Thursday, October 20, 2011

So, a few encoding benchmarks: Phenom II X3 710 (2.6GHz) versus overclocked FX-8120 (4.0GHz)

I finally got my new computer in a state where I could do some encoding, so it was the perfect time for some quick (and non-authoritative) benchmarks comparing my old encoding computer to my new encoding computer.

(Incidentally, this MacBook is not a valid encoding computer.)

For those of you who don't know how I do encodes, my basic procedure is to do a master lossless raw with most of the preprocessing and filtering done, and from there use that to encode the hardsubbed final encode.   I am benchmarking the final encode step.  All my encodes are two-pass encodes to a target bitrate.

Basically, I loaded two-pass encodes for the following episodes into MeGUI, which I use for the frontend:

1. Pollyanna episode 10: This one had an uncompressed YV12 master raw.  The resolution is 640x480, at 29.97fps.
2. Pollyanna episode 51: This one had a YV12 master raw that was compressed with Lagarith.  The resolution is 640x480, at 29.97fps.
3. Les Miserables episode 52: This one had an uncompressed YV12 master raw.  The resolution is 1280x720, at 23.976fps.

They were all encoded with the 32-bit version of 10-bit x264, release 2074r2 from JEEB's site.

First, let me show you the results from my old encoding computer.


If you were wondering, I was not able to unlock the 4th core on the processor. The 4th core simply would not work at all.

Now for my new encoding computer. CPU-Z gives an erroneous reading on the processor. The processor is a AMD FX-8120, overclocked to 4.0GHz.


Basically, we can draw a few conclusions:

1. x264 is capable of utilizing all 8 of the FX-8120's cores on the second pass of the encodes.
2. The IPC of the FX processor is lower than the Phenom II processor. However, the number of cores and the overclock I applied to that easily erases the difference.
3. With this setup, I'm able to encode 720p at twice the speed of real-time.
4. Notice how the first pass for Les Miserables 52 is not all that much faster on the FX versus on the Phenom II. One of the things I noted was that during the Pollyanna first passes, each of the 8 cores was evenly loaded about 20%. With Les Miserables' first pass, only three of the 8 cores was loaded. I do not know right off the bat why this is the case.

So, this is the performance of my encoding computer. I hope it will last me a really long time and allow me to encode more releases in a shorter amount of time.

We will be back from hiatus next Sunday; I will begin the translation of the first of the three selections for our next WMT project then.

(Edit: Yes, I know about the performance of Intel's processors based on Sandy Bridge.  However, I choose this option because the nature of the encoding done here requires more cores over IPC.)

10 comments:

  1. Hope you'll enjoy the new pc for a long long time from now on! Wish you all the best and cheers until next Sunday. :)

    Glad for the Edit note, or else, we would have started an IT topic related :))

    ReplyDelete
  2. Can't wait to see which WMT series were chosen. :)

    ReplyDelete
  3. One of these days I must learn more about encoding video, but first I must have a PC capable of doing such thing, think I did have to save up some of my wages' money over the next few months to build one capable of doing anything you are doing, it's been a while now can't remember which of the three WMT we could be expecting?

    ReplyDelete
  4. 2 bad i'm changing jobs, as i could asked for a free server at my older one. Haven't thought about this possibility till now :(

    You are doing a great job, nevertheless!
    Cheers!

    ReplyDelete
  5. By next Sunday, do you mean October 23rd? Or October 31st?

    ReplyDelete
  6. Julia, we're technically off hiatus yesterday, but we don't have anything ready to release yet.

    ReplyDelete
  7. Okay. Just wanted to confirm, that's all.

    ReplyDelete
  8. OVERCLOCK THAT THING. PAINFUL TO SEE MISERABLE 1.4GHZ

    ReplyDelete
  9. Thanks for the software notes. I have gotten out of the loop on the latest and greatest for encoding.

    On a different note, have you tried the 64-bit encoder? On some internal testing we were doing, we found that compiling 64-bit code gave a 10%-20% speed boost on all 64-bit CPUs we tested. All the coder had to do was change one flag. Of course, this was under Visual Studio. No telling with other complilers.

    http://brian.harrison.org/Benchmark/BodyBenchmark.html

    if you are curious.

    ReplyDelete
  10. I'm currently not set up to do encodes using 64-bit x264. People have reported that 64-bit x264 does give about 10% increase in speed.

    The main problem is that I can't go to 64-bit AviSynth because I use quite a few filters that are not available in 64-bit. I could rework my processing pipeline so I can still encode using the 64-bit version of x264, but it would make my process much less efficient.

    ReplyDelete