AMD and NVIDIA GPU Vive VR Performance in Raw Data @ [H]

[FrgMstr]

AMD and NVIDIA GPU Vive VR Performance in Raw Data - Both AMD and NVIDIA have had a lot to say about "VR" for a while now. VR is far from mainstream, but we are now seeing some games that are tremendously compelling to play, putting you in middle of the action. Raw Data is one of those, and it is also extremely GPU intensive. How do the newest GPUs stack up when it comes to Raw Data performance

 

[magoo]

very cool. thanks.

 

[JosiahBradley]

How can they not have mGPU working, it was supposed to be the perfect setup for VR, one GPU per eye. Both AMD and nVidia released full SDKs for this awhile back.

 

[FrgMstr]

How can they not have mGPU working, it was supposed to be the perfect setup for VR, one GPU per eye. Both AMD and nVidia released full SDKs for this awhile back.

Seems to be a little bit harder to get it working CORRECTLY in UE4 engine than expected from what I have been made to understand. Does not seem to be a game issue, which I guess is a good thing if you look at the big picture, since once they get it working on UE4, it will cover a LOT of current and upcoming games. I am told it is being worked on assuredly and is a priority.

 

[MotionBlur]

How can they not have mGPU working, it was supposed to be the perfect setup for VR, one GPU per eye. Both AMD and nVidia released full SDKs for this awhile back.

The initial work I've done on that with my 2 480s (1 for each eye) in some early custom DX12 engine tests it's been fairly painless, that could change as I add more and more modules (only got basic sound and input in conjunction with my graphics renderer right now). It's really just getting used to submitting work across what is available and architecting around that model verses relying on NVidia's SLI or AMD's CrossFire to do the splitting of work and controlling the race conditions.

 

[Quix]

I wonder if that simultaneous multi-projection feature that Nvidia has hyped to high heaven will make a difference once that starts to be implemented.

 

[trajan2448]

After all the blather about how "next generation" superiority favored AMD, once again turns out to be hot air.
with the RX 480 I did notice some texture flickering that was not present on the GTX 1060, oops.

 

[Sipster]

Nvidia's VR Funhouse is running on Unreal engine and it supports VR SLI. They suggest using 2 1080's for the best experience.

 

[FrgMstr]

Nvidia's VR Funhouse is running on Unreal engine and it supports VR SLI. They suggest using 2 1080's for the best experience.

Actually it seems to only support "video" SLI if you have three cards. If you have two, it dedicates one card to PhysX. If you have three cards then two cards get used for video, and one dedicated PhysX card.

NVIDIA VR Fun House Experience - The NVIDIA VR Fun House Experience

 

[Nenu]

Nice review, very interesting.

Also interesting is that the 10xx series of cards are not using Simultaneous Multi Projection yet.
The 1070 should have nailed the 980ti but is a tiny bit slower!
SMP was touted as a huge feature of pascal.

Are you aware of why this isnt in use yet?

 

[Ultima99]

Hmm, Pascal Titan X time.

 

[FrgMstr]

Nice review, very interesting.

Also interesting is that the 10xx series of cards are not using Simultaneous Multi Projection yet.
The 1070 should have nailed the 980ti but is a tiny bit slower!
SMP was touted as a huge feature of pascal.

Are you aware of why this isnt in use yet?

Short answer, I do not think it is fully baked currently. And that is just my personal opinion. Listening to the devs at Valve, they have bluntly said that SLI/CF is going to be the biggest way to bump perf right now and I think more resources are being pushed this way currently to get SLI working in UE4. Again, AMD is not talking to me on these things at all, so I can't really included CF in that statement.

 

[Shintai]

Lots of talk, only one delivers. R&D matters more than PR slide

 

[leppermessiah1]

Very interesting and informative! I would suggest not combining the metrics of dropped frame rate with reprojection rate, instead measuring dropped frame rate while reprojection is turned off, and measuring reprojection rate only when it is turned on.

