SteamVR Performance Test: 16 GPUs Compared

Last week, Valve released its SteamVR Performance Test. We just finished testing 16 GPUs in order to determine what you really need for a VR-ready gaming PC.

Two weeks ago, in preparation for HTC Vive pre-orders, Valve released its SteamVR Performance Test. This free utility is meant to evaluate whether your computer is ready to support VR content in its current form. And if it's not, the test will suggest your best upgrade path.

We spent the last week running this utility on a PC with specifications similar to the requirements of both HTC's Vive and Oculus' Rift VR HMD. The test system is equipped with a Core i5-4670K running at its stock 3.5GHz, 8GB of DDR3-1600 memory and separate SSDs for the OS and game installs. In other words, we tested using a fairly modest system (GPUs aside), rather than trying to bowl you over with high-end specs.

We tested the PC with 16 different graphics card configurations, some from AMD and some from Nvidia. Most of the GPUs meet the requirements for VR. But we also tested a few extra boards that don't satisfy the lowest recommendation. Like you, we're wondering whether there's any wiggle room when it comes to an enjoyable experience.


Test System
CPUIntel Core i5 4670K @ 3.5GHz
MotherboardAsus Z87 WS
Memory8GB Corsair Vengeance Pro DDR3-1600
StorageSanDisk Ultra 128GB x 2
PSUCorsair RM850 850W
OSWindows 10 Pro

Test Method

The two-minute SteamVR Performance Test runs several different scenes from the Aperture Science Robot Repair demo, measuring how many frames are rendered, how many frames drop below 90 FPS, average fidelity and if a frame's output is limited by the CPU. Once the benchmark completes, it spits out a somewhat vague "Not Ready," "Capable" or "Ready," with suggestions about how to improve performance.

We took the test one step further by running FRAPS and recording the minimum, maximum and average frame rates, as well as collecting frame time data. 

The SteamVR Performance Test takes a different approach than most of the benchmarks you see on Tom's Hardware. Rather than presenting the same sequence with the same detail settings, Valve's metric adjusts the image quality (fidelity) in an effort to maintain at least 90 FPS. Your score is determined by the average fidelity your hardware maintains, combined with the average frame rate (although Valve hasn't shared the complete equation used to generate the result).

When Valve and HTC finally revealed the price and launch information for their consumer Vive VR system, the two companies also disclosed hardware recommendations. But they purposely avoided defining minimum requirements.

MORE: Virtual Reality Basics
MORE: AMD's LiquidVR Puts Processing Muscle Behind Virtual Reality
MORE: Nvidia GameWorks VR Multi-Res Shading And Other Parlor Tricks

To use an HTC Vive, Valve and HTC recommend that your computer be equipped with an Intel Core i5-4590 paired with either an Nvidia GeForce GTX 970 or AMD Radeon R9 290 graphics card. Those are steep recommendations, and most PC gamers simply don't have that much graphics horsepower under their hoods. No doubt many enthusiasts are hoping to get by without upgrading and still enjoy the VR experience. Unfortunately, our results show that you probably shouldn't bother.

Lower Than Recommended

Valve, HTC and even Oculus recommend high-end graphics hardware, so it isn't a surprise that the GeForce GTX 950 wasn't up to par. We didn't expect it to return a "capable" result, though. With a score of 2.3, the GM206-based board suffered low fidelity in exchange for a high-enough frame rate.

The only real reason we included the 950 was to harness two of them together in SLI. A pair of 950s offers similar gaming performance as a GeForce GTX 970, so we were hoping for more from this configuration. But SLI doesn't work the same in VR and Nvidia hasn't perfected its mechanism for rendering individual screens with separate GPUs yet. As a result, our system's score dropped to 1.2, yielding a "not ready" rating.

The GTX 960, also based on GM206, is marginally better. A single Zotac GTX 960 Amp! Edition claws its way up to 3.3, enough to achieve medium fidelity. Add a second board, however, and watch your score fall to 2.2. And because nearly 25 percent of the time is spent under 90 FPS, Valve's tool calls this dual-GPU setup "not ready."

AMD's mid-range GPUs fare somewhat better than Nvidia's, especially in multi-GPU configurations. XFX's R9 380 4GB manages medium fidelity and kicks back a score of 3.3, which the SteamVR Performance Test deems "capable." We added a second R9 380 (a 2GB model) and ran both cards in CrossFire. The score jumped slightly to 3.7, still in the medium fidelity range. Again, this configuration was rated "capable."

