Skip navigation
Currently Being Moderated

PPBM5 - questions

May 25, 2013 4:51 AM

Hello,

I have been using premiere for years and finally decided to do this ppbm5 test as my system doesnt seem to be as fast with CS6 as it should be.

Specs are

 

i7 2600k @ 4.6GHz

32GB

1600mhz ddr3 @ 9,9,9,24, 3t

primary drive OCZ Agility 3 64gb windows 7 64bit

primary premiere drive RAID 0 4TB- Western digital blacks 2TB X 2

Media cache drive OCZ Agility 128GB

 

all running on a p8p67 sata 6gb/s

 

Graphics card is geforce 285 1gb but i also run with 460 1gb  but have determined 285 is 15% quicker with testing for what i do... so 285 is in currently

 

When rendering on cs6 to h264 output with some effects that are not cuda enabled i am getting 50% cpu utilization and around 15% GPU does that sound about right?

 

I am running the ppbm5  benchmark at the moment so i will get back with results..

I couldnt find microsoft avi in the export dialogue but i chose AVI

 

I dont have bluray preset HDTV 1080i 29.97 High Quality. so i used HD 1080i 29.97

 

please let me know if this will affect my benchmarks

 

thankyou

Oliver

 
Replies
  • Currently Being Moderated
    May 25, 2013 5:10 AM   in reply to oliewdj

    Oliver,

     

    I suspect both your PPBM times and real-world work would be better served by putting all files on your 2x HD RAID. SSDs until the more expensive, latest generation models are slow on large file writes.

     

    Regarding your 50% cpu utilization and 15% GPU for some effects that are not cuda, that sounds very possible.

     

    For CS6 and your pretty beefy system, I would suggest Harm and Bill's relatively new and very nice PPBM6 benchmark for testing and tweaking your build; you can download it from ppbm7.com. CS6 outputs are faster than those set up from the AME queue and PPBM6 takes advantage of this difference.

     

    Regards,

     

    Jim

     
    |
    Mark as:
  • Currently Being Moderated
    May 26, 2013 7:53 AM   in reply to oliewdj

    [moved to hardware forum]

     
    |
    Mark as:
  • Currently Being Moderated
    May 26, 2013 9:53 AM   in reply to oliewdj

    Oli,  thanks for using out new PPBM6 test.  I do not know if you have registered yet but if you do you can view the results.  I have a system very similar to yours.  the result of (about) number 19 show my MPEG with GPU assistance at 45 seconds.  Now I am running a GTX 660.  So youi 70 seconds is especially bad considering that test the CPU was only running at 3.9 GHz.  I would definitely try your GTX 460 but make sure you enable it by adding the "GeForce GTX 460" to the Adobe file "cuda_supported_cards.txt".

     

    More later, I have to run, right now but I suspect your CPU at 4.6 GHz is not optimizes for editing from the H.264 results

     
    |
    Mark as:
  • Currently Being Moderated
    May 26, 2013 1:21 PM   in reply to oliewdj

    Oliver,

     

    Your system does seem to be working properly. And, as you might expect, your aging GTX 285 does appear to be the weakest link in your present system for CS6; it is most probably holding back your render to DVD speed.

     

    You can use Windows task manager to monitor cpu (and memory) and a tool like MSI Afterburner or EVGA Precision to monitor your GPU usage during various tasks in CS6 to better understand what tasks require what hardware resources.

     

    If your goal was to make sure your current hardware is configured properly for CS6, you have probably completed that already.

     

    If your goal was to consider adding to your system to gain speed in a particular area what exactly in your workflow is currently a bother?

     

    For example, to dramatically drop the time to render a HD timeline to DVD you could buy a newer, faster GPU.

     

    However, if you wish to dramatically drop the time to render a HD timeline to Blu-ray or do work with 4k (its a beast), think fast 6 core or a dual Xeons.

