Announcement

Collapse
No announcement yet.

Disk test doesn't max NVME speeds.

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #16
    1s sampling:
    Click image for larger version

Name:	Drive Performance 1s.png
Views:	250
Size:	35.6 KB
ID:	53086
    Interesting. I did only change the sampling time and the Throughput Read/Write did increase considerably. Artefact?

    Comment


    • #17
      Some more values. The model average value is 44.462, I get 11% less performance. Sequential read I am above the average, all other values are below. Maybe I can switch off some more services without impacting my usage of the computer. But it does not make sense to switch all services off, many are required anyhow.
      Take home message: Secure Boot off has a big performance penalty. Runing the empty NVMe as non-system drive has no impact on performance. Some services have surprisingly an impact on disk performance.
      Disk Mark Setting used---------------------------------
      38.106 Start to compare, System Drive, Secure Boot on, CSM off, UEFI only
      33.304 System Drive, Secure Boot off, CSM off, UEFI only
      34.127 Non-System Drive E with 1.2 TB data, Secure Boot on, CSM off, UEFI only
      37.892 Non-System Drive E completely empty, Secure Boot on, CSM off, UEFI only
      37.165 Cloned again, System Drive, Secure Boot on, CSM off, UEFI only
      39.389 System Drive, Secure Boot on, CSM off, UEFI only, stopped all ASUS update services

      Comment


      • #18
        When you are looking at small performance differences, it is very easy to see patterns in what are actually random variations.
        (and by random I don't really mean random in the mathematical sense, I mean small variations due to external factors that to a large extent are out of your control and not predictable. So deterministic but so complex as to appear random)

        So if you are looking a 6 different scenarios of which some have 0.5% variation, you really need to be running lots of samples. Maybe 5 per scenario. So 30 runs in total. Then either take the max or the median.

        Stopped all ASUS update services
        Some of the ASUS Bloatware is pretty awful.
        Some of it also required a clean install of Windows to remove.

        Secure Boot off
        While we have never looked at it detail, we aren't aware of any difference in disk performance caused by secure boot, at least not after the machine has fully booted. (except a possible link to full disk encryption). I couldn't find anyone on the internet with this issue either. Maybe there is some interaction with other software (e.g. anti-virus scanning) when secure boot is on / off. It would be a bit of a shock to the whole world if this was reproducible on multiple machines.

        Throughput Write of 5463MB/sec
        This is an excellent result for this test scenario. Unheard of just a couple of years ago.

        Comment


        • #19
          This is in the Help file, but I will post it here as well.

          These are the details for each test scenario from the Drive Performance advanced test window (screen shots above)

          Throughput write only has a queue length of 1 (just like 99% of software applications). So it won't get the same performance as a deep queue with multiple threads.
          All test in this group are also single threaded.

          Click image for larger version

Name:	Drive-Performance-settings.png
Views:	237
Size:	14.7 KB
ID:	53090

          Comment


          • #20
            So if you are looking a 6 different scenarios of which some have 0.5% variation, you really need to be running lots of samples. Maybe 5 per scenario. So 30 runs in total. Then either take the max or the median.
            Fully agree. At least in some cases I re-run some single tests multiple times without any signifcant change on data. And the first two data are 8,7% apart. So I assume this is already significant.
            And running the new disk in addition to SATA-SSD seems at first to be negative, too. But with the empty disk attached, the values are getting close to the original (1 versus 4).

            Some of it also required a clean install of Windows to remove.
            See my results here

            This is an excellent result for this test scenario. Unheard of just a couple of years ago.
            I know. But my posting here was solely due to the fact that my Disk Mark was way below the average reported here. Many users are already on Windows 11, which may have an effect - or not.

            Comment


            • #21
              Secure Boot off
              Was a little tricky to find that again because I first tried only searches in english. But that one in in german.
              use google translate: https://answers.microsoft.com/de-de/...d-606a8e65ec93

              The combination of CSM off and Secure Boot set to Windows as a booster I have seen many times in general. Maybe I need to do more measurements. But also others could try this out and maybe there is something to it.

              Comment


              • #22
                Many users are already on Windows 11, which may have an effect - or not.
                Windows 11 had / has a bunch of performance issues. Some are fixed with patches some not.

                There are 4 pages of discussion at the moment.
                https://forums.passmark.com/performa...-varied-wildly

                Comment


                • #23
                  This shows the drive is under load and not getting its score from ram, if we are to go by the images you posted. Why can't the regular sequential test just do the same test?

                  Click image for larger version

Name:	passmark advanced vs stock test.jpg
Views:	243
Size:	262.9 KB
ID:	54276Click image for larger version

Name:	next.jpg
Views:	223
Size:	84.5 KB
ID:	54277
                  Attached Files
                  Last edited by flyercan; Jan-08-2023, 08:48 AM.

                  Comment


                  • #24
                    There isn't just one way to access a hard drive. There are many different options. Including API in use, synchronous or asynchronous access, queue length, block size, reading or writing, sequential or random access, cluster size, disk format, caching or uncached, compressible data or uncompressible, test length (for max temperatures and trim), etc...

                    When you put all these options together, there are a huge number of permutations. The standard tests were picked as they seemed a reasonable collection of options that are in common use and do a reasonable job of representing the real life performance of the drive.

                    We didn't pick a set of options just to make marketing claims.

                    Also the optimal set of options might not be the same for all drives.

                    But as you have noted you can use the Advanced tests to explore the performance in a wide range of scenarios.
                    Also for PerformanceTest V11 we are going to add a few more options to the advanced disk test Drive Performance window to get bigger numbers. As we do acknowledge that this is what people what to see.

                    Comment

                    Working...
                    X