Skip navigation
9storyteller9
Currently Being Moderated

Feedback required to setup CS6 based edit machine

Mar 27, 2013 11:31 PM

Tags: #nvidia #cuda #premiere_pro #cs #6 #gtx #titan #cuda_card #films

Hi

 

We are an independent film making unit and are about to start the edit process for our feature documentary (mixed medium). We have footage shot on DSLR, avchd format, iphone, ipad etc. Thus Premiere Pro (with GPU accelaration) was our first choice for an edit platform since it saves us tonnes of conversions.

 

We are trying to put together an edit machine, which will later be used for animation purposes too. This is the config we have come up with. Feedback regarding the same in context to CS6 support would be great.

 

1) Intel i7 3930K 3.2GHz 6 core

2) Motherboard: ASUS P9X79 PRO

3) graphic card: EVGA GTX Titan superclocked

4) Corasir 1200 W PSU

5) Corsair H100i cooler

6) Gskill 16GB (8GB x2) DDR3 2400MHz x2= 32GB ram

7) hdd : Seagate Barracuda 1 TB Desktop Internal Hard Drive 7200 rpm

    hdd2 : Seagate Barracuda 3 TB Desktop Internal Hard Drive 7200 rpm

 

A major confusion we have is regarding the graphic card. We have heard that the Titan is good for gaming purposes but not many reviews are out there in context to edit renders and animation renders. Please Help.

 
Replies
  • Currently Being Moderated
    Mar 28, 2013 9:19 AM   in reply to 9storyteller9

    The Titan is  very, very good and would be my choice if I were to build a new machine today. But you have to realize that the Titan is severely bottlenecked with your disk setup. That is the weak point in your suggested configuration. Only two disks is not enough for decent performance. Take a look at Final Results - Reflections

     

    The only reason for my top position is my disk setup and with only two disks, even if you have a faster video card, you will never be among the top-50.

     

    If you do not invest in a good raid controller and many more disks, you can just as well get a GTX 660 Ti or a 670, since the performance difference will be negligent, and the Titan seems overkill in relation to the rest of your setup.

     

    Have you read To Raid or not to Raid, that is the question

     

    Every system is as fast as its weakest link. I suggest a Titan in systems that are otherwise fully optimized, at least a hexa core overclocked, lots of memory (64 GB), dedicated raid controller with a large cache and at least an 8-disk array, or better, like dual octo cores. On 4-core systems the Titan is overkill. In general the cost of the video card in a well balanced system is around 10 - 12% of the total cost.

     

    Look at the picture here:

    PPBM6_results.png

     

    These are initial results from our new benchmark for CS6 and CS-Next from my own system (top) and Bill Gehrke's X58 system (2-nd).  Interesting is that both use an EVGA GTX 680 card and both use hexa cores. The differences are memory and disk setup. Even more telling is that Steven Gotz (4-th) did the test as well and he came out with 404 s total and 148 s (251 MB/s) for the disk test, despite a much newer system than Bill and despite his 2 disk raid0 array.  With non raided setups, the disk test alone would take around 310 seconds, assuming a sustained transfer rate of 120 MB/s.

     

    What I'm hoping to show you is that disk setup has a major impact on overall performance.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 28, 2013 9:42 AM   in reply to 9storyteller9

    Bear in mind, Harm's benchmarks are measuring export speed.  If 95% of your time is devoted to time-line editing, and your track requirements don't exceed the playback ability of the disks, you may see no obvious operational improvement from either a RAID or a high-end GPU like the Titan, as compared to single media disks and lower-end GPUs.  Those 5% of tasks where your system is 3x slower than someone else's, may not add up to much in the end.

     

    One place you will see an obvious daily improvement is using a SSD, even a small one by current standards, as your system drive, instead of a traditional HDD.  Things just go a lot faster.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 28, 2013 11:58 AM   in reply to 9storyteller9

    [moved to hardware forum]

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 28, 2013 4:08 PM   in reply to jamesp2

    Bear in mind, Harm's benchmarks are measuring export speed.

     

    Not true. Export speed is only measured in the disk test, the other tests measure CPU, GPU and memory performance.

    How can you make these statements when you have not even tried it yourself and understand what the testing entails?

