AMD & NVIDIA GPU VR Performance: Valve's Robot Repair

[Unreal123]

Let's start with the "good." The NVIDIA TITAN X, GTX 1080, GTX 1070, and GTX 980 Ti provided the best possible experience while playing through Robot Repair. All four of these NVIDIA GPUs provided a low enough average GPU Render Time that we "never" went into Reprojection on our Vive headset. (I did put "never" in quotes there because technically five hundredths of one percent is not "never" and I am sure some fanboys would want to call me on that.) From a gameplay standpoint you would not be able to discern between the four GPUs in a blind test.



The second group of GPUs down the scale in Robot Repair performance include both the GTX 1060 and R9 Fury X. In terms of GPU Render Time, the Fury X is a bit faster in this game, but overall, both provide very close to the same experience. If you spend some time truly evaluating what is going on while you are playing this, I think the Fury X provided a bit better of an experience, as it was not falling in and out of Reprojection during the first half of the game. The GTX 1060 was constantly bumping up close enough to or over our 11.1ms mark that Reprojection was being turned, then off, then on, then off, etc. This is noticeable while gaming in this title. This is the first game I have been sensitive to this in, but assuredly I did not like it. Again, not a "dealbreaker," but worthy of notation and still better than being in Reprojection "all the time."



Lastly we get to the "Radeon™ RX 480 set to drive premium VR experiences into the hands of millions of consumers..." The RX 480 "premium experience" showed us the worst GPU Render Times and the worst percentage of Reprojection (99.48%). I would suggest that if you are looking for a premium frame judder experience, the RX 480 fits the bill.



We talked some about Adaptive Quality on page 1, and it seems that a profile is used to determine when this dynamic scaling is implemented in Robot Repair. I saw no evidence of this framerate enhancing technique being used with the GPUs reviewed here today.

http://www.hardocp.com/article/2016...erformance_valves_robot_repair/6#.V7JzNq2XolA

 

[Bacon1]

Wake me when they are going to test some Vulkan / DX12 games

If that sounds like a great idea—and pretty much all the major developers agree that it is—the second way ATW helps is that rather than only warping old frames in cases where a new frame isn't ready, it can simply warp every single frame. So all the rendering takes place, and that can take somewhere around 8-9ms, and then right at the end before updating the display, you grab the telemetry data and do a time warp for the past 8ms. In many cases the resulting image will hardly change at all, but for fine-grained movement the perceived latency is now well under 10ms.

Now here's where AMD's async shaders come into play. ATW is a compute shader that can be scheduled right along with all the other graphics tasks, and there's no need to preempt other graphics processing to make it happen. Without async shaders, ATW needs to take priority over other work, which might mean a graphics context switch. If that happens at the appropriate time, all is well and ATW works as expected, but in a worst-case scenario the preemption takes too long and the window for warping and updating a frame is missed, leading to a repeated frame and some stuttering in the VR experience.

http://www.pcgamer.com/amd-liquidvr-vs-nvidia-vrworks-the-sdk-wars/

Also with Asynchronous Time Warp instead of rendering the frame right away, if it's under 11ms (for 90hz) it will try to update the frame based on any head movement / tracking that has occurred which is a very nice feature.

I find it ironic that [H] still is touting the "For premium VR" when AMD was clear what they meant by Premium VR wasn't full settings, but PC based VR, not Console / Phone VR.

Start ~4:30 in

 

[Phynaz]

Premium:
of exceptional quality or greater value than others of its kind; superior:

http://www.dictionary.com/browse/premium

I'm not seeing AMD meeting that definition when it comes to VR.

 

[exar333]

Wake me when they are going to test some Vulkan / DX12 games



http://www.pcgamer.com/amd-liquidvr-vs-nvidia-vrworks-the-sdk-wars/

Also with Asynchronous Time Warp instead of rendering the frame right away, if it's under 11ms (for 90hz) it will try to update the frame based on any head movement / tracking that has occurred which is a very nice feature.

