Loading build...
Your Shops
Radeon VII Capable of DLSS Technique Using DirectML Says AMD, 'Too Early to Use Raytracing'
By Jon Sutton on January 18th, 2019 at 03:08pm - original article from game-debate

AMD could have a few interesting tricks up its sleeves with its upcoming Radeon VII graphics card according to Adam Kozak, senior marketing manager of AMD’s GPU division.

Kozak has claimed AMD has been “experimenting” with Windows ML/DirectML and Luxmark and suggests AMD's Radeon VII is capable of DLSS-like super-resolution support. He even claims AMD’s own Radeon VII can outstrip the GeForce RTX 2080’s performance by 62% in Luxmark’s OpenCL-based ray tracing render tests.

Windows ML, or Windows Machine Learning, is a method by which machine learning can be used to provide GPU hardware acceleration through the DirectML API using DirectX technology. There are a number of techniques that DirectML can take advantage of, including support for super-resolutions, animation techniques, and denoising. The super-resolution support is a technique that sounds very similar to Nvidia’s own DLSS (Deep Learning Super Sampling), utilising deep learning techniques to identify repeatedly used polygons and textures, essentially pre-calculating where the GPU performance needs to be directed.

The AMD Radeon VII has already certified for use with DirectML, the graphics extension for DX12.

The original interview with Kozak comes by way of 4Gamer.net, a Japanese-language gaming site. This means the translation from English to Japanese and back again is definitely a little rough, and something to bear in mind.

“[At] last year's GDC 2018, Microsoft announced a framework "Windows ML" for developing machine learning based applications on the Windows 10 platform, and "DirectML" that makes it available from DirectX,” said Kozak. “We are currently experimenting with obtaining the evaluation version SDK of DirectML, but Radeon VII shows excellent results in that experiment.

“By the way, Radeon VII scored about 1.62 times the GeForce RTX 2080 in "Luxmark" which utilizes OpenCL-based GPGPU-like ray tracing renderer. Based on these facts, I think NVIDIA's DLSS-like thing can be done with GPGPU-like approach for our GPU.”

While not strictly the same as Nvidia’s DLSS, if AMD can get an approximation of DLSS up and running without the use of Tensor Cores this potentially bodes well for the Radeon VII’s capabilities. At the moment, DLSS is the key differentiator between the Radeon VII and the GeForce RTX 2080, aside from the $100 price different.

In addition to the DLSS chatter, Kozak also had a few words to say on ray-tracing, a technology which he believes just isn’t ready yet as it can only be used by a tiny fraction of PC gamers.

“I understand the importance of DXR, but I think it is too early to use it in games. Ray tracing is useful for professional video production sites, but there is Radeon Pro Render for such applications. The [only compatible game] is Battlefield V, and the 3DMark Port Royale benchmarks also. It will be unrelated to 99% of gamers. It is an understanding that there is little benefit for gamers so far.”

It's a fair point, although if Nvidia continues to make impressive strides with RTX optimisation and performance it could become moot. It's the DLSS-like features that have us interested for now though, and whether AMD is actually planning to support the technique anytime soon.amd