     

    One place you will see an obvious daily improvement is using a SSD, even a small one by current standards, as your system drive, instead of a traditional HDD.  Things just go a lot faster.

     

    Another myth that just is not true for editing. Boot time may be reduced from 65 seconds to 60 seconds, but in daily editing there is no discernible difference in performance, especially not in the 95% time-line editing case as per your example.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 28, 2013 5:21 PM   in reply to Harm Millaard

    Harm, I'll happily concede I'm no authority on these benchmarks, but I have never heard anyone, including you, claim or demonstrate that the differences they reveal (one machine versus another) are necessarily consequential, if the lesser system is already more or less real-time for most of the editor's tasks.  In your own example, you argue that systems without RAIDs simply can't score in the upper ranks.  But for editors you aren't constantly exporting material, the edge provided by disk arrays may not be significant.  Similarly, the difference between high end and middling GPUs may never be experienced for many or most tasks.

     

    As for the SSD -- if I gave the impression that SSDs improve editing speed, mea culpa, that would be for lack of clarity.  I was simply referring to boot time, load time, installation time. etc.  This is noted and appreciated on a daily basis, at least by this user; I didn't think it was a controversial point.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 29, 2013 9:47 AM   in reply to Harm Millaard

    Harm Millaard wrote:

     

    Not true. Export speed is only measured in the disk test, the other tests measure CPU, GPU and memory performance.

    How can you make these statements when you have not even tried it yourself and understand what the testing entails?

     

    I agree. The test that's deemed of greatest importance to those "timeline editors" is the H.264 Blu-ray portion (and to a lesser extent the MPEG-2 DVD portion) of the PPBM6 test suite: If all of the other tests come out competitive but the H.264 test is particularly long in comparison with otherwise comparably equipped PCs, and if no resource-hogging background programs were running during testing, then the CPU and/or RAM might need overclocking and/or upgrading.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 29, 2013 10:54 AM   in reply to RjL190365

    "If all of the other tests come out competitive but the H.264 test is particularly long in comparison with otherwise comparably equipped PCs, and if no resource-hogging background programs were running during testing, then the CPU and/or RAM might need overclocking and/or upgrading."

     

    I'll say this much, and then bow out of the discussion:  I would expect most editors can tell whether an NLE which is supposed to be "real-time", is in fact real-time, without a benchmark program, assuming most of their work consists of actual time-line editing.

     

    OTOH, the added expense and time devoted to seeking better scores without a compelling editorial need to do so -- purchasing, installing, tuning and maintaining hardware, the constant preoccupation with the system, keeping up with the latest developments, consulting this forum, etc. --  beyond what's needed for actual productive editing work should be added to the benchmark results, if we're going to be fair about it. 

     

    By that measure, the faster machine will likely have no chance of ever repaying the time put into it.  Just waiting for the delivery man could easily defeat 6-months' of  measurable speed gains.  Not to mention the usual difficulties with drivers, setup, etc.


     
    |
    Mark as:
  • Currently Being Moderated
    Mar 29, 2013 11:01 AM   in reply to jamesp2

    I would expect most editors can tell whether an NLE which is supposed to be "real-time", is in fact real-time, without a benchmark program, assuming most of their work consists of actual time-line editing.

     

    Where do you get the 'Supposed to be "real-time" from..'? The supplier, the neighbour, other sources?

     

    OTOH, the added expense and time devoted to seeking better scores without a compelling editorial need to do so -- purchasing, installing, tuning and maintaining hardware, the constant preoccupation with the system, keeping up with the latest developments, consulting this forum, etc. --  beyond what's needed for actual productive editing work should be added to the benchmark results, if we're going to be fair about it.

     

    So what are you doing here? And please quantify - in your specific case - the expense and time that should be added to the benchmark results in such a way that they are comparable to other systems.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 29, 2013 11:00 AM   in reply to jamesp2

    Actually, if a CPU is too slow, you will not experience a slowdown in frame rate dirung timeline editing - but instead you will experience dropped frames on playback. The slower the CPU is, the greater the number of frames that are dropped during playback. If a system drops too many frames during playback in the editor, I would consider that particular system unsuitable for editing that particular material.

     
    |
    Mark as:
  • Alex Gerulaitis
    490 posts
    Jun 9, 2009
    Currently Being Moderated
    Mar 29, 2013 12:17 PM   in reply to Harm Millaard

    Where do you get the 'Supposed to be "real-time" from..'? The supplier, the neighbour, other sources?