     

    Regards,

     

    Jim

     
    |
    Mark as:
  • Currently Being Moderated
    May 27, 2013 6:14 AM   in reply to oliewdj

    Oli,

     

    Here's the reason why the GTX 460 underperformed compared to the GTX 285 when rendering video clips with lots of effects: The 460's lower memory throughput. Depending on the revision, the GTX 460 has a memory throughput of only 115.2 or 96.2 GB/s versus the GTX 285's 159 GB/s. But the GTX 285 underperformed versus the GTX 460 in the MPEG-2 DVD test with CUDA enabled due to its low number of CUDA cores: The GTX 285 has only 240 CUDA cores versus the 336 CUDA cores in the GTX 460. For comparison, the GTX 460 SE has 288 CUDA cores and 108.8 GB/s memory throughput.

     

    And the relatively poor disk performance score suggests that you might have a bottleneck somewhere in your system, or that your system configuration might need some tuning (to eliminate unneeded background processes). Or it just might be that your two Hitachi 2TB hard drives just do not work toghether very well when RAID 0'd.

     

    By the way, my system that you referenced to actually has a 448-core version of the GTX 560 Ti with a memory throughput of 152 GB/s. As such, it is actually a slightly crippled GTX 570 (the 570 having the same memory throughput as my 560 Ti 448 but with 480 CUDA cores). The "normal" GTX 560 has the same number of CUDA cores as the GTX 460 (336) but with slightly higher clock speeds. As such, it is more than 50 percent slower than the higher-end GTX 560 Ti 448 (as tested in my secondary rig using the PPBM6/7 script).

     
    |
    Mark as:
  • Currently Being Moderated
    May 27, 2013 6:10 AM   in reply to oliewdj

    Oli,

     

    Regarding why is the GTX 460 faster at times and the GTX 285 is faster at other time, consider the # of cuda cores and memory bandwidth:

     

    GTX 285 (flagship gamer's card from may years ago:

    # cuda cores = 240

    memory bandwidth = 159 GB/s

     

    GTX 460 (newer, but much lower price point card when introduced):

    # cuda cores = 336

    memory bandwidth = 86 GB/s

     

    So, with the PPBM6 benchmark for rendering HD to DVD the GTX 460 with more cores is producing a better result. And when doing work that does not tap into the CUDA core capabilities it is probably reasonable that the GTX 285 with almost twice the memory bandwidth.

     

    Just as a reference for nVidia's flagship gamer's card from 2012, the GTX 680 (in Bill's PC tested) has:

    # cuda cores = 448

    memory bandwidth = 192 GB/s

     

    And for their current flagship in 2013, the GTX Titan:

    # cuda cores = 2688

    memory bandwidth = 288 GB/s

     

    Now, regarding your workflows:

    • mpeg2 speed would likely go up with a newer faster video card - Bill is your man to help here as he has done a lot of experimentation with a PC having exactly the same cpu as yours
    • h264 speed would not likely be helped much by a newer faster video card as you are much more CPU bound in this case
    • now for your Corista II workflow, you should monitor your cpu and gpu so see if either or both is max'd out; it is possible, I suspect even likely, that the Red Giant software is NOTmaxing out either one. Adobe has worked hard for various MPE components in CS6 to fully utilize all CPU cores and GPU capabilities all at the same time. Third party add-ins, not so much. If neither cpu or gpu is exceeding say 50% during those operations there may not be much you can do other than changing your workflow away from non-MPE filters and effects.

     

     

    Regards,

     

    Jim

     
    |
    Mark as:
  • Currently Being Moderated
    May 27, 2013 6:19 AM   in reply to JEShort01

    One correction, Jim:

     

    The GTX 680 has 1536 CUDA cores, not 448. And the GTX 660 that's in Bill's secondary (i7-2600K) rig has 960 CUDA cores and a memory throughput of 144 GB/s.

     
    |
    Mark as:
  • Currently Being Moderated
    May 27, 2013 6:19 AM   in reply to RjL190365

    Randall (RjL),

     

    Looks like with cross-posted; at least we agree!

