18 Replies Latest reply on Jan 11, 2012 8:32 PM by JEShort01

    Strange results from PPBM5.5... anyone got ideas why?

    davidbeisner2010 Level 3

      Ok, so I just finished running PPBM5 five separate times, each time with a slightly different disk configuration, and here are my results:

       

      • ASUS P6T6 WS Revolution Mobo
      • i7-920 @ 2.67GHz
      • 12 GB Memory
      • GTX-285 Video Card
      • All drives SATA 7200
      • FWIW, no reboot between tests (might explain the slowly growing H.264 test numbers?)

       

       

      Configuration
      Disk I/O Test
      MPEG2-DVD Test
      H.264 Test
      MPE-Off
      MPE-On
      MPE-Gain
      Score

      Normal setup:

      1 3-disk RAID-5 Array on internal controller for Projects/Scratch

      1 4-disk RAID-5 Array on external RocketRaid 231x controller for Data

      1 internal disk for exports

      12117399127914.1111529

      Next three are testing specifically for disk I/O:

      Everything on 3-disk RAID-5 Array on internal controller

      139178100126815.75

      551

      Everything on 4-disk RAID-5 Array on external RocketRaid 231x controller97175101121815.125502
      Everything on internal single disk124173101125815.625531

      Here's the weird one:

      1 3-disk RAID-5 Array on internal controller for Projects/Scratch and export

      1 4-disk RAID-5 Array on external RocketRaid 231x controller for data

      582173102125913.8889991

       

      That last one I actually ran first, and started it running before I realized I had forgotten to switch my export disks. After running the others and noticing how way out of whack the disk i/o test was, I ran it a second time and got a similar result (first run was actually 559 for the disk i/o, second and reported was 582).

       

      So here are my questions:

      1. Why would my disk I/O be so stinking slow when projects and export are on the same array and data is on a second array, when the disk I/O times with everything on each disk are so much lower?
      2. Why would my disk I/O time for everything being on the external RAID be so much lower than it is with everything (projects/scratch, data, exports) being on their own drives? The internal RAID array and single disk tests came in right close to the three-drive number.
      3. Based on what you see here, would you recommend I continue with my normal three-drive setup, or run with everything on the single external array? (My primary output is H.264, and I literally NEVER output avi files. My primary source files are XDCAM-EX and RED 4k.)

       

      Thanks!

        • 1. Re: Strange results from PPBM5.5... anyone got ideas why?
          Harm Millaard Level 7

          David,

           

          First of all, thanks for all your testing. I received all your results, but have not yet had the opportunity to add them to the database.

           

          Let me try to explain the Disk I/O results, but start off with the other results first.

           

          MPE On results are 8 or 9 seconds. What you clearly see here is statistical measurement error. Windows only allows us to measure in discrete seconds. If more accuracy were possible, it is not unthinkable that the actual results here would have been something like 8.73 - 8.41 - 8.39 - 8.46 - 8.52, so I would not worry about these differences, but attribute them to measurement error. BTW, that is the reason that the timelines in PPBM6 will be longer and even more complex, in order to reduce these measurement errors. In my own case when doing this test 10 times, I got 7 times a 3 second score and 3 times a 4 second score and that is why my MPE score is listed as 3.

           

          The same more or less applies to the MPEG2-DVD, H.264-BR and MPE Off tests. Plus or minus one second is statistically the same result.

           

          The Disk I/O test is a completely different matter. Your disk setup is crucial here. Let me put down some sustained transfer rates for your various disks or arrays and please remember that these are rough, indicative figures, but may help you understand what is going on here.

           

          Sustained transfer rates on your various disks are around:

           

          Single disk             100 MB/s

          3 x R5 on-board      140 MB/s (or less)

          4 x R5 RocketR      225 MB/s

           

          So, from that perspective it is no surprise that your best score is from the 4 x R5 RocketRaid and the test from the 3 x R5 On-board is the slowest, because of the CPU overhead that this on-board raid entails. The weird one is of course hampered by the fact that you have the media cache (scratch) and export on that slow internal raid, which puts an extra burden on the CPU for the parity calculations. You see the effect of the CPU overhead clearly in the MPE Off score when you run everything off the RocketRaid.

