EEVblog #726 – Dual Xeon Video Editing Machine Build


Dave builds and tests the 24 core dual processor 2.6GHz Intel Xeon E5-2630 v2 video editing machine and benchmarks against his current 8 core i7 3770K machine at 3.5GHz.
This is one hour of PC building and software testing, if you find this stuff boring, DON’T WATCH IT!
Forum: http://www.eevblog.com/forum/blog/eevblog-726-dual-xeon-video-editing-machine-build/

UPDATE:
– Even doing a 1 hour render at 240W consumption, the heatsinks do not even get warm to the touch.
– Yes, I need to fill all the blue slots (4 channels) to get best memory performance
– No, EEC memory is not slower than Non-EEC.
– No, the CPU’s are not thermal throttling. The core temps barely reach 40degC
– I put the video card in the slot for the CPU2, not the main CPU1.
– Direct uncompressed output using the Sony YUV Video For Windows codec actually works (it had issues on my old machine)
1:46 for the 1min test video which is pretty good and the fastest yet. That was to the SSD. File size was 12.163GB for the 1 min, so that means a 1 hour video would need 730GB.
Such an output would need about 114MB/s write speed, so that makes a 7200 rpm drive a WD Black suitable. 1GB+ of SSD would of course be very expensive and of no improvement.
So this seems like the best solution for now.

Be Sociable, Share!

    About EEVblog

    Check Also

    EEVblog #1035 – Flaming DIY Power Supply

    Smoke in the EEVblog Lab! What component failed and caught alight in the RD Tech ...

    • Synthetase

      Looks like a nice machine you have there 🙂

      With all the focus on video, in my opinion the audio could use a little attention, too. I’m hearing a lot of compression artefacts here.

    • TheSlider2

      Doesn’t Movie Studio have a rendering thread limitation ?
      I believe even the x64 version might be limited to 8 or 16 threads. There is probably a way to force it to use all cores by setting a value in some ini file.

    • Martijn Veen

      Hi Dave, the E5-2630 only has 6 cores per cpu and 12 treads per cpu = 2 treads per core,
      http://ark.intel.com/products/64593/Intel-Xeon-Processor-E5-2630-15M-Cache-2_30-GHz-7_20-GTs-Intel-QPI

    • Dave, I think your problem is with workflow much more than the machene’s speed. Rendering is something best done in your downtime, i.e. overnight. Set it going and it’s done in the morning ready to upload, as long as it takes less than 12 hours it’ll be done when you get in.

      Rendering is inherently computationally expensice and in the genuenly pro space (big films) nobody is trying to do it on a single machene. What you want to do is break up the tasks into smaller chunks and have that processed by multiple machenes at the same time. Then combine the output. I think ffmpeg/handbrake is capible of joining multiple streams.

    • Griffin

      I know this might be a shot in the dark but a few things I would suggest:

      1. Under power options make sure you are set to high performance and not balanced as this could also have an effect on the hardware utilization of the OS

      2. I also think you should try doing this on Windows 8.1 or Server 2012. Server can be downloaded as 180 day evaluations at :

      http://www.microsoft.com/en-us/evalcenter/evaluate-windows-server-2012-r2

      3. As Robert mentioned try rendering this on multiple machines especial now that you have the i7 and this machine. Keeping in mind that a singe machine can only do so much.

    • H:S

      I’m afraid that in this case the low speed gain may be more related to a fundamental problem with parallel algorithms rather than slow memory or unsupported Hyper Threading.

      Take a look at Amdhal’s law (http://en.wikipedia.org/wiki/Amdahl%27s_law ), I think it applies to your situation very well: HandBrake is about twice as fast, so it shows that the new build indeed IS ~2 times faster than the i7 machine and HandBrake uses a well written parallel algorithm that makes a good use of that. At the other hand, as you saw, Sony Vegas did not utilize the second CPU very well all the time — meaning that there are simply some tasks that cannot be effectively split between many cores (or, less likely, that all the codecs are poorly written).

      In short, parallel algorithms are hard. We hit the “power wall” half a decade ago and the only way forward seems to be adding more CPU cores, but there is no easy way to make full use of them — meaning that in the near future for most applications (including Sony Vegas) it will be much more beneficial to use a CPU with higher “clock speed” rather than higher number of cores. (But hey, make your own mind, it’s just my assho–, I mean, opinion. 😉 )

      • Sebastian

        I was going to point out the same things as well.
        Especially Amdahl’s law.
        Some software just can’t be made faster by throwing more cores at it.

        As a computer science student I have access to some rather big machines at the university.
        We are talking about something in the area of 64 cores with 128GB.
        In my experience whenever I encounter a software that is rather slow on my i7 desktop machine it won’t be of any use to run it on one of this big machines.
        Because either the problem I’m working on can not be split in smaller parts to be dealt with in parallel or the software I’m using is poorly written.
        I’ve encountered cases where more thread even made things slower due to communication and management overhead.

