3

I have been perusing this page: https://openbenchmarking.org/showdown/pts/ffmpeg

Something doesn't make sense to me, I am looking at the following CPU's:

Intel® Xeon® Processor E5-2650L v3 (30M Cache, 1.80 GHz)    
Intel® Xeon® Processor E5-2690 v3 (30M Cache, 2.60 GHz)     
Intel® Xeon® Processor E5-2640 v3 (20M Cache, 2.60 GHz)     
Intel® Xeon® Processor E5-2670 v3 (30M Cache, 2.30 GHz)

enter image description here

According to the benchmarking, the ideal processor is the E5-2640 from that list. I don't see the correlation between the benchmark numbers and the CPU specs, Now the page didn't specify which version of that particular model, so I picked a pretty new one that is almost identical to the newest v4.

I would have assumed that when picking a CPU for video encoding I would be looking the either the:

  1. Highest base frequency
  2. Largest onboard cache
  3. Fastest bus speed

Or some other common metric. What is your experience?

3
  • 1
    a) That pts test is using an old version of ffmpeg b) it's encoding to DV, which is an intraframe codec. Not sure how much that taxes modern CPUs. It'll tax disk I/O a lot more.
    – Gyan
    Commented Nov 1, 2016 at 17:31
  • I had wondered about the disk I/O issue considering all these CPUs can't be on the same system. While yes it is an old version, the comparative numbers between CPUs seemed like it should be similar across all ffmpeg versions.
    – Alan
    Commented Nov 1, 2016 at 17:36
  • do your own testing... I have a quad core E3 v5 being 4 times faster than a quad code E5 v2 at x264 encoding. Ram isn't the reason as the DD3 is 1866 while the dd4 is 2100. I don't get it either.
    – Ray Foss
    Commented Jun 18, 2019 at 14:43

2 Answers 2

1

The DV encoder used in the cited test is useless for gauging modern CPUs.

I ran the test command with the referenced media on a 4-core i5 and the difference between -threads 4 and -threads 3 is 0.5% and with -threads 2 is 1%. I see a much larger variance when I use H.264 encoding.

Of course, the bigger flaw is that the command is resizing the video. FFmpeg's internal scaler isn't multi-threaded!

1

A big question is will your encoder use multithreading.

I have a dual E5-2697 v2 System, 48 Cores.

So if I need to just encode video which does not or cannot, due to effects applied, use the GPU (cuda); then encoding in AE rather than PPro is much faster, as it can thread up to 32 cores, and render much faster.

Otherwise; go with a processor that can be overclocked to the highest level. If your encoder can only use 1 core, you want the fastest clock speed possible... not more cores.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.