• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
  • Hey Guest. Check out the NeoGAF 2.2 Update Thread for details on our new Giphy integration and other new features.

|OT| AMD Radeon RX6800/RX6800XT Reviews/Benchmarks Thread |OT|

supernova8

Member
Jun 2, 2020
1,643
2,279
415
As I said, RT hardware step alone doesn't produce anything palatable, you can't disable everything else and get it rendered, you could, perhaps, disable some of the "other steps".



That's curious, ignoring what wccf is, but far from obvious, what it is being tested to begin with.
It's the demo that was first shown on Vega GPU, with no hardware RT being used, I'm puzzled if it is used in this demo at all.

The hardware intersection bit I'm referring to, is about "up to" number, of NV claimed "up to 10 billion intersection" for Ampere (which is 10 times more than Tesla :)) per second, vs AMD's... 380 billion on XSeX. (again: up to)

Fishiness of "up to" aside (e.g. imagine "it's that many if all structures fit into LL1 cache") "It is largely not about intersections at all" explains, why we do not see even remotely as major RT perf jump as it is claimed.

Last, but not least, how many Crytek RT engine based games are out there? Not even one, right?

They gave a breakdown of the specs. Besides, if as you suggest there no was RT in the demo at all, AMD would win anyway because they have better raster performance, no?


You can find the details here. I personally don't care to read it.


DF did their own video on it anwyay. The RTX cards clearly beat the RDNA cards and again if you look at the WCCF video, the new RTX cards beat the RDNA2 cards.

I'm not entirely what you're criticizing about the ray tracing demo to be honest. In the DF video they explain that it's using "SVOGI" or some sort of method of using a simplified version of the geometry to generate ray tracing information (which is not as accurate as "proper" ray tracing) and then adds additional ray tracing info on top for moving objects.

Either way, the NVIDIA cards seem to win.

I think I sort of get what you're trying to say but the whole "in theory AMD should be better" counts for very little until AMD is actually (consistently) better.
 

Ascend

Member
Jul 23, 2018
3,262
4,666
525
There likely is a bug with AMD's RT, either in software or in hardware. After RT was enabled in Quake II RTX for the radeon cards, a comparison was made between the RTX 3080 and the RX 6800XT. Why this game? Because it has a nice feature that breaks down how much time each aspect of the rendering is taking, and in that game in particular, the 3080 is over 300% faster at BVH Update compared to the 6800XT;



Aside from that, the differences with regards to RT are mostly minor. Even more interesting is the fact that a 1080 Ti, which has no RT cores, does the BVH update at 0.47 ms, compared to the 1.38 ms of the 6800XT. That is the main bottleneck in RDNA2, because otherwise it's not really (significantly) worse on average.

As seen above, AMD is better at shadow maps and refraction in general, while nVidia is better at rays in general. But something is going on with AMD's BVH Update. Games will perform better on nVidia if they use more rays, while they would perform better on AMD if they use more refraction, except the BVH update throws that in the dust. And that is not part of the RT cores, because, as mentioned earlier, the 1080Ti does it faster than the 6800 cards.
 
Last edited:

llien

Member
Feb 1, 2017
9,483
7,467
935
They gave a breakdown of the specs. Besides, if as you suggest there no was RT in the demo at all, AMD would win anyway because they have better raster performance, no?
No. :)
As my argument was actually that it was non-RT stuff optimized for NV that gives it edge in NV sponsored RT games.

You can find the details here.
It is even titled "SOFTWARE ray tracing" :).

I think I sort of get what you're trying to say but the whole "in theory AMD should be better" counts for very little until AMD is actually (consistently) better.
I'm saying "it's not about RT hardware... at all".
AMD would need to bother and write code that runs better on their platform, or just onboard more developers, the way they did with Dirt5/GodSomething, but those bits have nothing to do with "ray tracing hardware".

In the chart Ascend Ascend put above, BHV update is the bit that shows the biggest difference, yet it is not something that is accelerated by RT hardware, it's just piece of code developed by NV and, of course, they aimed to optimize it for the own hardware.
And "but it's faster at raster" doesn't really apply here, AMD being faster at code developed specifically to use green card strength would mean worse than Intel's Zen3 situation in GPUs.
 

smbu2000

Member
Mar 6, 2012
925
115
385
Well I ended up selling my vanilla 6800 and managed to pick up a 6800XT. (MSI Gaming X Trio)
Ended up being not too much to upgrade to the 6800XT, so I went for it.
Thought about the 6900XT, but there was a much bigger price difference for one of those, so I didn't get it.

Happy with the 6800XT so far!
(My System is a 5900X with an Aorus X570 Xtreme)