5 Replies Latest reply on Apr 17, 2013 9:36 AM by J. M. Oden

    PPBM5 Disk Layout Question

    J. M. Oden

      Greetings all,

       

      First - a big thank you to everyone who has taken time to document and post information about their computer builds and optimizations.

       

      I have just completed a build that is currently at #154 on the CS6 list and I'd like to see if further optimizations are possible.

       

      Current relevant stats are:

       

      ASUS P9X79 WS Motherboard

      Core i7 3930K (stock at time of PPBM5 submission, 3.8 OC now)

      32GB Corsair Dominator GT RAM

      GTX 680 4GB (GeForce)

      Samsung 840 SSD (OS & Programs) 6Gbps SATA - C:

      WD Veloc 10K on 6 Gbps SATA - D:

      WD Veloc 10K (x2) on Marvell RAID 0 - E:

       

      Seasonic 1200w Gold PS

      Caselabs Magnum 8 Case

      Noctua NH-D14 CPU Cooler

      Noctura 120mm Case Fans (2 in, 1 out)

       

      First Question - When I ran PPBM5, I only had the OS SSD Drive C: and one WD Velociraptor 10k Drive D:, so I put the project on D: and followed the instructions and submitted the result. Now that I have the RAID 0 Drive E:, I don't see how, per the benchmark instructions, to take advantage of the multiple drives that I have by splitting up media, scratch, export, etc. It looks like everything must reside on one drive, or the VB script that collects the stats can't find what it needs.

       

      Secondly, when I ran the benchmark again, I moved everything over to the RAID 0 Drive E: and it didn't seem to improve the results at all. Per CrystalDiskMark, I'm getting about 190MB/s Read & Write on the Single WD Veloc 10K (Drive D:). CrystalDiskMark shows 357MB/s Read but only 175 MB/s Write on the RAID0 Drive E:. Is this asymetrical Read/Write speed normal, or am I missing something obvious in my setup?

       

      Any suggestions or tips to get this PPr noob on the right track would be greatly appreciated.

       

      Thanks again for everyone who contributes to this body of knowledge. And a special thanks to Harm and Bill for putting together the benchmarks.

       

      John

       

      Edited to add: I am aware of the inherit risks of RAID0. This system is not in "production" yet, and if the entire box went up in smoke, I have everything backed up in multiple other places. If the collective wisdom indicates that I don't have enough disks to make RAID worthwhile at this point, I'm perfectly happy to ditch the RAID and use the disks individually. In either case, I'm quite particular about backing up critical data and have procedures in place to make sure everything is protected.

       


        • 1. Re: PPBM5 Disk Layout Question
          Harm Millaard Level 7

          John,

           

          To be accurate, in the overall ranking your system is at rank # 154, but in the CS6 ranking, your system is at rank # 36.

           

          It is a nicely balanced system. See below:

           

          John Oden.png

          All the scores are around the upper quartile, Q3 and - not surprisingly - the Disk test is the relative weakest link, but of course that can be attributed to the single disk when testing. The cheapest and quickest way to further improve your results is by overclocking. Maybe you can reduce some overhead by diligent tuning, setting unneeded services to manual according to the Blackviper list and reducing processes at startup, but nothing more comes to mind with your hardware, unless you are willing to spend a lot more on disks and a dedicated raid controller.

           

          You have a well performing and balanced system that ranks #3 among non-overclocked systems.

           

          Per CrystalDiskMark, I'm getting about 190MB/s Read & Write on the Single WD Veloc 10K (Drive D:). CrystalDiskMark shows 357MB/s Read but only 175 MB/s Write on the RAID0 Drive E:. Is this asymetrical Read/Write speed normal, or am I missing something obvious in my setup?

           

          I can only guess that this is caused by the rapid depletion of the cache memory on the disks, causing the write speed to reflect more of its real physical speed. It could also be influenced by fragmentation of the disks. When I use Crystal Disk Mark on my array, I get results like:

           

          CrystalDiskMark E.png

          In my case I have, next to the 64 MB cache on each disk (21 of them), 4  GB of dedicated cache on the controller. That makes a difference.