 

[Ieldra]

Nice review, very interesting.

Also interesting is that the 10xx series of cards are not using Simultaneous Multi Projection yet.
The 1070 should have nailed the 980ti but is a tiny bit slower!
SMP was touted as a huge feature of pascal.

Are you aware of why this isnt in use yet?

Not yet supported!

I wonder if that simultaneous multi-projection feature that Nvidia has hyped to high heaven will make a difference once that starts to be implemented.

It's going to make a big difference.

 

[Activate: AMD]

Interesting review. I can definitely see that its going to take some time to get all the tools and figure out the best way to handle VR benchmarking, but I think you did a pretty solid job to start. I have to say, I'm a bit surprised that AMD didn't do better, especially compared to the 1060. It is only a single data point, and as things develop we may see the tables turn. That said, I would maybe have guessed that the 480 would beat the 1060 and the Fury X would have done substantially better. Considering all the "VR Ready" stuff coming out of AMD these days, this is underwhelming. Then again, who out there has the disposable income for a Vive then pairs it with a $200 GPU outside of somebody who is stretching their budget to the absolute limit?

 

[tybert7]

Used unreal engine, no wonder it does so poorly on amd. What's next, project cars VR? Get back to me when one of these vr titles uses an engine that's actually designed to allow amd cards to perform.

Fury X below a 1060 = totally not optimized to run well on amd.

 

[Dunnlang]

Kyle, did you guys try plotting the dropped frames as a rolling sum? You could tweak the sum range to dial it in to something that looks presentable. The sum could be represented, on a second y-axis, as a percentage of 90 or 45 fps or by raw number. You would still have similar data display issues, but maybe this display method would help highlight sections where the frame drops build up and other areas where it alleviates. Maybe one or two dropped frames aren't bad, but a section with many is worse (I am only guessing, not from personal experience). The rolling sum could help highlight the volume of dropped frames better than the small lines.

 

[FrgMstr]

Used unreal engine, no wonder it does so poorly on amd. What's next, project cars VR? Get back to me when one of these vr titles uses an engine that's actually designed to allow amd cards to perform.

Fury X below a 1060 = totally not optimized to run well on amd.

What titles do you suggest testing that are extremely GPU intensive. UE4 is holding a LARGE share of games right now, so not running well on UE4 is sort of a copout.

 

[ebduncan]

Well that is some clear bias in hardware performance. I didn't know UE4 was so favored towards Nvidia performance.

1060 beating the Fury X? lol , that should never happen.

oh well, hopefully as VR matures things will perform more equally like other games.

Thanks for the review guys, i've yet to dive into VR.

 

[FrgMstr]

tybert7 - Serious question here. What games do you suggest I use? You are pointing out that we are using the wrong game, so I am asking what can I do to make this right. I am very sincere in this question.

I basically picked this game in a "blind" test after going through and find around 15 VR titles that were GPU intensive. I basically did no research on the game till I picked it out using a GTX 1080 as my test GPU.

 

[Trepidati0n]

Go ahead and keep testing, I just won't put much stock in the results until the games/engines start being designed to run properly on amd parts and make effective use of dx12/vulkan. That might be a long time if devs want to just release dx11 games in vr.

The not much stock comment is a bit disingenuous. Kyle gave you the data..flat out, as clean as it can be, based upon actually game play. The data is GENUINE and worth taking stock of. Anybody who buys a piece of HW based upon the #'s on the box is ignorant of the fact that a graphic cards in, of itself, needs support from several other pieces (process, OS, drivers, etc).

 

[Activate: AMD]

Go ahead and keep testing, I just won't put much stock in the results until the games/engines start being designed to run properly on amd parts and make effective use of dx12/vulkan. That might be a long time if devs want to just release dx11 games in vr.

But if they are going to stick to an api designed and released from 2011-2013 and expect me to look at the performance results as something representative of a properly coded game, they failed with me.