    None of the above.  James already said it rather eloquently and clearly, I think.  The systems' ability to fulfil editor's exactations in terms of responsiveness of said system - is the only real criteria of "realtime".  You could count dropped frames and try to measure lag, or do something as outlandish as measure exports of 2-hr DV timelines, or something as irrelevant to realtime editing as H.264 encoding.  None of that approaches what realtime editing is.  It really does seem that high PPBM scores became an agenda of its own for so many people on this forum, including Harm - where actual editing is left far behind and is an afterthought.  The forum became largely a PPBM advertisement.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 29, 2013 12:52 PM   in reply to Alex Gerulaitis

    None of that approaches what realtime editing is.

     

    Please tell me what realtime editing is, when we do not use H.264, forget about DV and MPEG2? Why is PR supplied with codecs for DV, MPEG2 and H.264, if they are irrelevant for editing? What is actual editing in your opinion? Keep in mind you just scratched DV, HDV, AVCHD, RED 4K, XDCAM and MXF as usable codecs for actual editing.

     

    You seem to be going for the good old gut-feeling about responsiveness. Well, that is your choice. Nobody forces you to look at any benchmark results. Some others feel different.

     
    |
    Mark as:
  • Alex Gerulaitis
    490 posts
    Jun 9, 2009
    Currently Being Moderated
    Mar 29, 2013 1:29 PM   in reply to Harm Millaard

    Please tell me what realtime editing is, when we do not use H.264, forget about DV and MPEG2?

    I've already proposed my view of it, and so did James.  I also said "encoding", not "editing", which makes it clear to me that my words are either misinterpreted or simply ignored.  That makes it a very productive discussion, doesn't it?

     

    My only recommendation then is to try and listen to professional editors and integrators and see what they have to say about realtime editing, if you're not sure what it is.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 29, 2013 1:51 PM   in reply to Alex Gerulaitis

    Maybe you should read what I quoted from your post:

     

    None of that approaches what realtime editing is.

     

    I also said "encoding", not "editing", which makes it clear to me that my words are either misinterpreted or simply ignored.  That makes it a very productive discussion, doesn't it?

     

    Stop confusing people and be clear about what you mean and respond in a clear and understandable way.

     

    That makes it a very productive discussion, doesn't it?

     

    integrators and see what they have to say about realtime editing, if you're not sure what it is.

     

    I still have to meet an integrator that knows about editing. With the exception of Eric and Scott of course.

     
    |
    Mark as:
  • Alex Gerulaitis
    490 posts
    Jun 9, 2009
    Currently Being Moderated
    Mar 29, 2013 1:51 PM   in reply to Alex Gerulaitis

    Alex Gerulaitis wrote:

     

    You could ... do ... something as irrelevant to realtime editing as H.264 encoding.  None of that approaches what realtime editing is.

     

    Your response:

    Harm Millaard wrote:

     

    Please tell me what realtime editing is, when we do not use H.264, forget about DV and MPEG2? Why is PR supplied with codecs for DV, MPEG2 and H.264, if they are irrelevant for editing?

    I said, encoding into H.264 is irrelevant to realtime editing, you said, "Please tell me what realtime editing is, when we do not use H.264".

     

    Again, I didn't say anything about H.264 not being used in realtime editing, only that encoding into H.264 is irrelevant to it.  That made me realize you either didn't carefully read my post or chose to misinterpret it.  Was I wrong in that impression?

     

    Does it make sense now?

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 29, 2013 2:30 PM   in reply to 9storyteller9

    This thread got side tracked, bad.

     

     

    The 3930k is a great CPU for such a build. I'm not sure about the Titan, but in theory it should be very good for both tasks based on the specs.

     

    The biggest thing I would change about your setup is the disks. a 256gb for os/programs, 128gb if you're just using the CS suite and maybe 1-2 3D applications. This will only improve operating speed, which is really nice.