I find it ironic that [H] still is touting the "For premium VR" when AMD was clear what they meant by Premium VR wasn't full settings, but PC based VR, not Console / Phone VR.

Start ~4:30 in

Let's not kid ourselves. It's entry-level but definitely not premium. Works fine though...

 

[Magee_MC]

Premium:
of exceptional quality or greater value than others of its kind; superior:

http://www.dictionary.com/browse/premium

I'm not seeing AMD meeting that definition when it comes to VR.

They weren't referencing Polaris being better than other GPUs for PC VR, but were saying that the 480 would allow users to access PC VR which is premium VR compared to other VR forms like Google Cardboard and the like using phones.

 

[Bacon1]

Let's not kid ourselves. It's entry-level but definitely not premium. Works fine though...

I already explained what they meant by premium and even listed a video. They weren't talking about all settings @ max PC gaming in VR with a $200 card. They explained that "Premium" meant PC VR, not Phone VR or "cheap" alternatives.

These are DX11 games and many of them Gameworks VR ones that [H] is testing. These are worst case for AMD, and just tech demos at that.

 

[Phynaz]

They weren't referencing Polaris being better than other GPUs for PC VR, but were saying that the 480 would allow users to access PC VR which is premium VR compared to other VR forms like Google Cardboard and the like using phones.

I'm not familiar woth any of the Phone VR implementations. Do they have issues with reprojection?

 

[pj-]

I'm not convinced the testing methodology is accurate because the Fury X is terrible. Even in Valve's simplistic source 2 based Robot Repair demo which is supposed to use adaptive quality to keep the framerate above 90, the Fury X is reprojecting almost half the time. If that's actually happening then AMD's current offerings are worthless for VR.

Wake me when they are going to test some Vulkan / DX12 games

How are they going to test VR games that don't exist?

Also with Asynchronous Time Warp instead of rendering the frame right away, if it's under 11ms (for 90hz) it will try to update the frame based on any head movement / tracking that has occurred which is a very nice feature.

No, that is synchronous timewarp. ATW is used when the frame isn't going to be ready in time. The current frame's rendering is interrupted on the GPU and the previous frame is updated with new headset info. Regardless, only Oculus Rift has that now, so it is irrelevant in Vive/SteamVR based testing.

These are DX11 games and many of them Gameworks VR ones that [H] is testing. These are worst case for AMD, and just tech demos at that.

Yeah, welcome to the current state of VR. I have 50ish VR games/demos and I don't think a single one is DX12. Almost everything is an indie "game" made with UE4 or Unity. Should nobody test anything at all so AMD doesn't look bad? Or should AMD update their marketing to be: "Premium VR.*

*In 2017 or 2018 when the majority of games are DX12 and there are new headsets available that require far more power than a 480"

 

[AtenRa]

Tooooo early for VR wars, we dont even have a working full PC game to test. Also, i dont expect anyone to release a DX-11 VR game.

 

[pj-]

Tooooo early for VR wars, we dont even have a working full PC game to test. Also, i dont expect anyone to release a DX-11 VR game.

"anyone"? There are hundreds of released games and I am not aware of one that is dx12. Even the oculus funded games from bigger names like Insomniac and Crytek are dx11. If you are talking about AAA stuff then it will be years before the market is large enough to support games with 8 figure budgets. Doom VR will be the first Vulkan VR 'game', but it's still unclear at this point how much gameplay it is going to have.

 

[AtenRa]

Yes im talking about a real PC game, not Asteroids on VR

 

[pj-]

There are some larger games with retrofitted VR such as Project Cars and Elite Dangerous. Fallout 4 will be another one. These types of things will likely be the biggest budget games with VR support for the next few years.

There are also exclusively VR games that I would consider full games, but not full length games, such as The Gallery. For benchmarking purposes the length of a game makes no difference.