        So my guess would be that the codecs for Sony Vegas have been written with
        4 or maybe 8 Cores in mind, which a the typical configuration for your desktop machine.
        The size of the sub problems to work on in parallel, the communication between thread all these things are probably optimized for that number.
        Handbrake on the other hand is from the Linux world where machines with many cores are not that uncommon.

        What might be worth a try if you have a two stage process already:
        Export the video from sony vegas using a codec that does not compress things at all.
        And then do all the compression using handbrake.
        Of course the first step will result in a ridiculously large file, but you don’t need to keep it once you have the compressed one.
        I did something like this for a 2h 1080p video I needed to render this weekend,
        and it worked rather well for me.
        Still I’m just guessing over here.

        Also nice new machine you got there.
        Must admit it made me a bit jealous.

      • errsu

        An additional scalability issue might be the bottleneck between the two processors. In your video there were two groups of threads, the busy and the idle ones. Do they group along the chip boarders? In video rendering, esp. for uncompressed material, you could possibly trick out the scalability issues by running two renderers in parallel, each using 6 threads, assigned in a way that they are on one of the processors only, and then stick the results together. Question is if you can start the Sony SW twice.w

    • Alvin Andries

      Hi Dave,

      Hyperthreading is not floating point friendly. This means that you have a lot of contention for the FPU. So, as a rule of thumb, only use hyperthreading for I/O intensive workloads.

      Regards,
      Alvin.

    • Andrej Gelenberg

      You can try to render multiple videos at same time if your software don’t scale that good.

      More cores should help with that even if it doesn’t shorten rendering time for a single video that much.

    • Sue

      Faster memory will make no difference.
      It’s the software you are using.

    • tlhIngan

      You’re better off upgrading. Your 3770 is I think Ivy Bridge, while just going one up, to a 4770 (Haswell) is a *significant* improvement. Going to the latest 4790 improves things just a little more.

      Xeons are usually a generation behind because those are to go in super-stable products like servers, so long after the architecture has proven itself will Intel make a Xeon equivalent. So I don’t think there’s a Haswell Xeon part yet.

      A 4770 should be relatively cheap by now as it’s a generation behind, and Haswell has a huge speed jump over the previous gen – enough so that if you have previous gen, upgrading to haswell was definitely worth it.

    • Herman

      I have an opinion as well, but I’m sure Ill just be blowing it out of my a…..

    • brettm

      It could be your Sony software. one benefit of software aimed at pros is a focus on performance. not always, granted. but I would suspect the Sony software because Handbrake is >2x as fast. That is a clue that the machine is working to performance but the Sony software is not taking advantage of it. I am not a coder but from what I understand the software has to be coded to take advantage of CPU efficiencies and options. For $40US can you sign up for month with Adobe CC and download and try the latest version of Premiere. it would be worth it just to trouble shoot.

    • Peter Z

      Did you try disabling power management in BIOS? We’ve had a few dual 2630’s that just did NOT want to perform unless power management was off.

    • bill brant

      If the CPUs are less than 100% used, then they are waiting on the disk or the RAM. You say that SSD versus platter type disks does not matter, so your CPUs are waiting on the RAM. Faster RAM will help a little but I suspect using more DIMMs will be better because the mobo has so many DIMM slots (and I guess Memory Controllers). You need more lanes (not a higher speed limit) to get more bandwidth.

      I would try to get 100% CPU on one processor first by filling the DIMMS on CPU1 then try to use both CPUs. You need to experiment.

      The tools you are using are very basic, you need more data.

    • Mark Vaynberg

      Hi Dave,

      You should try Intel Quick Sync (Google it).
      It comes built in with every Core i7 from the first generation. Even your old 3770 have one,
      It’s part of the GPU.

      To make it work you need to take out the external GPU and to enable the internal GPU in the BIOS, some BIOS will have Quick Sync enable/disable option.

      Then you should use one of the codecs that support Intel quick sync, list in Intel website.

      The new Core i7 5XXX series has a huge leap in performance compare the old ones.

      This will probably be your best solutions for fast encoding.

      Regards, Mark.

      P.S. Upgrading the RAM to a faster one will not increase the encoding speed!!!.

    • blipton

      Any links to the videos (or episode #) where Dave talks about why the gpu can’t be used for rendering for Sony Vegas.. and/or why ssd doesn’t help with the performance? why mp4?

      Also can you run a benchmark test from passmark to see how it compares with others using similar hardware?

    • the eggroll

      Hi Dave,

      That machine doesn’t have 24 cores, it has 12 cores and 24 threads.

      Hyperthreading can be useful for some applications, such as integer math or widely varied tasks (some threads needing I/O, some needing memory bandwidth, some needing number crunching), but when you have a lot of copies of the same floating point task, hyperthreading just slows things down.

      Try disabling hyperthreading in the BIOS and run your encoding job on your 12 real cores to see how it compares.

    The EEVblog Store generally ships twice a week, on Tuesdays & Fridays, Sydney time. Dismiss