Drive Benchmark limiting performance?

[taryngar]

I've been noticing that on my newest computers with PCIe 4.0 SSDs, the Disk Mark tests are not reporting anywhere close to their true performance. It's even more evident when using a PCIe 4.0 RAID0 array of 980 Pro drives, which are pretty much the fastest out there right now. Even the Performance Test Drive Performance section reports much higher values than are assigned as part of the Disk Mark test. I've attached screenshots of both the Disk Mark vs Drive Performance as well as an ATTO Disk Benchmark of the same drive.

For me to be getting lower scores than a baseline 960 EVO when every other benchmark shows that it's transferring just fine (including in the same program) seems mind-boggling. The scores are very similar for a single Sabrent 2TB Rocket 4.0, which should also be much higher than Disk Mark seems to understand. It feels like it's not using a high enough I/O size, and therefore not taking full advantage of the drive's capabilities.

[David (PassMark)]

The manufacturers only advertise the peak performance for a drive (of course). Even if that peak performance is only available in some obscure set of circumstance that bear little relevance to real life. For example, they don't mention the fact that the drive only does a pitiful 75MB/sec in the fairly common 4K random read scenario. RAID with M2 drives doesn't provide much benefit and might even hurt performance in some cases.

You also aren't comparing apples to apples to apples as you have different file sizes, different queue depths and likely difference caching settings. (you might even be measuring RAM cache performance and not drive performance). These settings can be adjust the in the PerformanceTest Advanced Disk Test window.

See also this post
https://forums.passmark.com/pc-hardw...ax-nvme-speeds

[taryngar]

What are the exact settings being used to give me the score of 3029 sequential read and 2971 sequential write used for the Disk Mark calculation score? Those don't seem to match up to any of the numbers being reported in your app's own Drive Performance screen. It's kind of hard to compare apples to apples when I have no idea what apples you're using.

[Richard (PassMark)]

PerformanceTest use a IO queue length of 20 for the sequential tests with block size of 32 KB. File size depends on test and medium used.

Sequential Read: 1 GB for non solid state drives, 2 GB for solid state drives.
Sequential Write: 400 MB for non solid state drives, 800 MB for solid state drives.

More information and test settings for the Standard Disk and Drive Performance settings can be found in the help file found within PerformanceTest.

Windows is asked to turn off the RAM disk cache before the test, but there are a few device drivers (and some RAID systems) that ignore the request and cache to RAM anyway (either main system RAM, or RAM on the disk controller).