Next, we tried our luck with PowerColor's R9 380X Myst Edition, one step down from AMD's VR-ready Radeon R9 390. The Myst Edition card is overclocked though, so if any card under the recommended cut-off can get a toe in the door, it'd be this one. And while it does post the best result we've seen so far, the R9 380X is still only ranked "capable," driving average fidelity with a score of 4.6.

Capable, But Not Really 'Ready'

If you own a GeForce GTX 960, a Radeon R9 380X or even a pair of R9 380s, you might be thinking to yourself, "Excellent! I can run VR games on my system," and you'd be forgiven for thinking that. But don't get too excited yet. There's a big difference between VR-ready and VR-capable.

A VR-ready system meets the specifications set forth by HTC and Valve for developers to target when they build VR games. Incidentally, Oculus is asking for the same level of hardware for its VR system. Developers are free to optimize so that their games work on lower-end hardware, but most will likely put more effort into making games run smoothly on a GTX 970 or R9 290. Just because the SteamVR Performance Test says your system is "capable" doesn't mean you'll be able to run all the games, nor does it mean you'll enjoy the experience.

Here's what the Results page states for a system with a "capable" rating: 

"Your system is capable of rendering medium (or low, depending on the GPU) fidelity VR but does not meet the recommended spec! Some VR titles will have been tuned to display only at recommended visual fidelity and may not run on this system. Others may opt to display with reduced visual fidelity to support smooth framerates. Follow the recommended and minimum specs for each VR title."

That doesn't sound like the ideal situation to be in after spending $600-$800 on a brand new VR HMD. It'd be better to have a VR-ready machine if you really want to enjoy the experience this year.

Create a new thread in the Reviews comments forum about this subject
This thread is closed for comments
83 comments
    Your comment
    Top Comments
  • Sergio_9
    The score for the 295x2 is wrong, there is a manual setting for in SteamVR that enables proper Crossfire use, my 295x2 with a 2600K @ 4.8ghz gets a score of 9.7, after changing the setting, before making the change I get the score listed above.
    13
  • Other Comments
  • dark_lord69
    Thanks, these are performance benchmarks I wanted to see!
    It answers several questions I had...
    4
  • TechyInAZ
    Good read.

    With how expensive VR headsets are right now, I don't understand why you would pair it with such a budget friendly GPU in the first place. If you can pay $500 for a VR headset, you should be able to buy at least a GTX 970 too.
    4
  • turkey3_scratch
    You could buy well more than a GTX 970. For a true VR experience, for the most immersive, I think you will want a GTX 980 minimum, preferably an R9 Nano to 980Ti. Any type of slight frame drop is a major loss of immersion.
    0
  • TechyInAZ
    Anonymous said:
    You could buy well more than a GTX 970. For a true VR experience, for the most immersive, I think you will want a GTX 980 minimum, preferably an R9 Nano to 980Ti. Any type of slight frame drop is a major loss of immersion.


    I would probably favor the Fury Nano or Fury over a 980 since VR headsets have very high res displays.
    0
  • turkey3_scratch
    Yep. 980 is such a bad value IMO. The 390X value isn't too good either. The R9 Nano is the only card above the GTX 970 and R9 390 that seems to have good value IMO.
    -1
  • moogleslam
    Because I have an i7-2600k (running at 4.8Ghz), I apparently don't meet the minimum requirements for Oculus Rift. Really?
    2
  • lun471k
    Quote:
    Because I have an i7-2600k (running at 4.8Ghz), I apparently don't meet the minimum requirements for Oculus Rift. Really?

    They don't really bother checking your CPU. If it's older gen, they automatically assume it sucks. funny thing is that I also run a 2600k @ 4.8Ghz and even systemrequirementslab says my CPU can't run latest games (not in VR). Yet I play everything on Ultra...
    4
  • turkey3_scratch
    System requirements are subjective.
    0
  • rainyfeels
    Is some of this based off architecture? My 780 is superclocked and can easily compete with a 970 yet only gets 4.5 in the test and is "capable". Or will it be fine when it actually comes to playing games?
    3
  • kcarbotte
    Anonymous said:
    Thanks, these are performance benchmarks I wanted to see!
    It answers several questions I had...


    I'm glad you liked it.
    It was quite an effort to do, but I felt it was info that people would want to know.
    1
  • picture_perfect
    AMD./Nvidia should offer VR-ready cards at a discount to get people started. VR-only so as not to affect normal sales. It's definitely in their interest. Or they could apply for government subsidies considering VR's potential public services. I'd vote for it :)
    0
  • kcarbotte
    Anonymous said:
    Is some of this based off architecture? My 780 is superclocked and can easily compete with a 970 yet only gets 4.5 in the test and is "capable". Or will it be fine when it actually comes to playing games?