We have all seen what properly coded games/engines run compared to the stuff before, if you want to run a shooter on a dated api with no optimizations for specific hardware, you have lost the right to be taken seriously as far as performance is concerned in my eyes.

It's like winning a race against cripples. I want to see the results of a race when the runners are not hobbled, I want to see what performs better when all the runners are fed and ready to run at more or less their best. That is not dx11. If you want to see who wins the race filled with cripples, have at it.

Confusing perspective right there. If games running these engines are representative of the market, choosing to ignore them is just absurd. Do you also not pay attention to benchmarks of games that don't use TrueAudio or other equally irrelevant features? If the tools don't exist, it can't be tested. Should the blame fall on the reviewer for not having games that look good and are "fair" to both companies? If there is nothing available that shows one company's "strength" is it actually a strength? These are serious questions by the way, because as it currently stands you just sound like you'd prefer to be willfully ignorant of actual, real results because they don't fit your little narrative.

 

[ChosenUno]

I can't wait until SMP support is enabled, and even the lowly 1050 might get to beat Fury X in VR games. What then? Cry GameWorks?

Seriously, cry foul just because your favourite irrelevant hardware company is doing crap in VR. Guess who invested more in VR tech until now? Guess who's reaping the benefits now?

Nvidia right now is miles ahead in VR tech (performance as well as immersion) as well as marketshare, and so devs will follow suit. Since the cost just to get started in proper VR is so high ($800 for HTC Vive), it's squarely in the enthusiast category. And there's no competition there.

Maybe AMD should've spent more time with the Unreal guys as opposed to the Frostbite guys huh.

Great review Kyle, keep it up. You're reporting the reality of today, not the delusions of the future, and I for one applaud that.

 

[FrgMstr]

Before you blame the Vive's screen to much for VR's Aliasing issues, take a look at the lack of forward rendering/MSAA in the Unreal engine. Plenty of games look great in the Vive, such as Valve's Lab pack in title. The good news is that Forward Rendering is in the pipeline for Unreal. Thanks!

Just to be clear, I blamed nothing on anyone or anything thing. I am very aware of Valve's Lab Pack. We may use it in the future as well, but what I did not find it indicative of, was actual gaming performance. I think it is on the same plane as NVIDIA's VR Fun House. It is a tech demo.....a very impressive tech demo.....but I did not find Lab or Fun House to be representative of games that I have purchased overall.

 

[Stoly]

So what happened to Simultaneous Multi-Projection? shouldn't it give a Huge performance boost? IIRC nvidia claimed twice the performance of Maxwell in VR.

 

[FrgMstr]

So what happened to Simultaneous Multi-Projection? shouldn't it give a Huge performance boost? IIRC nvidia claimed twice the performance of Maxwell in VR.

It is not a supported technology yet. They are still working on it the driver side and implementation side of that.

 

[FrgMstr]

So does it require both dev support and driver support? I though it only needed to be implemented in the driver.

I am not sure, but I would guess both.

 

[CheeseCake87]

I'm not sure you should of done this review at this time to be honest. Hi, by the way, just joined, great to be here. My opinion; Not that it matters, is that I'm greatful for the work you have done here and yeah... good job; but, to early for this stuff, to be honest. No offence... You have taken an early access game (granted nothing is really out yet) and benched it. Not to mention its a GameWorks title. A closed source addition that none of us are allowed to poke holes in, and with nVidias track record... weeelll, you can kinda see where this is going.

I noticed a comment above that was saying AMD should just die bla bla yada. Lets not get into that much; horrid if you ask me. Fact is, companies have no soul. No feelings. No interest in you. They are only interested in the money. Some donate their work to the open source to progess technology and others lie about specs and pay reviewers gospel funds. It is very much a dog eat dog market (if dogs actually ate dogs)

Fact is folks, speculation is only that; Speculation. This review was done way to early. VR early access game with closed source GameWorks, in a market that is still very young. Not even SMP is out yet, and you done the review anyway lol Tame your fleshy flag poll for VR for now. - As the review stands just now, written in a rather bias view, just saying lol - Not that I care. I recommend parts towards clients set budgets; just last week I was plugging in a STRIX 1080 - but if someone wanted a budget system. It has to be AMD. The customer always comes first, your bias should never get in the way of whats best (just a little lesson for the AMD hater guy xD - LOVE YOU!) - k bye!

