hardware level as well.

ML techniques such as denoising and super-resolution will allow hardware to achieve impressive raytraced effects with fewer rays per pixel. I expect DirectML to play a large role in making raytracing more mainstream.


 
Last edited:
honestly don't think cost would be a huge detrimental factor for AMD.
even if they are not they still prefer not to spend any money and resource as much as possible and let others shoulder the burden. this has been very clear in several of their initiatives for the past decade.

This isn't hardware emulation, but more like Software.
that's the second problem.
 

ASK THE COMMUNITY

TRENDING THREADS