How We Tested The HTC Vive
We used two different computers for our Vive evaluation. The first one, our standard test rig, employs an Intel Core i7-5930K, 16GB of Crucial Ballistix DDR4 memory, a pair of 500GB Crucial MX200 SSDs and an MSI X99S Xpower AC motherboard. We used this system to record all of our performance metrics.
We tested a series of 10 different GPUs with the Vive, including several VR-ready cards and a few that likely have no business powering a VR HMD.
XFX provided us with one of its R9 390X cards and an R9 Fury for our VR testing. PowerColor sent an R9 380X Myst Edition, and Zotac sent its top dog, a GTX 980Ti AMP! Extreme, to represent a best-case scenario.
|Products||PowerColor R9 380X Myst Edition||XFX R9 390X||XFX R9 Fury||Zotac GTX 980Ti AMP! Edition|
|GPU||Antiqua XT||Grenada XT||GM206||GM200|
|Max Core Clock||1020MHz||1050MHz||1000MHz||1253MHz|
|Max GDDR5 Transfer||5900MT/s||6000MT/s||7220MT/s|
We also dug into our graphics card stash to provide results from a Gigabyte GTX 970 Windforce, an Asus GTX 980 Matrix Platinum and Sapphire’s R9 390 Nitro. We don’t have an R9 280 on hand to test the Vive's suggested minimum specification, but we do have an Asus R9 285 Strix that we'll throw into the mix.
Our testing utilizes current drivers, including GeForce Game Ready Driver 364.72 and Radeon Software Crimson 16.3.2.
After getting our hands on as many games as possible, we picked five for benchmarking the Vive. Our procedure calls for gathering performance data for two minutes at a time. Curiously, none of our game choices appear to offer configurable detail settings. Rather, each one launches straight into the action.
|Job Simulator||Two-minute test - Office Worker scenario|
|Tilt Brush||Two-minute test - Space Dragon drawing replay|
|Space Pirate Trainer||Two-minute test - Gameplay from round one|
|The Gallery||Two-minute test - Gameplay after opening scene|
|Vanishing Realms||Two-minute test - Gameplay starting after leaving training room|
For our real-world testing, we used the same rig and a lower-end machine with an Intel Core i5-4670K, 8GB of Corsair Vengeance DDR3, three 128GB SanDisk SSDs and an Asus Z87 WS motherboard. We armed that second system with a GeForce GTX 970 to represent the lower threshold of recommended hardware.
The VR benchmarking landscape is somewhat barren right now (for our insights on HMD performance testing — such as it is so far — refer to our explanations regarding the complexity of VR testing in our Rift review). Basemark’s VRScore isn’t due out until later this year, so we weren’t able to include it today. Last week when we tried to run VRMark on the Rift, we learned that Oculus' latest runtime isn't compatible with the metric. Fortunately, the Vive doesn’t suffer the same fate.
VRMark relies on SteamVR to access the HMD, which is what the Vive is built upon. Naturally, it just works. The press build is still very early though, and Futuremark tells us it's still evaluating which tests are valuable and which ones aren't as important.
The measurement we have access to at this early stage conveys the time between a draw call and an image appearing on-screen. VRMark sends a signal to the HMD and an external sensor is used to detect when the display initializes and when it goes dark again. The software then compares the measurements with the draw call to determine the total draw latency, response time and frame persistence. For a more detailed explanation of how the test works, see our VRMark Preview.
The sensor that Futuremark provides can test one lens at a time. We ran VRMark on each side of the HMD to facilitate comparisons between panels. Ideally, you want the measurements to be as similar as possible. It’s unlikely that you'll see two panels perform exactly in sync, but the closer the better.
The left eye of our review sample returns 45ms of total draw latency and 11ms for both response time and frame persistence. The right eye is slightly lower, reporting 44ms of total draw latency, 9ms response time and 10ms frame persistence. We used our Radeon R9 Fury for this testing.