           

          Last, the weird result may have been caused by background processes, like automatic updates, disk optimizing schedules, and the like.

           

          I know this is a complex matter and probably my explanation does not answer all your questions, but at least you can't say I did not try.

          1 person found this helpful
          • 2. Re: Strange results from PPBM5.5... anyone got ideas why?
            davidbeisner2010 Level 3

            Thanks a ton, Harm... that actually helps a lot. My curiosity, though, is why test 5 had a 580 disk speed, while test 2 had a 138, where EVERYTHING was on that one disk? It would seem to me that at least not having the data on that disk would help the disk i/o, and therefore it should be faster than test 2... Or is test 2 faster because the CPU doesn't have to handle moving things between multiple disks/raids, while on test 5, it has to move data between two different raids? But if that were the case, why wouldn't test 1 also be slow since data is moving between two raids and a third standalone disk?

             

            Also, please note the comments I replied on my data via e-mail before you post it... one of the submissions had an error in the raid submission, and the first one I threw out entirely...

             

            Also, can you explain why the 4 disk raid has such a higher data rate than the 3 disk raid? Is it purely due to the added disks?? And if so, would adding a 4th disk to my on board R5 speed it up in a similar fashion? Because I could take the 1TB standalone for exports and add it to that on-board R5... Or, should I do as I saw recommended in another post this morning, and take those four onboard disks and recreate them as an R10, maintaining redundancy but also gaining speed? Would losing that dedicated export disk not be worth the added disk in the R5 and/or the ability to create an R10?

            • 3. Re: Strange results from PPBM5.5... anyone got ideas why?
              Bill Gehrke Adobe Community Professional & MVP

              David, with CS5.5 running with hardware acceleration (MPE-On) on  a GTX 285 is like (pardon me Harm) wondering why your speed boat with the anchor overboard is going so slow.  My testing has show that while the GTX 285 and GTX 275 were of value with CS5, they are practically useless with CS5.5 especially if you are overclocking your CPU.  You are running an i7-920 at stock speed but you might try running PPBM5.5 with MPE software only  Notice my SB i7-2600K results below and compare the MPEG2-DVD results with and without GPU assistance.

               

              Despite all the expert opinions on multiple disk drives it turns out that all my x58 scores are with essentially 3 drives/arrays.  This system has room for 18 removable drives

              1. OS/Applications/paging--1 drive
              2. Big RAID 0 for everything project-wise--8 drives (I now admit this is overkill but I had to prove it to myself)
              3. Backup/archiving multiple drives

              GPU-test-PPBM5.5-new.jpg

              • 4. Re: Strange results from PPBM5.5... anyone got ideas why?
                davidbeisner2010 Level 3

                Interesting, thanks for the info, Bill. How would I set AME to not use the GPU so I could pull numbers to test my own setup?

                 

                Regardless of what the numbers show, simple usability says a lot, and I've found over the last several days that with two-three layers of RED 4k footage on the timeline, with CC applied (RGB CC and Three-Way CC are my go-to effects), and the playback resolution set to 1/4 (lowest available for me), I cannot play or scrub smoothly without MPE hardware turned on. I turn it on, and I can play and scrub perfectly smoothly, though not above 1/4 resolution.

                 

                Unfortunately I work for a rather penny-starved non-profit educational institution, and I can't just requisition a 570 or 580... I have to put it as a line-item in my annual budget request, and maybe it will be approved... I've already over-spent my budget by nearly $3000 this fiscal year due to changes that had to happen to support the Epic footage, and as long as my system is usable, I'll not be likely to get another penny until next fiscal year (July 1). So I'll have to live with what I have for now. I did notice, though, that the three things that seemed to really pick up the scores for folks with my proc and mobo, were the GPU, the memory, and the OS drive speed... so a 10,000 rpm drive or an SSD for OS/Programs will also be in the budget request for next fiscal year, along with 24GB of memory.

