Carzy Zarx’s PC Gaming Emporium - Catch Up on All the Latest PC Gaming Related News

[Bofferbrauer2]

Captain_Yuri said:

Here's something interesting regarding 6000 series and Ray Tracing from DF's weekly talks.

They discuss how the top tier 6000 series seems to perform similar to Turing in the new Cyberpunk patch when you have RT Ultra enabled. Now the interesting thing is what Alex says.

Developers have more programmability when it comes to BVH Traversal aspect on RDNA 2 cards on the consoles thanks to their low level access. While it's very use case dependent, it can certainly help with the performance if the developers can optimize for it since you can see ps5 being quite capable in Spiderman. The kicker is that on the PC side, while RDNA 2 supports it at a hardware level... Neither Direct X Ray Tracing or Vulkan Ray Tracing currently allows this. So it's something that will probably not happen on PC for a while and no one knows how much that will help.

This kinda reminds me of that whole "Mantle" situation that AMD tried to do way back when. They had issues with Direct X 11 performance but if developers specifically coded for Mantle which was AMD's low level API, they could have had quite a nice performance boost. The problem is, on the PC space, majority of the devs choose Direct X 11 as that would cover both GPUs. And since Nvidia's market share was huge, there wasn't much incentive to do so.

We will see if AMD can get Microsoft or Kronos Group to enable such programmability for their cards on the PC side anytime soon. And of course, the age old question of... Whether or not the devs will bother outside of AMD sponsored titles.

Funny you mention mantle, since it's at the same source as the weakness of current NVidia cards with older and weaker CPUs.

Back in those days, NVidia had a better scheduler, and it was running on the CPU instead of the GPU, freeing up even more ressources. Those were the days where a 4c/4t i5 was largely enough and even a 2c/4t i3 was pretty sufficient for gaming. AMD then brought in Mantle as a low-level API to get closer to metal and rendering it's weakness moot. Microsoft also felt inspired and made DX 12 much more low-level than previous iterations. The result of all this: Nvidia suddenly fell behind in DX12 titles because of their scheduler not being apt for it. 

Now, instead of adapting and letting the API schedule everything, NVidia overhauled it's scheduler to keep up and to keep their advantage in DX 11 titles. The problem: To do so, it started needing a substantial amount of CPU cycles, which is at the heart of the current problem: Due to needing so much CPU time, a computer with an NVidia GPU hits the CPU limits much earlier than with a Radeon GPU, the more so the bigger the GPU is.

It took close to 10 years, but AMD turned their scheduling weakness into an asset now. Let's see if they can do a similar thing with raytracing now...