Hi All -
We are trying to leverage Xeon E3 based micro-server as video server for encoding the raw data YUV into H.264. It will be current multi-session based.
We are trying to understand the Max throughput capability of the hardware encoder for H.264 in therory : number of megapixels per second...
We can not find any information regarding this ...
We understand that they are many factors which will impact the actual performance and the trick of using Intel Media SDK how to handle the task in parallel ...
Is there any way to get the the max throughput in terms of : number of megapixels per second ?
Very similar , we understand cpu capability using the parameters like : flops or instructions issued per cycle ...
For video encoding, benchmark metrics are commonly measured in 'frames per second', and based on specific content and using fixed encoding parameters for size ("level"), features ("profile"), and quality. The goal of encoding is not to encode 'a pixel', but is instead to encode something that when decoded appears similar to the original (unless the target is 'loss-less encoding' which Intel Media SDK does not support.). For example, the performance can vary greatly based on the amount of motion in the content.
Also, there are many benchmarks that capture performance of the overall task, instead of just the 'video encoding' part of an operation (since audio, disk or network I/O, and other factors play a significant role in most use cases).
There are other factors as well, including the specific platform being used. For example, the specific Xeon E3 model used, the memory configuration, etc. is all important.
Hope this helps.
Feel free to use our free H.264 Quick Sync benchmarking tool. [no install, email address, or registration is needed]
I believe it may answer your performance questions, if you have the computer system in question.
It can be configured in the following ways:
-TargetUsage [speed/quality tradeoff]
-Resolution [640x480 to 4096x2160]
-Direct3d memory / system memory
Hopefully you will find it useful. [The first time you run it, it does a 90 second comprehensive benchmark, after that, you can try any parameters, settings you like]
Best regards, Cameron