Regardless of what you think the VR market lacks in terms of content, the performance of the games that do exist on hardware that also exists is important to the people who have or are planning to jump into VR.

 

[Bacon1]

Project Cars and Elite Dangerous.

And how does their performance in VR compare to their existing performance?

 

[Stormflux]

It's good that HardOCP are doing these VR tests, that no one else seems to want to do quite yet. Their results should kick AMD in the rear to start doing something visible to the public for their VR support.

The use of these early access games and smaller experiences are debatable. But good regardless to show the state. I for one would love to see these tests done for the games mentioned above. Elite: Dangerous, Dirt Rally, Assetto Corsa, Project Cars are all larger games that support VR and are relatively larger releases.

 

[Bacon1]

Sad that Kyle still doesn't understand what Premium VR meant. Maybe if he was invited to their Launch Event he would have understood.

 

[pj-]

He's clearly being a dick to rile up AMD fans.

It will be interesting to see how established games with VR retrofits perform, but the current crop of "cockpit" games don't really interest me a whole lot. They are the shallowest in terms of VR-ness, and can cause motion sickness in some people.

Adding VR to fallout and doom is much more interesting to me, and I hope it starts happening with strategy games. They are begging to be played on a virtual table top.

 

[Stormflux]

Is that HardOCP picture for real? Can you get any less professional? Vive/Rift (Premium) > GearVR (Casual) > Google Cardboard (Entry).

Cockpit games are probably the best use case for the current VR headsets right now unfortunately. But little is shown on this front. Room Tracking, and Standing experiences are largely in development and lacking, but are coming.

It will be interesting to see how they implement VR into Fallout and Doom, if they don't resort to a teleportation method of movement, cause if you think the motion sickness in cockpit games can be troublesome...

I still think the 2nd generation of headsets will be the real time for the masses to join in on the "premium" VR movement though. It's still really rough all around.

 

[pj-]

As immature as that picture is, I don't think it's entirely baseless.

If these results reflect real world performance I would not consider Fury X or 480 adequate for PC VR. Reprojection compromises the experience quite a lot. 45fps rendering is worse than what you get with gear VR or PSVR

 

[bystander36]

Is that HardOCP picture for real? Can you get any less professional? Vive/Rift (Premium) > GearVR (Casual) > Google Cardboard (Entry).

No, that is not real. That's Bacon1 doing his thing.

This is the real one:

 

[Bacon1]

No, that is not real. That's Bacon1 doing his thing.

No, this was posted by Kyle on his forums:

https://hardforum.com/threads/amd-n...robot-repair-h.1908149/page-2#post-1042483698

 

[bystander36]

No, this was posted by Kyle on his forums:

https://hardforum.com/threads/amd-n...robot-repair-h.1908149/page-2#post-1042483698

That was a joke, in response to someone complaining about the real one. That was not a real leader board.

 

[Bacon1]

That was a joke, in response to someone complaining about the real one. That was not a real leader board.

And? I was pointing out that he still doesn't understand what "Premium VR" meant. He created it, not me. You owe me an apology for the callout.

 

[bystander36]

And? I was pointing out that he still doesn't understand what "Premium VR" meant. He created it, not me. You owe me an apology for the callout.

I apologize for thinking you made it. It still is not real, and you clearly are trying to make people think it was.

 

[Bacon1]

It still is not real, and you clearly are trying to make people think it was.

How is it not real? I didn't modify it at all. If anyone thinks it's a real leaderboard then that is Kyle's fault and he posted it in relation to this topic on his forums. When you have the tester creating such immature content it really makes you question their objectivity while testing.

 

[bystander36]

How is it not real? I didn't modify it at all. If anyone thinks it's a real leaderboard then that is Kyle's fault and he posted it in relation to this topic on his forums. When you have the tester creating such immature content it really makes you question their objectivity while testing.

It's on a forum for starters, not on the leader board or in any review.