    That's a great question.
    I didn't have a GTX 780 to test, so I couldn't test that out.
    I was very curious myself about the score, but your CPU will play a roll in the overall score. If you're running an AMD CPU it may not have the single thread computation power to keep up.

    It's also importatnt to keep in mind that your GTX 780, while comparable to a 970, does not actually keep up with a 970 in all tests.
    According to Nvidia's own graph, even the Titan is bested by the 970 in 3DMark FireStrike.
    http://international.download.nvidia.com/webassets/en_US/shared/images/products/shared/lineup-full.png
    1
  • kcarbotte
    Anonymous said:
    AMD./Nvidia should offer VR-ready cards at a discount to get people started. VR-only so as not to affect normal sales. It's definitely in their interest. Or they could apply for government subsidies considering VR's potential public services. I'd vote for it :)


    You can't make a VR-only GPU. Any GPU capable of VR games, would be capable for regular games.

    The price of VR ready hardware will come down as technology progresses. It's no different than it was before.
    The graphics card that launch this year will be more powerful (as they always are) at lower price points. VR-ready GPUs will be available for much cheaper in half a year to a year.

    Making GPUs more affordable now doesn't do anyone any good. VR HMDs are already selling out well beyond thier release date. There's only so many of them that can be produced at a time.
    that will change over time too, but there's no incentive to lower prices when they are already selling out what they can produce. GPU manufacturers are aware of this fact.
    2
  • jaber2
    This will be the experience of VR on a slow video card http://i.imgur.com/Ot87GQF.gifv
    6
  • Realist9
    I think the benchmark is awful and fairly useless...misleading at best. The article here is still useful though, in as much as you can get an idea of RELATIVE performance between video cards.

    I have a system I consider in NO WAY capable of VR. It's a i7 2600K, SSD, 16 GB ram, and 780ti (upgraded from a 580ti 2 yrs ago). It comes out at the top of the 'capable' range.
    Running this test implies my system would be fine if I upgraded the video card. That's ridiculous. If I did that, it would likely puke all over the floor the first time I tried to actually use it with a VR game.
    Valve, who I think highly of, wiggles out of any responsibility by saying "oh, but we didn't test your CPU".
    I wonder how much backlash they are going to get when people reference this benchmark, while complaining about vomit (theirs or their systems) from trying to play an actual game in VR.
    -8
  • Sergio_9
    The score for the 295x2 is wrong, there is a manual setting for in SteamVR that enables proper Crossfire use, my 295x2 with a 2600K @ 4.8ghz gets a score of 9.7, after changing the setting, before making the change I get the score listed above.
    13
  • tpapas_papas
    I concur with Sergio_9. Here is my 295x2 score http://imgur.com/6b7QiO9. Article is dead wrong! Please fix!
    2
  • InvalidError
    Anonymous said:
    With how expensive VR headsets are right now, I don't understand why you would pair it with such a budget friendly GPU in the first place.

    For VR to become relevant, it has to be affordable enough for the non-enthusiast to get interested in it, which means budget-minded gamers need to know how much it might cost them to get there.

    If VR ever becomes a mainstream thing, you might see headsets going for as low as $100: even entry-level smartphones have most of the technology bits required built-in. Rip out the SoC, RAM, battery, WiFi/3G/4G, SIM, SDCard, touch sensor, etc. and you have plenty of budget left to mold the head mount and optics from. There might even be enough budget for a higher resolution display or a pair of 2" ones instead of splitting a single 5".
    -2
  • picture_perfect
    Quote:
    VR HMDs are already selling out well beyond thier release date. GPU manufacturers are aware of this fact.


    But are they upgrading thier PCs for a good experience. Bad experiences could turn people off and then there goes PC sales anyway. AMD/ Nvida could think of discounts as insurance...well just throwing that out, yea I know, not going to happen.
    0
  • Sergio_9
    Quote:
    The score for the 295x2 is wrong, there is a manual setting in SteamVR that enables proper Crossfire use, my 295x2 with a 2600K @ 4.8ghz gets a score of 9.7, after changing the setting, before making the change I got the score listed above.


    I believe the setting is the -vrforce set in Steam command line setting for the benchmark. I don't have access to my PC at the moment to confirm. The Steam forum is where you can find more details. Also, for reference, the AMD driver version I ran this benchmark was Crimson 16.2.1, on Windows 10 Pro, current regular build.
    0