           

          BTW, how do you like the Magnum M8? I have the TH10 and consider it the best chassis I have ever seen. Like to hear your opinion about it. You may have already seen this, but if not, look at Intro Part 1

          1 person found this helpful
          • 2. Re: PPBM5 Disk Layout Question
            J. M. Oden Level 1

            Harm,

             

            Thank you for taking time to look over my configuration. Its great to get confirmation that I am at least in the ballpark. My new system is very stable and feels very quick, so I'm looking forward to putting it to use with Premier Pro, Photoshop and LightRoom.

             

            Since my first run at PPBM5, I've applied a minor 3.8 OC which I will bump up a little bit to maybe 4.0. I"ve also ordered another 256MB SSD and will rely on that to boost my disk throughput until I can get a proper RAID card ordered and installed. I'll submit another benchmark run when I've gotten the new SSD installed and the OC configured.

             

            As you can probably tell based on the components I've selected, I have been heavily influenced by your build. In fact, it was only by reading your build writeup that I learned about Case Labs. I can't praise Case Labs and the M8 highly enough. This chassis is superior to anything I've ever seen this side of a rack-mount server case. I am thrilled with the M8 and would highly reccomend Case Labs to anyone that is interested in a new chassis for their build. Its definitely worth the extra investment.

             

            An aside for anyone who might be considering the Samsung 840 SSD: Be sure and get the 840 SSD Pro. I made the mistake of ordering the non-pro version and while its read speed is on the mark at around 500 MB/s, its write speed is only half that (~250 MB/s) due to the NAND technology it uses. I have my Samsung 840 SSD deployed as the C: OS/Boot drive, so the slower write speed isn't much of an issue, but even so, the warranty on the Pro model is longer and if you intend to use your OS/Boot drive as a scratch drive also, you'll definitely want the Pro. If you intend to use your drive for anything that would be affected by write speed, you'll absolutely want the Pro model.

             

            Thanks again,

            John

            • 3. Re: PPBM5 Disk Layout Question
              Harm Millaard Level 7

              John,

               

              As promised, your (OC 3.8) results have been published.  I have made a similar chart as yesterday, but to show you what the overclocking has done to system performance, I have included the rankings with stock speed, shown with the yellow shorter dial and the new overclocked results with the longer grey/white dial, so you can see in a glance where the differences are. Values displayed are for the OC system.

               

              John Oden-OC.png

              As you can see, the major gains from overclocking are in the CPU intensive tests, MPEG and H.264. The deterioration in Disk performance is not due to clock speed, but is mainly dependent on fragmentation of your disk and background processes.

               

              Even this slight overclocking brought your systems performance from barely Q3 to a solid Q3 contestant.

              • 4. Re: PPBM5 Disk Layout Question
                Bill Gehrke Most Valuable Participant

                Harm Millaard wrote:

                 

                John,

                 

                The deterioration in Disk performance is not due to clock speed, but is mainly dependent on fragmentation of your disk and background processes.

                 

                I would just like to add a little more information to Harm's statement above.  For that test you are dealing with a rotating mechanical device you have to expect minor changes from run to run.  Many times numbers that you have seen from me are averages of up to 10 runs.  I have even seen variability when writing to a SSD device.

                 

                With our new benchmark since some of the tests do take longer to run, I will have to cut back to fewer runs but still do some averaging.

                 

                I do like Harm's new "speedometer" presentation, it really shows results very well.

                • 5. Re: PPBM5 Disk Layout Question
                  J. M. Oden Level 1

                  Thanks Bill. Your point about averaging runs is well taken. As I continue to try various optimizations, I'll make multiple runs before drawing any conclusions.

                   

                  Regarding the Disk I/O on my last run - My Disk I/O score actually dropped by a few seconds when I attempted to optimize disk throughput by moving things around to different drives. Is this even a valid approach? I notice that the benchmark submit form allows for OS, project disk and an output disk, but it wasn't exactly clear how to accomplish that layout when running the benchmark.

                   

                  What I attempted was to use the AE output path to direct the output files to another location on my output disk, and then move them back into the project folder after the run, so the script could find the files and collect the stats. Is this a valid approach?

                   

                  After doing the above, I found that my Disk I/O score actually dropped by a few seconds. The score was a little better with scratch, project and output all directed to the same drive.

                   

                  Harm & Bill - big thanks again to you both for taking time to help me with these optimizations.

                   

                  John