I was developing Vulkan code against the Apr 19 beta driver. I recently upgraded to the latest drivers for my Radeon 270X and saw performance of my test case go from 1500 fps to 500 fps. A friend tried the latest driver on his 290, and saw an even more extreme problem -- the demo went from 2500 fps to 100 fps (one hundred!)
Is the 16.7.3 driver really that much slower than the Beta for Vulkan? Why?
Thanks for any info.
FPS is not a valid performance metric, it would be more helpful in even surmising what could have happen if you posted actual wall clock times. This gives the reader a more accurate perspective as to the severity of the problem. Saying the app goes from 1500 fps to 500 fps, as a developer, does indicate to me any massive problem...as something as simple as missing vsync will essentially half your frame rate. Gamer's care about fps, if you are a developer, use accurate timing to convey performance differences as that is directly measurable and less subjective.
It would be useful if you could shed some light on your approach to measuring FPS. Also, without having access to actual source code of the application, it is difficult to elaborate on potential reasons behind the performance degradation you are describing.
My test case turned out to have some errors in it, but the effect persists even after these are fixed. In fact, you can run some of the better test cases, like Sascha Willems Vulkan code at
On my NVidia 1060, the instancing test case runs at 1900 fps. On the Radeon 270X, it runs at 240 fps. This is a much slower performance than it should have, and slower than it did have on the Apr 19 beta.
I have no idea when the performance changed. I was using the beta for months and then upgraded only recently when a friend told me the production drivers now handled Vulkan.