No announcement yet.

2D graphics test is very flawed

  • Filter
  • Time
  • Show
Clear All
new posts

  • 2D graphics test is very flawed

    First this is my specs and benchmark scores.

    When I ran PerformanceTest first time, I scored 99th percentile in every mark except the 2D Graphics Mark, which was only in the 95th percentile. I was very much confused by what's causing the poor score when all other components are at the top of their game. Then I noticed that every 2D test was done in a very small window (my desktop was in 4k). I changed my desktop resolution to 1080p and reran the test. Boom, this immediately got the score to the 99th percentile. I ended up submitting my baseline with the 2D test done in 720p as the score was even higher than 1080p.

    This really shouldn't be the case as it punishes people using high resolution displays. Hopefully you guys would consider changing how the 2D test is done.

  • #2
    which was only in the 95th percentile. I was very much confused by what's causing the poor score
    95th percentile isn't a poor score.

    We have noticed over the years that the cost of the video card and monitor have almost no relationship to their 2D performance. So generally speaking you don't need to buy high end parts to get good 2D. (though a fast single threaded CPU can help)

    As you noticed the 2D tests run in a fairly small window (compared to 4K resolution anyway). This was done deliberately to keep the number of pixels displayed equal on all displays. So if everything else is equal, the 2D score should be about the same, regardless of the resolution. But everything might not be equal. For example Windows might be changing the font scaling as you drop the resolution or the behaviour in the driver might not be the same.

    In our testing here we saw a 7% difference in the overall 3D mark going from 4K resolution to 720p. So not very much of a difference. But enough to move the percentile score a few points. Hardly flawed.

    We spend a bit more time today to confirm it isn't our code causing the difference. In the end however it is a job of the benchmark to measure the performance of the system. If the system is slightly faster as a lower resolution, then it is important we report this (not hide it as you are proposing).