While the QuickSync encoding engine doesn't seem to use up much CPU, it seems to use up a lot of GPU cycles (according to ProcessExplorer).
Also, when I'm rendering and encoding at the same time, rendering performance suffers.
Is this a known and expected behavior? I was hoping that the encoding is done in a separate unit that can work indepndently in parallel..
Andras
Hi,
The degree of GPU load depends on the workload. It's hard to say if what your experiencing is normal or not without a better understanding of your workload. For instance if you're running an efficient transcode pipeline your GPU will likely be under high load.
The Quicksync encoding component does use both GPU execution unit (shared with generic GPU operations) and GPU fixed function units handling encoding specific tasks. If you'd like to explore the GPU load further I suggest checking out Intel Graphics Performance Analyzer (GPA) which will give you a momentary view of the GPU load (per unit) and also allow fetching GPU workload traces.
http://software.intel.com/en-us/vcsource/tools/intel-gpa
Regards,
Petter
I see. We do have a very high workload on the encoder. I was hoping that encoding would have no effect on rendering performance, but it seems like it does. I cannot use GPA, since our application is using OpenGL for rendering..
Aha, you're using OpenGL
.
That will certainly lead to at least some CPU/mem bandwidth overhead. The primary validated use of Media SDK/Quicksync is for applications using DirectX. From a functional perspective there should be no issues using OpenGL. But I do not think the concurrent use of DirectX and OpenGL driver and how it impacts GPU performance is well understood. If you are concerned about coexistence of OpenGL and DirectX and the impact on performance I suggest you raise the issue on Intel graphics driver forum instead.
Regards,
Petter