     

    Regarding Oli's RAID 0 array, I think he is OK there. In the PPBM6 test which I believe to be a more telling true I/O test (vs. PPBM5), Oli's time of 225 is actually faster than Bill's time of 228 seconds.

     

    Regards,

     

    Jim

     
    |
    Mark as:
  • Currently Being Moderated
    May 27, 2013 6:21 AM   in reply to RjL190365

    My bad on the GTX 680 cores, agree it is indeed 1536!

     

    My own personal GTX 480 has 448 and I misread some of my notes in an Excel speadsheet .

     

    Thanks,

     

    Jim

     
    |
    Mark as:
  • Currently Being Moderated
    May 27, 2013 6:34 AM   in reply to JEShort01

    225 seconds is still on the slow side, considering that I get 180-ish second (about 200 MB/s) scores with a two-disk RAID 0 array of older-generation Seagate 7200.12 1TB hard drives in my secondary rig - and especially slow when compared to the over 270 MB/s (130-ish second) results with the RAID 0'd pair of 1TB Samsung F3 hard drives in my main rig. And even a RAID 0'd pair of 1TB Western Digital Black WD1002FAEX drives still achieved a result of over 240 MB/s in that same PPBM6 disk I/O test.

     

    I think the underperformance of some of the drives in the PPBM6 Disk I/O test tells me one thing: The cache (buffer) memory in such underperforming drives might be on the slow side.

     

    By the way, that 228-second result from one of Bill's systems was with a single drive. (And a fast single spinning drive, a 3TB Seagate 7200.14, at that.)

     
    |
    Mark as:
  • Currently Being Moderated
    May 27, 2013 7:18 AM   in reply to RjL190365

    Regarding Bill's 228 second I/O on the PPBM6 results table, the Speccy listing clearly shows two ST2000DM001 drives (2TB model, additional to 2x 840 pro SSDs).

     

    Bill, could you possibly clarify more test details for your 228 sec. I/O run (PPBM6 Computer ID = BillG SB 660 8GB 3.4)?

     

    Thanks,

     

    Jim

     
    |
    Mark as:
  • Currently Being Moderated
    May 27, 2013 7:24 AM   in reply to JEShort01

    I stand corrected on the drive model. And as I learned just from that Speccy, the two 2TB drives were configured as separate volumes. No RAID involved. Thus, the 228-second result was within one of the two 2TB drives (the second 2TB disk wasn't used at all for anything besides caching in the PPBM6 benchmark).

     
    |
    Mark as:
  • Currently Being Moderated
    May 27, 2013 9:55 AM   in reply to RjL190365

    I am sort of back in the f0rum game-my back is driving me crazy.

     

    Randall I agree that Oliver's disk result are not first class because in my example I got the same Premiere write rate with a single very new generation disk drive (not a RAID0 array).   Also I am not familiar with that motherboard and its SATA ports.

     

    For that 4.6 GHz clock speed, I think that your OS my not be optimized.  At 4.4 GHz I get 260 seconds for the H.264 test, where you 4.6 GHz score is 303 seconds.  While H.264 testing is Heavy CPU usage their is a little assistance from the GPU.

     

    If you can plan ahead for an upgrade it sure looks like  something in the mid 600 series like my GTX 660 (~$200) would help your configuration significantly.

     

    Message was edited by: Bill Gehrke

     

    As you might have figured out I had a message ready to send before the Randall and Jim messages came in.