     

    A single 3tb HDD for your projects is going to be a serious bottleneck for your editing, I would suggest at least 2 HDD's in raid O or, depending on the size of the project data, another SSD. Raid 5 with 4 1TB HDDs would give you speed, security, and lots of storage, but if you go with large HDD's the risk of data loss is pretty high. With something like this I would think about RAID 1+0 with 4 2tb HDD's, 4tb of storage, around double the speed of a single HDD, and a mirror incase of disk failure.

     

    IMO for video editing, if you don't go at least RAID 0 with two decent HDD's, the titan and 3930k aren't going to get utilized enough to justify the cost. If money is a thing, drop down to a GTX 680 or even a GTX 580, from what I've seen the 580 is a great card for editing. You could even go down to a GTX570, and you'd probably get better overall performance with that and a good disk setup than you would with 2 HDD's and a the titan.

     

    Disk speed doesn't only affect render speed, it affects how quickly premiere can digest the footage. Dragging and dropping 30-50gb of footage into a premiere project can take forever on a single disk system, but with an SSD or good RAID setup, that time can be cut down dramatically. Same goes for playback, if you've got multiple streams of dense material going through the editor you'll end up with dropped frames and skipping because the HDD can't feed the system fast enough. With a large project like yours I'm sure having better disk performance would make the whole process more smooth.

     

    If you look at the results of the new PPBM, mine is the last, and it uses a single HDD for projects. Now my system is probably slower in other areas also, it's powered by a 3770k at 4.4ghz and a GTX 660, but even on my system I'm bottlenecked by my HDD. Then again, I work on 1-2 minute projects containing maybe 3-4gb of footage from a DSLR, so a better disk setup would kind of be overkill because my render times are already short. But if I had a large project to work on, I'd be waiting around for footage to import, waiting forever for renders, and having to deal with skipped/dropped frames with high bitrate/multicam footage.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 29, 2013 2:57 PM   in reply to Paul Lewis

    Paul, you are not last any longer, there have been new submissions. I'm working on how to make them available to you without compromising some confidential information. Anyway, here is a screenshot of the latest results:

     

    PPBM6 Results.png

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 29, 2013 3:12 PM   in reply to Harm Millaard

    Thanks for the update Harm! Hope you get the table all sorted out.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 29, 2013 3:46 PM   in reply to 9storyteller9

    Abhay,

     

    Many people are using the 560 video card to complete satisfaction. There are 86 people using that card in our benchmark, about 7% of the observations. Out of the most popular video cards, it occupies rank # 4 in popularity. Going over to the Titan will give you a performance increase of around 8% over a 560 Ti (plus or minus some %) but at a significant cost. It still will not guarantee it will run trouble-free. The fact that the 560 Ti does not run trouble-free is telling, because there are so many others without problems. I think you have to figure out first what is causing your problems, before you invest huge amounts into a new video card. Are you running the 314.21 driver, for instance?

     

    PS. Have you tried to overclock the video card or have other cooling problems (being in India) that might be the cause of this occasional problem?

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 29, 2013 3:50 PM   in reply to 9storyteller9

    The GTX 580 is officially supported by Adobe Premiere pro and is a great card, and in some cases it actually performs better than the newer GTX 680. The Quadro cards are way overpriced IMO, not worht it. If you check out the PPBM results there are a lot of "consumer" cards being used over Quadro cards, because in our case they work better. If possible, go for the 3gb version. They aren't cheap, but they are a lot less than a Titan.

     

    I use a GTX 660 and it works just fine, I'm not sure what the problem is with your GTX 560 ti, which is in theory better than my card for this application. I did find that Adobe doesn't like OCed cards, I tried pusshing mine another 100mhz and it would continously crash with Premiere, but not with any of my games.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 29, 2013 4:01 PM   in reply to Paul Lewis

