- Jan 28, 2014
TechPowerUp has tested resizable BAR performance on Ampere cards across all games that support it in their current testing suite. They found that the performance delta can vary wildly between games and resolutions, with resolution obviously playing a big role. They show how resizable BAR can be enabled on other games through NVIDIA Inspector, but per usual there may be bugs and anomalies that could pop up in titles that are yet to be officially supported.
Out of our 22 game test suite, that's 12 titles, which is pretty decent. The gains vary wildly, though—around 20% (!!) in Gears 5 and almost nothing in other titles. On average, we saw a 2–4% performance improvement from Resizable BAR when counting titles with BAR support in the driver. Now, we of course have 10 more titles in our test suite without BAR support, which drags down the overall average, but I feel that looking at "BAR-only" games is the better approach since almost every title will benefit from Resizable BAR, it's just a matter of game support. While "a few percent" doesn't sound like much, it's free extra performance for a minimal investment and no loss in image quality. I wish I had insight into how many man-hours GPU makers spend on BAR support vs. traditional driver game optimizations.