                • 5. Re: Strange results from PPBM5.5... anyone got ideas why?
                  RjL190365 Level 4

                  Bill,

                   

                  Also note that the MRQ is turned off by default with MPE Off. On my overclocked i7-2600K system @ 4.6GHz, I got an MPEG-2 DVD result of around 110 seconds with my current GTX 470 - but with MPE Off (and thus no GPU acceleration), it produced an MPEG-2 DVD result of 115 seconds with MRQ off. However, my tests with MRQ turned on produced a much slower result: 850 seconds for MPEG-2 DVD. In my system, turning off MPE also increased the H.264 times, as well: It increased from around 61 seconds with GPU acceleration to 106 seconds without GPU acceleration and MRQ off, and a whopping 300 seconds with MRQ on. But the PPBM5 instructions call for leaving MRQ turned off, so the GTX 470 level is the absolute minimum GPU that's required for an i7-2600K overclocked to 4.5GHz in order for the MPE performance to be "faster" than software-only performance.

                   

                  Note that this applies only to my system, and not necessarily to others. And different CPUs have different minimum GPU requirements in order for a good balance.

                  • 6. Re: Strange results from PPBM5.5... anyone got ideas why?
                    JEShort01 Level 4

                    David,

                     

                    For 3 layer of RED, you are likely being most hammered by your CPU. You can validate this by calling up CPU, GPU monitor during your work and note when you maxing you CPU cores out.

                     

                    You talk about budget issues; invest the pennies involved to get a CoolerMaster 212 cooler and overclock your 920 to at least 3.7GHz. Done right, this will be WAY faster for what you are trying to do and will not fry your CPU. At 3.7 with a good cooler, your CPU will still likely be way cooler than your GTX 285!

                     

                    Also for scrubbing and playback, try from the menu sequence / render entire work area. While this takes time to render, I find this an essential step to work with 4-layer RED 4k on my pretty fast 6-core GTX 480, 24GB system.

                     

                    Jim

                    1 person found this helpful
                    • 7. Re: Strange results from PPBM5.5... anyone got ideas why?
                      davidbeisner2010 Level 3

                      Thanks for the suggestions, Jim!

                       

                       

                       

                      Do you have any more advice on overclocking? Back when I first got into

                      computers in the late ‘90s I used to OC my processors (I was an AMD man

                      then) and I ended up destroying two procs because they overheated… and that

                      was with water cooling! I’d love to be able to OC this thing and get more

                      juice out of it, but I’m honestly a little scared to. I don’t know much

                      about OC’ing a machine, and all the different things that have to get

                      played with (I know there’s all sorts of memory timings etc that have to

                      change too).

                       

                       

                       

                      Is there a good OC’ing program out there that will do it for you? Or is it

                      still a game of testing and pushing, one step at a time, until the machine

                      becomes unstable?

                       

                       

                       

                      From: JEShort01 forums@adobe.com

                      Sent: Wednesday, January 11, 2012 7:19 AM

                      To: davidbeisner2010

                      Subject: Strange results from PPBM5.5... anyone got

                      ideas why?

                       

                       

                      Re: Strange results from PPBM5.5... anyone got ideas why?

                       

                      created by JEShort01 <http://forums.adobe.com/people/JEShort01> in *Hardware

                      Forum* - View the full

                      discussion<http://forums.adobe.com/message/4132011#4132011

                      • 8. Re: Strange results from PPBM5.5... anyone got ideas why?
                        Harm Millaard Level 7

                        David,

                         

                        See: Overclocking the i7, a beginners guide

                         

                        The motherboard supplied utilities for overclocking have a huge tendency to increase voltages beyond what one would consider acceptable, thereby decreasing life expectancy. Good overclocking is a time consuming and manual endeavor.

