GeForce RTX 30 vs. AMD Radeon RX 6000 Series Vulkan Ray-Tracing On Linux

[phoronix]

Phoronix: GeForce RTX 30 vs. AMD Radeon RX 6000 Series Vulkan Ray-Tracing On Linux

Given this week's launch of the Radeon RX 6600 XT and that also bringing the new Radeon Software for Linux 21.30 driver, I was curious to see how the Vulkan ray-tracing performance compares now against the NVIDIA GeForce RTX 30 series on Linux.

https://www.phoronix.com/vr.php?view=30444

[Myownfriend]

AMD did way better than I thought they would in these tests! Also I always like seeing those performance per watt tests.

[MadeUpName]

A bigger question than speed for me is if you render the exact same scene do you get the same image or are they producing wildly different scenes?

[Masush5]

Would be nice if the amdgpu-pro vulkan driver could actually run any raytracing game other than Q2RTX on linux. Metro Exodus hangs the GPU, Doom Eternal crashes on launch with both 21.20 and 21.30 and vkd3d-proton DXR tests still segfault, which means no DXR game is playable on AMD.

[uid313]

Does ray tracing on AMD hardware suck only on Linux, or does it also suck on Windows as well?

[schmidtbag]

Weird how AMD either does fantastic or horrendously bad, but no in-between.

[zxy_thf]

Does ray tracing on AMD hardware suck only on Linux, or does it also suck on Windows as well?

I heard of gaming people saying RDNA2's ray tracing performance is roughly the same as of NVIDIA's first Gen RTX.

Not bad, but also not not very competitive.

[piotrj3]

I heard of gaming people saying RDNA2's ray tracing performance is roughly the same as of NVIDIA's first Gen RTX.

Not bad, but also not not very competitive.

That depends what you compare it too. If you use just raytraced shadows like in shadow of tomb raider, raytracing performance might be competitive, but that is only because there is not much RT stuff going on. Then there is more RT stuff going on in raytracing reflections, then there is even more going on in global illumination and then there is even more going on in full path tracing like Minecraft/Quake2/Marbles demo.

In for example Minecraft RT, RTX 2060 is a lot faster then even 6800XT. In Shadow of Tomb Raider raytraced shadows, 2060 is just a tad slower then 6600XT.

And that is even skipping that a lot of truly advanced RT games, also have support DLSS and DLSS neural network is optimized also for raytraced titles, so DLSS on windows when you use raytracing is truly good argument.

[pipe13]

A bigger question than speed for me is if you render the exact same scene do you get the same image or are they producing wildly different scenes?

That's an excellent question whose answer does not, alas, readily lend itself to an automated test suite.

[sdack]

That's an excellent question whose answer does not, alas, readily lend itself to an automated test suite.

Actually it does. One can use ImageMagick and its compare command to compare images based on PSNR, RMSE, and other metrics. It prints out a simple number that lets you quantify how different images are from one another. It should be an easy thing to add to an automated test suite.

https://imagemagick.org/script/command-line-options.php#compare