Is AMD's poor VR performance because AMD hasn't solved their frame latency issues?

ChrisC

Weaksauce
Joined
Jun 14, 2016
Messages
96
Having seen a couple of VR benchmarks so far, we know the current score is Nvidia 8 - 0 AMD. This is despite the Fury X having 8Tflops being beaten by a 1070 GTX which has only about 6Tflops. Something is not right.

In many games, most people will easily point out that the Fury X will run close to the 1070 GTX in frame rates and even far exceed the 1070 GTX when we run DX12/Vulkan games.

However it does not explain VR adequately. So far the most common excuse is that VR games are all DX11 and thus are putting AMD on a handicap.

By some chance, I came upon this old article. Radeon HD 7950 vs. GeForce GTX 660 Ti revisited

All the VR benchmarks have consistently shown the 1070 GTX to have lower frame latencies than the Fury X, despite the Fury X's 33% computing performance advantage. Even for low graphics loads such as Google's Tilt Brush, the Fury X still puts in higher frame latencies.

Back then even AMD admitted that frame latency was not considered. A driver update to reduce Radeon frame times. To AMD's credit, they said they will be addressing it and there have been driver updates to improve on this issue. But this has been a long while back. Has this issue basically been forgotten? Has VR now forced AMD's frame latency issues to rear its ugly head and is now biting AMD back?
 
I think they are saying whatever they can to buy time. I am firmly in the camp that believes both Polaris and the upcoming Vega are boat anchors from the previous regime (they used to be Island cards) that are weighing the RTG down, and they still need to sell them to stay profitable.

I don't think you will see anything truly competitive from AMD until Navi - the first 100% RTG card.
 
  • Like
Reactions: DPI
like this
To me Compute power is akin to horsepower on a sports car. Its pretty much meaningless by itself.

AMD has been the king of specs since the days of Northern Islands, but unfortunately it doesn't translate in high performance in games. It does great for mining though.
 
The problem depends on how AMD designed it's hardware, it might very well be they can't fix the latency unless they redesign the hardware. You could think of it as AMD focused it's hardware solution for performing better in a processing capacity so as to edge out the competition via it's specs in tflops but then under designed it's internal data paths (kind of the whole XBox One SRAM thing) and thus gained a latency issue. If something like that is the cause, it is not going to be solved by any software patch. That will require a newly designed board to fix. They'll do the best they can with software fixes but if there is a bottleneck in the hardware you'll only be able to go so far till you have to give it up and just put out new hardware to really fix it (XBox One Scorpio).

To use Stoly's example, regardless of how much more horsepower your vehicle has, if you are running on square wheels the lower horsepower car with circular wheels is going to win every time, and nothing you do will fix that till you change to circular wheels.
 
I am not thoroughly convinced that it is purely the hardware though.

Both PS4 and Xbone are now running AMD gpus, and PS4 is pushing PS VR really hard. If it is the hardware, is the GPU on the PS4 substantially different from the GCN architecture in the Fury X and 480? The reasoning is if it is the hardware, then latency on the PS4 should be just as high. This is why the old frame latency issue raised back in the 7970 days rang a bell and I thought that might be the most likely cause. That AMD has basically not really addressed the frame latency issues at all.

Besides, in that whole frame latency furor, it was never really revealed what caused the frame latency issues. Was it driver or hardware?
 
No, it's entirely a software problem. Induced by them redesigning the memory controller in GCN, and simultaneously having to cut money spent on drivers (cause the company was losing tons of money).

More complex memory controller = more eyes required to write the optimized support code.

Driver software to be tweaked to reduce Radeon frame latencies in series of updates

The frame times were fixed when AMD actually got around to optimizing their memory controller drivers for each game:

A driver update to reduce Radeon frame times

The problem is, all games require some amount of custom tweaks, so the problem has never gone away completely. And they've fallen completely off the wagon optimizing for VR, since it's such a small market. They'd rather get more mainstream games working properly given their limited budget.

But in the meantime, their marketing department will say whatever it takes to make the sale. So "VR for the people" :rolleyes:

Nvidia just has way more people to spare on driver optimizations. So the only games where AMD has a sizeable lead are the ones designed on AMD hardware.
 

Thanks for those links. I didn't catch those follow up articles by Scott.

One thing Scott says in his article:

One of the tougher questions we had for AMD, in the wake of our discovery of these latency issues and their subsequent move to fix them, was simply this: how can we know that we won't see similar problems in the future? Dodd addressed this question directly in our correspondence, noting that AMD will be changing its testing procedures in the future in order to catch frame latency problems and prevent them:

Up until this point we had mostly assumed that there were occasional flickers in frame rate, but we had thought these were related to the fact that modern games mostly have streaming architectures and limitations of scheduling in the OS. We definitely will start regular measurements to ensure we track improvements, and stop regressions. Long term, we want to work with game developers and Microsoft to ensure these kinds of latency issues don't keep cropping up.

That's exactly the sort of answer we want to hear, and we'll be watching and testing future Radeon drivers and GPUs in order to see how well AMD executes on that plan.

I think AMD has been let off the hook too much recently then. Many reviews seem to be lacking any frame rate latency measurements. Even [H]'s own review of the Radeon 480 did not have frame latency measurements although the VR reviews do make up massively for them.

Maybe a question for the [H] editorial crew, do video card reviews have to take their VR performance into account, or is VR performance going to be strictly kept within VR benchmark reviews?
 
Yeah, I agree: the playable settings comparative reviews have been ditched by [H]. I've stopped reading those reviews just because they now provide less information than TechPowerUp and Anandtech.

TechReport still does the latency tests though, and the RX 480 had pretty decent latency scores.

AMD's Radeon RX 480 graphics card reviewed

So yes, AMD is keeping on top of these things even today. They've obviously kept their word that they would fix their QA.

But for anything not mainstream, you *could* see higher latency.
 
Maybe a question for the [H] editorial crew, do video card reviews have to take their VR performance into account, or is VR performance going to be strictly kept within VR benchmark reviews?

These will remain two separate entities since the way each is evaluated is tremendously different.
 
Back
Top