     

    The one problem we face is that there is no way to indicate in Speccy which drive is the drive being used as the project drive and with my ultra flexible configuration  i7-2600K Sandy Bridge system adds much to the confusion.   Here are the three Sandy Bridge entries with their rank (as of today):

     

    • BillG SB 660 8                #4       Disk I/O=46 seconds      RAID 0 with two each Samsung 840 Pro
    • BillG 660 32 3.9              #5      Disk I/O=118 seconds    RAID 0 with two each  Seagate ST2000DM001
    • BillG SB 660 8GB 3.4   #19      Disk I/O=228 seconds    Single Seagate ST2000DM001

     

    The Disk I/O is fairly independent of the CPU clock rate, and of course completely independent of GPU.   Also I believe that the ST2000DM drives above are the 2-platter versions of this confusing Seagate mess.  You can tell that the last result is a two-platter drive by looking at the Speccy data on the drives.  Looking at the Speccy data on the first drive that is listed with that test you can see that for the serial number it lists "Serial Number        Z1E1ME4V" note that the third digit of the serial number is "E",  those two drives on that test run definitely were both two platter drives.   Three platter drives are just a little bit slower than the two platter versions. 

     
    |
    Mark as:
  • Currently Being Moderated
    May 28, 2013 5:39 AM   in reply to oliewdj

    Oliver,

     

    I just checked the Speccy on your two PPBM6 results. There are a few issues with regards to your rig:

     

    1) It appears that the BIOS version on your motherboard has not been updated at all since that board was shipped after the B3 stepping came into existence. There have been nine BIOS versions that came out since that particular BIOS version (1401) came out. The current BIOS version is 3602.

     

    2) Which version of the Intel Rapid Storage Technology driver are you running? (If you do not have iRST software installed, you must have been using Microsoft Windows-provided software RAID, which is often the cause of such abysmal RAID 0 performance.) Unfortunately, Speccy does not reveal any information at all whatsoever about the SATA or RAID controller or software used.

     

    3) If you have antivirus disabled, there is a chance that your PC might have become infected with malware. Malware infections can significantly slow down overall system performance.

     
    |
    Mark as:
  • Currently Being Moderated
    May 28, 2013 9:49 AM   in reply to oliewdj

    Oliver,

     

    You did not submit your PPBM5 results yet, so the chart here may not be 100% accurate, but shows about where your system is: Oliver's Results

     

    Overall this is not as good as I would have expected. With an overclock to 4.5 GHz I would have expected your scores to be over 50% on Total Time and RPI. Your disk results are disappointing, but you would not be unique in having some problems with the Intel Rapid Storage drivers. It may require testing which version of those drivers give the best results. Bill had similar problems with these drivers on his 1155 platform, where transfer rates were far lower than one would expect.

     

    Another thing I noticed is that you have quite a lot of processes running that are not needed. You may want to check with Process Explorer and kill processes not needed, like acrotray, the Cineform GoPro utility, jusched, daemon, and the like. Ideally you would have less than 50 processes running. Also, have you checked whether there are services you can set to manual according to the BlackViper list?

     

    You did not tell where your previews are located, but I assume on your 128 GB OCZ Agility 3 SSD as is the media cache. This is an older SSD that suffers from serious 'stable state' performance degradation, and unless it has been trimmed very recently may give you no more than 200 MB/s write speeds, often even less.

     

    What can you derive from the chart linked to above?

     

    • Your disk performance is disappointing. It could be caused by the IASTOR driver version, 'stable state' degradation of the Agility 3 SSD, the running Microsoft Security processes and not having defragged your disks recently or a fill rate that is too high and now shows the regular performance degradation on your HDD's.
    • Your MPEG score is relatively good, showing that your GTX 285 is performing nicely, confirmed by your MPE score.
    • Your H.264 score is somewhat disappointing, which indicates that your CPU is hampered by background processes.

     

    On the new PPBM6 test, one sees a clear difference between the GTX 285 and GTX 460 H.264 tests and disk I/O tests, but no difference in the MPEG test. That means the 285 and 460, which are heavily involved in the MPEG test, perform comparably.  The H.264 test where the 460 is clearly faster than with the 285 indicates that the number of cores on the 460 (336 versus 240) and the efficiency of DDR5 versus DDR3 makes the difference when rendering and exporting the GPU compute intensive 'red' bar areas of the timeline (Red 4K).