 

[FrgMstr]

Fact is folks, speculation is only that; Speculation.

No, this is not speculation or "bias," this is fact. Probably the first time you have ever seen any VR performance backed up with objective data. It is obvious from the totality of your post that you did not fully read the review as well.

 

[Cobalt35]

Thanks for the article, very informative

I love how AMD marketed the shit out of the 480 as a card that was designed for VR. Once again (again, again, again, again, again, again, again, again, again for AMD marketing) the actual unbiased numbers, as tested and portrayed by an unbiased site, put the lie to the marketing spin.

nVidia don't even have to do anything - AMD just keeps on shotgunning themselves in the face and then sitting back and wondering where the pain is coming from.

 

[Trepidati0n]

The customer always comes first, your bias should never get in the way of whats best (just a little lesson for the AMD hater guy xD - LOVE YOU!) - k bye!

Calling Kyle biased is like saying water is dry. Kyle has been posting hard data for over a decade and it has NEVER been biased. You may read it as biased since it doesn't support your viewpoint...but then that would make you biased and thus your post biased? -k Bye...

 

[Remon]

Thanks for the article, very informative

I love how AMD marketed the shit out of the 480 as a card that was designed for VR. Once again (again, again, again, again, again, again, again, again, again for AMD marketing) the actual unbiased numbers, as tested and portrayed by an unbiased site, put the lie to the marketing spin.

nVidia don't even have to do anything - AMD just keeps on shotgunning themselves in the face and then sitting back and wondering where the pain is coming from.

In a Nvidia biased engine... Lol, unbiased.

 

[Cobalt35]

In a Nvidia biased engine... Lol, unbiased.

If you think the results for the 480 were anything other than bad you need to go re-read the article. Or read it properly for the first time. As Kyle stated, UE4 is one of the most commonly used engines today. Why should they not use it?

 

[RAutrey]

It's a tad bit early to break out the fucking pitchforks... The VR review process is in its infancy.

 

[Remon]

If you think the results for the 480 were anything other than bad you need to go re-read the article. Or read it properly for the first time. As Kyle stated, UE4 is one of the most commonly used engines today. Why should they not use it?

And that somehow makes it unbiased. Nice logic there.

 

[FrgMstr]

It is too early to test though. Really, come on... If someone is buying a $700+ card, they are buying to keep it for a while. An investment. It is a little early to be pushing what is best for VR at this moment.

It is a real game that real people are playing and I would suggest that many buying new GPUs now will like the information we are sharing. Your thoughts are noted.

 

[FrgMstr]

And that somehow makes it unbiased. Nice logic there.

Yes, testing a game using a widely adopted VR engine is absolutely unbiased. I keep asking, but getting now answers....so how about now? What VR game that is GPU intensive would you like to see reviewed? I am all ears. I bought 5 new games this weekend that I need to dig into now at as well. Next up will likely be Elite Dangerous.

 

[CheeseCake87]

Calling Kyle biased is like saying water is dry. Kyle has been posting hard data for over a decade and it has NEVER been biased. You may read it as biased since it doesn't support your viewpoint...but then that would make you biased and thus your post biased? -k Bye...

That wasn't aimed at Kyle.

We all know that GameWorks favours nVidia, that makes them naturally bias. I am not bias myself, I just care about - whats best for what - at what time, if that makes sense

It is to early to test, especially on an early GameWorks title. Just saying.

 

[FrgMstr]

I hear (read) you Kyle. I just feel VR is still premium right now, the review you have is fine for people with the enthusiast dollar... Again, kudos for the data.

Hehe, another point that I directly addressed in the review that agrees with your comments.