                        • 9. Re: Strange results from PPBM5.5... anyone got ideas why?
                          davidbeisner2010 Level 3

                          Thanks, Harm! That's a very helpful guide...

                           

                          I went ahead and downloaded the system information tools and Prime, went into the BIOS and disabled all the power saving stuff like it said, and I'm running the initial Prime test right now to see where my stock numbers lie. Even in turning off the EIST and setting the multiplier to 21x (had been Auto), I've seen a jump from 2.65GHz to 2.81GHz, so even without doing much I've seen a little bit of a speed jump.

                           

                          My temperatures after 20 minutes under the Prime load have gone from 43-45C per core up to 81-86C per core, leaving me only 13-20C left before I hit the TJ Max... I'm going to go ahead and quit Prime, reset the temp stats, and run the RealTemp program while I edit to see what my standard workload does to the temps... if they stay in this range, I'll wait to go any further until I can buy an aftermarket cooler (the cooler in there is a stock Intel heatsink and fan) as I think the temps I'm seeing under the Prime load are too high already. If my normal editing pushes the temps up that high, I'll probably go ahead and put everything back to default (multiplier and EIST) until I can get a better cooling fan.

                           

                          I looked at that CoolerMaster 212 and it looks pretty good... huge thing, though, and I'm hoping it'll fit in my case! It's a big case, but it's also a big heatsink and fan!

                           

                          Quick question... what's the recommended tool for watching the temp and voltages from the GPU?

                           

                          Also, I think the memory I have now is 1333, but how would I know for sure? I looked on the label on the memory stick, and it didn't say... is there somewhere in the system information I can find out what it is? The memory tab of the CPU-Z program just says it's DDR3 with an NB Frequency of 2138.2 and a DRAM frequency of 534.5.... I don't really know what any of that means, or how to deduce anything about the speed of the memory from that information.

                           

                          Thanks!

                          • 10. Re: Strange results from PPBM5.5... anyone got ideas why?
                            Harm Millaard Level 7

                            My temperatures after 20 minutes under the Prime load have gone from 43-45C per core up to 81-86C per core, leaving me only 13-20C left before I hit the TJ Max...

                             

                            That is pretty high, David. Using the torture test for 6 hours, my (OC) 3.7GHz i7-920 reaches only around 75 degrees with ambient temperatures of 21 - 22 degrees. Under load while encoding my temps remain below 67 degrees. The Cooler Master H212 is pretty decently sized. The Noctua D14 is even bigger!!! Here are some of my favorite tools: Adobe Forums: My favorite tools

                             

                            DRAM frequency at 534.5 means you are running memory at default speed, 1066. The SPD tab in CPU-Z can give you info on the rating of your RAM modules. If in doubt, just post a screenshot from the various CPU-Z screens. What does the label say on the modules?

                            • 11. Re: Strange results from PPBM5.5... anyone got ideas why?
                              davidbeisner2010 Level 3

                              I just rendered a multi-layer RED Epic timeline with some heavy CC, and my temps after about 12 minutes of rendering peaked in the upper 70's... looks like I'll definitely be getting a better cooler!

                               

                              The SPD tab has the following info:

                               

                              Module size: 2048MB

                              Max Bandwidth PC3-10700H (667 MHz)

                              then the timings table has a ton more info that I can post if you need it... The label on the memory just gives the part number (which returned no hits on Google) and a barcode... not much info.

                               

                              A quick google search for PC3-1070 reveals that it is indeed what is known as 1333 memory, though apparently manufacturers are a little misleading in their labeling info, as 1333 refers to the MT/s data rate, not the MHz speed. However, since I have a triple-channel memory system, would the 667 MHz be tripled to give me an effective 2000MHz speed? And since you're only as good as your weakest link, is there really any point to OC'ing my system with memory at only 2GHz (already slower than my proc)? I really don't know much about the ins and outs in the hardware world

                              • 12. Re: Strange results from PPBM5.5... anyone got ideas why?
                                Bill Gehrke Adobe Community Professional & MVP

                                David,

                                 

                                If your idle temps are at 43°-45° C (in a typical 20°+ C ambient) that probably indicates that your current heatsink is not properly mounted, or really lousy.  Here are my three favorite free tools:

                                 

                                To answer you other question on how to turn off GPU Hardware Acceleration, it is in Premiere under the Project/Project Settings/ General.

