SteamVR Performance Test grades PCs on their VR readiness

After HTC's price and pre-order announcement for its Vive VR headset this weekend, Valve has taken the wraps off its SteamVR Performance Test, a simple tool that lets aspiring VR junkies determine what kind of experience to expect from their PC. This test is free, and it doesn't require a VR headset to be hooked up to the host PC to run.

After running a truncated version of Valve's Aperture Science Robot Repair demo for the Vive, the benchmark tells us how many frames are tested, how many frames were delivered "below 90 FPS"—or, we're guessing, how many frames took longer than 11.1 ms to render—and how many frames were CPU-bound.

The test also assigns a performance index to the host system. My Core i5-4690K-powered, GeForce GTX 980 Ti-equipped main desktop scores a 10.9, for example. Swapping in the GeForce GTX 970 from our recent Breadbox build drops that number to a 6. From what we can tell, both tests delivered all of their frames above 90 FPS, and only a couple of frames were CPU-bound. 

AMD brought this test to our attention, and the company actually sheds a bit more light on what this benchmark does than Valve's official release notes do. According to AMD, this test uses a technique called dynamic fidelity to adjust image quality to the level needed to maintain a high enough frame rate for a good VR experience.

Going by that description, the rather vague graph in Valve's report seems to tell us how much time cards can spend at various image quality levels. Unsurprisingly, then, the GeForce GTX 980 Ti's flat line at "Very High" means it didn't have to adjust image quality at all, while the GeForce GTX 970's more variable graph might indicate that more on-the-fly graphics settings changes were happening behind the scenes to keep frame times down.

Source: AMD

For its part, AMD is pleased about the results its products are delivering with this test. The company has already benched a variety of its graphics cards against the Nvidia competition, using a PC with an Intel Core i7-6700K, 8GB of DDR4-2666 RAM, a Z170 motherboard, Radeon Software 16.1.1 drivers for the Radeons, GeForce driver version 361.91, and Windows 10 64-bit.

All of these results are gleaned from a system with a single graphics card save for one: the dual Radeon R9 Nano configuration you see up top. AMD says those results are the product of an early version of its Affinity Multi-GPU technology, part of the company's LiquidVR SDK. Affinity Multi-GPU lets developers break up the work of rendering a VR scene across multiple graphics cards, and the cards then work together to composite that work together for the final scene. The company warns that its work with Valve on incorporating Affinity Multi-GPU into this demo is far from finished, but it's pleased with the "significant" scaling that Affinity Multi-GPU already exhibits over a single card. 

While we're extremely skeptical of the usefulness of an index like this for comparing graphics card performance, the numbers AMD provided do give us a rough picture of how the current players graphics card market might stack up to the challenge of the Vive. Without knowing the exact models of cards that AMD tested, though, it's hard to say much about how well the results of this test map to the real-world options graphics card buyers can choose from.

As expected, though, anybody with a Radeon R9 390, GeForce GTX 970, or better card should be able to enjoy a decent VR experience with the Vive, assuming all titles will target tradeoffs between performance and visual fidelity similar to those that Valve is making in its benchmark.

Comments closed
    • LightenUpGuys
    • 4 years ago

    i5-2320, 16GB DDR3 1333, 780 GHz Edition. 5 so im almost in the green. Not bad for a 2012 PC and a 2013 GPU.

    • Voldenuit
    • 4 years ago

    GTX 970 in my system running @ 1429 MHz: 7.7

    [url<]http://tinypic.com/view.php?pic=5druw5&s=9[/url<] EDIT: PS According to Afterburner, my GPU was rendering at a flat 120 fps throughout the test, so it looks as if there is some IQ headroom if Valve were aiming for 90 fps.

    • Chrispy_
    • 4 years ago

    Today’s work PC got a fail.

    It run the test and it looked like a fluid framerate to me, but apparently the Kepler-based GT640 is not enough for it.

    • southrncomfortjm
    • 4 years ago

    I’ll pass.