    May I just add one (long) comment.  Back in the early 2000's when I, a hobbyist, started visiting the Premiere forum all I saw was a lot of opinions on what made a good editing computer with no justifications for those opinions.  I looked around for some benchmarks to justify opinions and really found nothing.  For instance at that point some people were recommending expensive Graphics cards when there were no GPU assisted coding.which did not appear in Premiere until version 5.0 on April 30, 2010.  Therefore since I was a retired computer systems engineer I decided (about 2005) that since video editing needed a hardware evaluation benchmark that could be done in Premiere and hence I developed the first PPBM.  The goal was to evaluate your hardware and compare it to other users and guide new systems for optimal configurations.  It never was about who has the fastest system, although how else do you learn except by push the state-of-the-art.  And guess what we now have a whole forum dedicated to Premiere Pro Hardware, can you say that about any other software.  Thanks to Harm who has made the web site what it is today and will be tomorrow.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 29, 2013 5:09 PM   in reply to Paul Lewis

    Paul Lewis wrote:

     

    I use a GTX 660 and it works just fine, I'm not sure what the problem is with your GTX 560 ti, which is in theory better than my card for this application. I did find that Adobe doesn't like OCed cards, I tried pusshing mine another 100mhz and it would continously crash with Premiere, but not with any of my games.

    Actually, not all GTX 560 Tis are better performers than the GTX 660. The standard GTX 560 Ti is less powerful and has lesser memory bandwidth than the GTX 660: Only 384 CUDA cores and 128 GB/s memory throughput for the standard GTX 560 Ti versus 960 CUDA cores (effectively 480 CUDA core-level performance versus the older GTX 5## series) and 144 GB/s memory throughput for the GTX 660.

     

    And it seems like the better scores from the GTX 560 Ti have all come from the 448-CUDA-core variant (which is actually a slightly crippled GTX 570, with a memory throughput of 152 GB/s, rather than a souped-up GTX 560 Ti). Poorer results from the GTX 560 Ti have come from the standard version (whose GPU performance is clearly held back by the lower memory throughput).

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 29, 2013 5:56 PM   in reply to RjL190365

    RjL190365 wrote:

     

    Actually, not all GTX 560 Tis are better performers than the GTX 660. The standard GTX 560 Ti is less powerful and has lesser memory bandwidth than the GTX 660: Only 384 CUDA cores and 128 GB/s memory throughput for the standard GTX 560 Ti versus 960 CUDA cores (effectively 480 CUDA core-level performance versus the older GTX 5## series) and 144 GB/s memory throughput for the GTX 660.

     

    And it seems like the better scores from the GTX 560 Ti have all come from the 448-CUDA-core variant (which is actually a slightly crippled GTX 570, with a memory throughput of 152 GB/s, rather than a souped-up GTX 560 Ti). Poorer results from the GTX 560 Ti have come from the standard version (whose GPU performance is clearly held back by the lower memory throughput).

    You will notice that on our testing of GPU's (down almost to the bottom of this page) thr good results of the GTX 560 Ti is a 448 core version

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 30, 2013 5:54 AM   in reply to Bill Gehrke

    Bill,

     

    I found your history lesson regarding the "when and why" of how the PPBM benchmark started to be quite interesting.

     

    Maybe this summary should be added to the latest site along with Harm's words on why and when he joined you in support of the benchmark and the web site.

     

    Thanks again Bill and Harm!

     

    Jim

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 31, 2013 3:47 AM   in reply to JEShort01

    Ha! I see there is a new "Historical Perspective" section on the PPBM6 site home page!

     

    Thanks Harm! (assuming you did this)

     

    Jim

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 31, 2013 4:32 AM   in reply to JEShort01

    Yeah, I found your suggestion a good one, so I added that. Yesterday I again had a look at the PPBM+ site where my involvement started, then the PPBM4 site and then the PPBM5 site and I think we have made some progress in the years. But I never expected that buidling websites was so time consuming when you have no experience at all and you really have to learn the basics of .html, .css, .php, .js, Joomla, MySQL and the like.  Now the struggle in the automated exploding of the Output.csv file and the XML loading of the Speccy file into the database and this is all a one-man-show.

     

    Personally, I think the learning curve is a lot steeper than learning PR, but that is also caused by the fact that my profession was HR consultant, nothing technical or IT related.

     
    |
    Mark as:

More Like This

  • Retrieving data ...

Bookmarked By (0)

Answers + Points = Status

  • 10 points awarded for Correct Answers
  • 5 points awarded for Helpful Answers
  • 10,000+ points
  • 1,001-10,000 points
  • 501-1,000 points
  • 5-500 points