news basemark gpu benchmark crosses api and os boundaries
News

Basemark GPU benchmark crosses API and OS boundaries

The most common metrics for graphics card performance tend to be games, but modern GPUs are tasked with doing a lot more than creating pixelated representations of explosions and car crashes on Windows PCs and smartphones. The fine Finns at Basemark have taken the covers off Basemark GPU, a utility intended to test the mettle of graphics chips using a variety of 3D APIs on multiple desktop and mobile platforms. 


Was this inspired by Unigine Heaven?

The most difficult scenario in Basemark GPU performs an offscreen test at a resolution of 3840×2160 and requires 4 GB of video memory. A less-aggressive version of the test needs one-quarter of the memory and targets the common 1920×1080 resolution. As is common among benchmarking tools, Basemark GPU is available as a feature-limited free app and a paid version with additional features. The free version always submits results to Basemark's Power Board, something hardware and driver developers probably avoid under most circumstances. Both versions of the tool can test onscreen and offscreen results, a feature that should help make comparisons between desktop-bound and mobile platforms somewhat more direct.

Besides its work making benchmark tools for VR and web browsers, Basemark also offers an “industrial-grade graphics and compute engine” it calls Rocksolid. That engine runs on a variety of APIs including OpenGL 4.5, OpenGL ES 3.1, and Vulkan 1.0. In turn, the Rocksolid-powered Basemark GPU test supports those same APIs. The company says its GPU test tool also supports embedded and automotive OSes like QNX and Integrity, and that future releases will add support for more APIs like Microsoft's DirectX. Getting onto iOS and macOS will require supporting the Metal API given Apple's recent deprecation of all third-party graphics APIs on those platforms. For the time being, Basemark is offering its GPU benchmark tool for Windows versions 7 through 10, Debian and Ubuntu Linux, and Android.

0 responses to “Basemark GPU benchmark crosses API and OS boundaries

  1. Still working on it!

    (More like life is complex and it’s this year that I’m getting around to working on it seriously, but it is now going somewhere. I forget what I’ve mentioned where – it’s a subset of a much larger project.)

  2. BaseMark? Rightware? What relationship do these companies have to each other, and to FutureMark, which is now technically UL?

  3. I would like to see one of these benchmarks that actually does some sort of frametime capture while monitoring CPU and GPU usage and temperatures in the meanwhile. Then, at the end, produce an actual analysis of the performance instead of “LOL AVG FPS / 10 THERE’S YOUR SCORE BRAH BUY MORE GEFORCE”.

    Tell me why my system is slowing down or hitching, don’t just give me a dumb number. Why doesn’t this utility exist yet? (´Д⊂ヽ

  4. i7-6700K (stock) – GTX 1080 (stock) – Offscreen
    Vulkan – 6484 :: Avg 64 :: Min 45 :: Max 195 (3840×2160)
    OpenGL – 6532 :: Avg 65 :: Min 36 :: Max 155 (3840×2160)

    Pixel 2 XL – Snapdragon 835 – Offscreen
    Vulkan – 2047 :: Avg 20.48 :: Min 12.37 :: Max 41.30 (1920×1080)
    OpenGL – 2083 :: Avg 20.84 :: Min 13.42 :: Max 41.06 (1920×1080)

    I’m not quite sure how that PC score works out being higher for OpenGL but /shrug

  5. Downloading this now.

    Can I run it on my S7 and Vega56 at the same settings?

    I just want to see what the difference is between Android and PCMR because I’m pretty sure that difference gets smaller every year….

  6. It does seem a little uniform considering the balance of CPU and GPU hardware is different across the three SoCs. I might have expected more from the S6 and less from the Pixel since those CPU cycles on the older hardware are more precious.

  7. It seems strange that it’s a consistent 25% across all three. I’d expect the results to be different, but not to see the same difference across all of them.

  8. I’d be curious to see if a similar pattern holds on a desktop system with a discrete GPU assuming there’s a competent OpenGL implementation being used.

    I wouldn’t be surprised if a big part of that 25% is reduced CPU usage that frees up power for the graphics to run in a constrained environment.

  9. Just for fun, I ran this in both Vulkan and OpenGL ES versions across three Android phones: Galaxy S6 (Snapdragon 810), Galaxy S7 (Snapdragon 820), and Pixel 2 (Snapdragon 835). The actual numbers are meaningless, but there’s a consistent 25% increase on all three devices going from OpenGL ES to Vulkan. It’s expected, I suppose, but nice to verify.

  10. Sorry BaseMark… nobody wants to post about you… so I will… and maybe I’ll even download you tonight so don’t look so underwater.