    • Wirko
    • 4 years ago

    [quote<]AMD brought this test to our attention, [/quote<] WAS that SOmeoNe that we know?

    • Anovoca
    • 4 years ago

    I love how AMD threw that 2x R9 on that benchmark but no sli cards for comparison and then for purely scientific and unbiased reasons placed it as a winner in its tie with the 980ti in their charts.

      • ImSpartacus
      • 4 years ago

      The 2xNano is probably the new dual fiji “Gemini” card (or it’s really 2xNano, but they wanted to basically match Gemini).

      Recent rumors have pegged Gemini at 375W (the official max for the pcie spec, i.e. 2×8-pin). Since the Nano’s official tdp is 175W, that means Gemini could conveniently be 2xFiji with Nano clocks.

      Note that the Fury X uses 250W, so it would take a 500W Gemini to have dual Fury X performance. The 295X2 used 500W, but it was considered kinda sketchy because it was intentionally breaking the pcie spec (but most power supplies can still work with it).

      So yeah, that’s the probable reason for why the only dual gpu setup is 2xNano when you’d almost never use dual Nanos in a real build.

      Since amd has been pushing Gemini as a vr solution since day 1 (dual gpus are good for vr, one gpu per eye, see amd quantum). And in particular, amd has worked with valve and vive, so I wouldn’t be surprised at all if valve has early Gemini boards (or at least knowledge of Gemini’s “dual Nano”c configuration).

    • madmanmarz
    • 4 years ago

    [s<]8.9[/s<] [s<]9.1, bring it![/s<] You still haven't seen my true power level [s<]9.6 with some overclocking[/s<] 9.8!! 6600k @ 4.7Ghz 2x4GB DDR3 Fully unlocked R9 290(x) [s<][url<]http://i.imgur.com/aqRioWe.png[/url<][/s<] [url<]http://i.imgur.com/UyqUOEP.png[/url<]

    • Deadsalt
    • 4 years ago

    I pulled a 1.

    i74790K@stock(4.0GHz)
    32GB of DDR3 1600 Mhz
    GTX670 OCed to 1020Mhz. 1098Mhz @ Boost. 1502Mhz VRAM.

      • Vaughn
      • 4 years ago

      I got 2.3

      i7 970 @ 4.2Ghz
      12GB of DDR3 1600 Mhz Cas 7
      7970Ghz 1050/1500

    • brucethemoose
    • 4 years ago

    Here’s a fun game:

    How long until Intel has a “VR Ready” IGP?

      • travbrad
      • 4 years ago

      How long is a piece of string?

      • christos_thski
      • 4 years ago

      Five chipsets and bull***t marketing about “integrated graphics finally catching up” onwards. [and it will still suck for games of that period].

      But it will work for 2016 games. 😛

      • Andrew Lauritzen
      • 4 years ago

      There’s no need for that to be IHV-specific. How long until we see something “VR read” (~GTX 970 performance) in 15W-25W? Do the process math – it’s gonna be at least a few years still 🙂

    • moose17145
    • 4 years ago

    Huh… my system pulled a 6.7… actually a lot better than I was expecting giving the mish mash of old and new parts. I was expecting something in the like… 2ish range…

    i7-920 OCed to 3.2GHz (tends to stay turboed around 3.36Ghz from what I have observed)
    24GB DDR3 1600MHz Ram
    R9-290
    And a 500GB Samsung EVO 850

    • RtFusion
    • 4 years ago

    Sweet got a 10.2.

    Win 7 64-bit SP1
    nVidia Asus GTX 980 STRIX OCed 1500 MHz core, 8000 MHz VRAM
    Intel 3770K at 4.5GHz
    G-Skill 32GB DDR3 8×4 F3-12800CL10-8GBXL

    • Generic
    • 4 years ago

    I am shocked! Shocked, I say, that my GTX 460 [i<]isn't[/i<] up to the task. Perhaps I'll upgrade when xPoint is widely available in Knight's [place] consumer video cards...

      • USAFTW
      • 4 years ago

      You’re better off… My GTS 450 is handicapped by the capitalist pigs at Nvidia.
      …anyone?