                                • 13. Re: Strange results from PPBM5.5... anyone got ideas why?
                                  davidbeisner2010 Level 3

                                  Thanks Bill. That GPU-Z program is exactly what I was looking for.

                                   

                                  I just played through a sequence using CUDA in PrPro, with both CPU-Z and GPU-Z open, and found that my GPU load never peaked 10%, while my CPU load hovered around 98%, which I guess implies that the CPU is my hangup... at least with this RED footage, anyway.

                                   

                                  I'm going to put in a request for the cooler which will allow me to overclock a bit and hopefully balance things out a little better. Per my above post, is it worthwhile overclocking with my current memory?

                                  • 14. Re: Strange results from PPBM5.5... anyone got ideas why?
                                    davidbeisner2010 Level 3

                                    Also, per Harm's PSU calculator, with my current configuration (non OC'd), I need 574 watts of PSU... I currently have a 700 Watt PSU. Can I safely OC with 125 Watts of overhead in my PSU, or should I hold off OC'ing until I can get a 1000 Watt PSU?

                                    • 15. Re: Strange results from PPBM5.5... anyone got ideas why?
                                      JEShort01 Level 4

                                      David,

                                       

                                      I use MSI Afterburner (my GTX cards are all MSI) and used to use EVGA's Precision GTX tuning application to set GPU speed, monitor temperatures, etc.

                                       

                                      For CPU temperatur monitoring, voltage monitoring, etc. I use HWMonitor Pro from the cpu-Z people and love it! I think that it monitors GPU temperature too, but I prefer the trending and other features of the Afterburner / EVGA Precision GTX tuning applications.

                                       

                                      Jim

                                      • 16. Re: Strange results from PPBM5.5... anyone got ideas why?
                                        Bill Gehrke Adobe Community Professional & MVP

                                        davidbeisner2010 wrote:

                                         

                                        Thanks Bill. That GPU-Z program is exactly what I was looking for.

                                         

                                        I just played through a sequence using CUDA in PrPro, with both CPU-Z and GPU-Z open, and found that my GPU load never peaked 10%, while my CPU load hovered around 98%, which I guess implies that the CPU is my hangup... at least with this RED footage, anyway.

                                         

                                        You know that is something I never tried looking at playing the timeline, guess I will chave to try it while observing the GPU/CPU utilization.  I only have looked at it while encoding.

                                        • 17. Re: Strange results from PPBM5.5... anyone got ideas why?
                                          Bill Gehrke Adobe Community Professional & MVP

                                          Well I tried it and guess what?  No GPU utilization on playback, and I verified that MPE hardware was enabled (yellow line)..  But instead of your very high CPU utilization with my AVCHD 1920 x 1080 single timeline with Fast Color Correction, Full Resolution Playback and running my i7-2600K at 5.0 GHz (ok rather extreme but that was they way I left it after the last test) the CPU utilization was less than 10%.  When I switched to Software MPE (red line) the CPU utilization did go to about double or ~20%.  I guess I do not understand what is going on..I do not have any Red high resolution data to play with, I do have uncompressed 1920 x 1080 pixel 10-bit 4:2:2 files.

                                          • 18. Re: Strange results from PPBM5.5... anyone got ideas why?
                                            JEShort01 Level 4

                                            Bill,

                                             

                                            You can check out the following link for two things: links to free downloadable Red clips and my comments (reply #7) regarding CPU usage for Red playback. My overclocked 6-core was hitting 100% for un-rendered 4-layer playback at 1/4 resolution!!! RED media is so very intense, and also very beautiful.

                                             

                                            http://forums.adobe.com/message/3826526#3826526

                                             

                                            Jim