IPP H264 performance issues with multi-threads

[Karl_K_1]

Dear All,

I'm developing a transcoding unit with IPP libraries. So, I need multiple instances of encoder and decoder. Especially with H264, it seems not to be scalable with a multi-threaded environment. With one or two H264 encoder and decoder instances (30fps@VGA), there is no huge performance problem. But, if I use more and more H264 encoder and decoder instances, CPU increases dramatically.

The only answer I found in this forum is "DO NOT USE OPENMP". But, the problem is that with Xeon E5620 @ 2.4GHz CPU, I cannot decode and encode a H264 30fps@VGA stream in real time without OpenMP support.

Please tell me if I missed something.

thanks

[Chao_Y_Intel]

Hello, Disabling the internal threading is one way when you are using the nested threading case. Another way to reduce the CPU usage is discuss here: http://software.intel.com/en-us/articles/high-cpu-usage-and-intel-ipp-threaded-function Thanks, Chao

[Karl_K_1]

Thanks for your answers. I've already tried to call IPP H264 functions with only one thread to avoid the nested threading case, but I cannot use the whole CPU fully. I think disabling the internal threading is the only way i can use with multiple instances of H264 encoder & decoder. If you have any experience with multiple instances of H264 encoder & decoder, please tell me. thanks, Karl