    • travbrad
    • 4 years ago

    [quote<]The company has already benched a variety of its graphics cards against the Nvidia competition, using a PC with an Intel Core i7-6700K, 8GB of DDR4-2666 RAM, a Z170 motherboard, Radeon Software 16.1.1 drivers for the Radeons, GeForce driver version 361.91, and Windows 10 64-bit.[/quote<] It's a bad sign when even AMD is buying Intel CPUs for their systems.

      • Mikael33
      • 4 years ago

      Their GPU test configs have used them for some time, for obvious reasons.

    • jessterman21
    • 4 years ago

    My OC GTX 960 gave me 4.3; upper-end of VR-Capable 🙂

    [url<]https://a.disquscdn.com/uploads/mediaembed/images/3246/3340/original.jpg[/url<]

    • christos_thski
    • 4 years ago

    By the way, exaggerated CPU recommendations had me worried, but the test seems to find my Core i5 3470 adequate for VR. My Radeon 7870, on the other hand, was described as “VR incapable” even at low fidelity. Score…. 0,2. It’s as if I’m sporting the integrated gpu-equivalent rig of VR. 😛

    VR is really demanding, isn’t it?

      • DPete27
      • 4 years ago

      Two 1200×1200 screens, so more demanding than 1440p but not as bad as 4k.

        • Flapdrol
        • 4 years ago

        Why is 2400×1200 more demanding than 2560×1440?

          • RoxasForTheWin
          • 4 years ago

          I would assume it would be based on the need for constant 90 fps, although it is true the raw pixel count is lower than 1440p

            • moose17145
            • 4 years ago

            Also the fact that you are actually rendering the scene twice from two different angles… That plays a big part of it as well

          • aggies11
          • 4 years ago

          Both Rift and Vive render at a higher internal target resolution than the displays for various reasons. I think the Vive is around 1.4x and the Rift is 1.7x (times the total number of pixels).

          Plus as others have pointed out, the added requirement of 90fps means you are pushing a hefty amount of pixels.

          If you throw all the math together that’s roughly the same as driving a 4k display at ~50fps.

      • Flapdrol
      • 4 years ago

      Average fidelity score is 0.9 on my 670, which shouldn’t be that much faster. Scores seem messed up for anything below recommended.

      • djayjp
      • 4 years ago

      I would blame bad drivers/optimization. GPU makers have no incentive to optimize old cards…. For me, I strongly suspect that the erratic nature of PC hardware will= motion sickness for many. PSVR here I come.

    • anotherengineer
    • 4 years ago

    Was going to run it on my 955BE and HD 6850 for craps & giggles, but at almost 5GB, I’ll have to pass.

      • DancinJack
      • 4 years ago

      huh? Mine was ~1.9GB.

      • Topinio
      • 4 years ago

      It says 4.7 GB on the store page, but the download is “only” 1.9 GB

    • BobbinThreadbare
    • 4 years ago

    Valve really should have done something like this years ago. Still, it would be nice to have a standardize single number that people can refer to for how game worthy their systems are. It should make non-enthusiasts much more comfortable with joining in (and restrict crappy salesmanship like Geforce 710s with 2 GB of ram!).

      • christos_thski
      • 4 years ago

      That’s a great point. The only sort-of kinda universal metric for PC performance has been the Windows 7 Experience Index (unavailable in Win10 and win8, if I remember correctly), and that one left A LOT to be desired (in particular its graphics benchmark).

      Valve should extend this to a more general PC performance metric index, for parallel use with traditional requirements and specs.

        • nanoflower
        • 4 years ago

        It’s not enabled by default in Win 10 but apparently is still there. I used a tool to turn on much of the telemetry that Win10 has and it also allowed me to enable the Windows Experience feature. I think it was the Ultimate Windows Tweaker 4.0.1.0 that I used to do that.