     

    The differing Disk results are indicative of background processes that get in the way during these tests, because the GPU is not involved at all. It is only disk transfer rates at play here.

     
    |
    Mark as:
  • Currently Being Moderated
    May 29, 2013 5:48 AM   in reply to oliewdj

    p.s not sure why it says 276 for diskio when i ran the test and timed myself it was 2minutes 50 seconds and when i ran the output module it told me the diskio was at 230mb/s which i would expect from this raid0.

     

    When you run the test, the first step in the whole process is creating the destination file. Then the source material is fetched, all the clips are 'glued' together and as a last step, the final results (37 GB) are written to disk. When that writing is completed, Windows adjusts the modified timestamp of the file that was created as the first step. What is measured is the difference between the creation and modified timestamp, which is accurate to 1 second. It may look like the test has already finished, but Windows may still need time to do the last writing on the AVI file and adjust the modified timestamp after all the writing. This delay can depend on the degree of cache buffering that the system can use.

     

    There is one thing I'm not sure about and that is the internal intelligence of PR CS6. In comparison to CS5, where identical clips or sequences were cached, CS6 lost that capability. That is the explanation for the lousy MPEG2 results in PPBM5 for CS6 versus CS5. In this test we use over 1600 identical clips. If the program is smart, it will only read a single clip, but if it is not smart, it will read all 1600+ clips from disk. My feeling is that this is what happens, but I cannot substantiate it.

     

    This means reading 37 GB from disk, gluing them together and then writing one single 37 GB file to disk. What this basically means is that you have 37 GB of 'random' reads and one 37 GB sequential write in this test. However the results are presented as if writing the file is the only activity. The reported transfer rate is only based on the writing speed and disregards the reading speed.

     

    Let me give you an argument for this assumption. This is the HD Tune file benchmark result for my array:

     

    Areca E file benchmark 2GB.png

    The read speed is in excess of 3.5 GB/s and the write speed is almost 4.0 GB/s. Yet my PPBM6 results show only a transfer rate of 1,686 MB/s, less than half of what HD Tune reports, and Crystal DiskMark, AJA and other tests for that matter. Here is the Crystal DiskMark:

     

    CrystalDiskMark E.png

    Again the PPBM6 result is only around half of the reported speed on this test, so my feeling is that the 'dumb' reading of 1600+ identical clips causes these rather disappointing transfer rates in our test, but it is statistically more accurate to use these hard figures from Windows, than applying coefficients to the calculated figures.

     

    I'll have a closer look at your performance charts and let you know what my thoughts are.

     
    |
    Mark as:
  • Currently Being Moderated
    Jun 14, 2013 10:54 AM   in reply to oliewdj

    Without getting too specific, I would say in general 3GB is fine for rotating drives (only the fastest SSDs benefit significantly from 6GB ports) and Intel's RAID firmware/software is better than Marvell's.

     

    Jim

     
    |
    Mark as:
  • Currently Being Moderated
    Jun 14, 2013 12:26 PM   in reply to oliewdj

    oliewdj wrote:

     

    -Barracuda (new ones with redisigned platter for improved performance?) st2000dm001

    -F3 (1TB versions?)  like randall has that gets really good speed only 2TB raided but thats fine.

     

    Oli

    Oli. on the Seagate ST2000DM001 drives.  the performance level depends on where they are manufactured.  The third digit of the serial number must be an "E" (this has been true in past on all my drives).  But even if you have the three platter versions you should get very good performance,  I do not think I would run a mixed set.  So check your S/N's.before you open them  The last drive I received is a two platter version, so maybe yours will be also

     
    |
    Mark as:
  • Currently Being Moderated
    Jun 15, 2013 10:50 AM   in reply to oliewdj

    Sounds, great I just hope Seagate has not changed the system of marking but those are the better ones.

     
    |
    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