          • biffzinker
          • 4 years ago

          Displays the performance index score on Windows 8.1/10.
          [url=http://www.softwareok.com/?seite=Microsoft/ExperienceIndexOK<]ExperienceIndexOK[/url<] [url=https://www.virustotal.com/en/url/969045110f77e36114b95f2f85ce7e165ab509f2313c89452a367b7801e51dca/analysis/<]VirusTotal Scan Report[/url<]

        • BobbinThreadbare
        • 4 years ago

        The Windows experience score was a decent idea, but I don’t think MS knew what to test. I also don’t think they pushed retailers hard enough to provide the scores upfront.

      • slowriot
      • 4 years ago

      Eh….

      I think this is really dangerous. There’s no way to come up with a score that usefully reflects the various games out there, which vary widely in bottlenecks, resource usages, etc. Plus what use would this score be for people who are trying to decide on a new PC or build? Or when selecting say whether to upgrade your GPU or CPU? And which one to choose from? I don’t think it would do anything to help stop users from buying cheap, crap hardware. Worse, it could lead to these companies gaming the benchmark (which we’ve seen before).

        • BobbinThreadbare
        • 4 years ago

        [quote<]There's no way to come up with a score that usefully reflects the various games out there, which vary widely in bottlenecks, resource usages, etc.[/quote<] Actually there is, games do not vary *that* much from one another, and the point isn't to tell you which component of the system is holding you back (or is being held back), but just how the overall performance of the system is. It's not for people who feel comfortable upgrading components, but for people who buy pre-built systems. [quote<] Worse, it could lead to these companies gaming the benchmark (which we've seen before).[/quote<] How is this worse? Right now your average non-enthusiast consumer has literally 0 idea how a system will perform. Even if every hardware manufacturer started cheating on a test, we'd be exactly where we started.

          • slowriot
          • 4 years ago

          [quote<]Actually there is, games do not vary *that* much from one another, and the point isn't to tell you which component of the system is holding you back (or is being held back), but just how the overall performance of the system is.[/quote<] No, there's not. Hence why no one has ever come even close to a useful universal score. Do you think 3DMark and the like haven't been trying for decades? Why would you even care about this Steam news if companies hadn't failed numerous times at this goal? Oh that's right, because all these attempts at a comprehensive single score number have failed. Even ones from names like Microsoft. I sincerely doubt (in fact, I know) Valve hasn't cracked this impossible task. Games do vary quite widely. It's comical to me to try to deny this. I've seen far too many times people are convinced they barely need a CPU and then oh crap in a year they're seeing a real crap frame rate floor and wondering why that is... Currently this score is useful to no one. It's not advertised with a new computer at Best Buy or on Dell or HPs website. And frankly I sincerely doubt these companies have any interest in doing so. [quote<]How is this worse? Right now your average non-enthusiast consumer has literally 0 idea how a system will perform. Even if every hardware manufacturer started cheating on a test, we'd be exactly where we started.[/quote<] There is a cost to deceiving users. I think we're barking up the wrong try by trying to convince the average Joe to care. Instead, I think we would have far greater success in demanding more out of the media that covers these products and companies, who in turn would provide simpler, straightforward advice to consumers and also force OEMs into building more balanced configurations.

        • Bensam123
        • 4 years ago

        Sure you can… that’s why you have stats. Steam iis a giant data mining machine.

      • Bensam123
      • 4 years ago

      Described this thing exact thing over multiple years here, especially when they were talking about their defunt Steambox. They have the stats and the system available to do this.

    • UberGerbil
    • 4 years ago

    I want to try this on my Sandy Bridge with its IGP. How low do the numbers go?

      • morphine
      • 4 years ago

      -MAXINT

        • brucethemoose
        • 4 years ago

        ERROR: UNDERFLOW

    • chuckula
    • 4 years ago

    IT GOES TO 11.

      • DPete27
      • 4 years ago

      [url<]http://media.giphy.com/media/aqSl7Dw5HTojK/giphy.gif[/url<]

      • ImSpartacus
      • 4 years ago

      I also thought that was clever, lol.

Pin It on Pinterest

Share This