Bad Disk Mark Results (Samsung 970 Evo)

[DedPixlz]

My benchmark score in Passmark Software with the 970 EVO is between 7000 and 10,000. The average score is over 14,000. I closed every unnecessary processes running in the background to make sure nothing is writing to the disk I reinstalled Samsung's NVME drivers. I checked my temperatures for throttling. I have overprovisioning and my drive is far from being full. Everything seems fine.

My score looks normal in Samsung Magician's benchmark. It seems Passmark is having troubles especially with my Read speeds. What can be wrong?

[David (PassMark)]

What are the actual benchmark results for each of the tests?

[DedPixlz]

What are the actual benchmark results for each of the tests?

Sequential Read: 1336 (Should be in the 2600's)
Sequential Write: 1038 (Should be around 1450)
IOPS 32KQD20: 549 (Seems somewhat normal, numbers vary a lot in this test for this disk)
IOPS 4KQD1: 66 (Seems somewhat normal but could also be higher)

Disk Mark: 10631

This is one of the best results so far. Sometimes my Sequential Write will drop to 500. I had similar results when I first built my PC, then a few months later I did another benchmark and to my surprise the results were normal, even a bit above average. Then today I tried it again and I am getting bad results again. Previous tests were with Passmark 9.0. The latest tests are with version 10.

I also made sure I am running in PCI-E Gen 3x4 and rebooted several times.

If it can help, this is the rest of my system configuration:

Ryzen 7 2700x on Asrock X470 Taichi
32GB Teamgroup Dark Pro (B-die) @ 3200 14-14-14-30
MSI RTX 2060 Super Gaming X
EVGA Supernova G3 850

EDIT: Just tried CrystalDiskMark, and my read/write speeds are actually above average according to the results. It appears my disk is fine but Passmark Software can't benchmark it correctly for some reason.

[Richard (PassMark)]

Sequential Read/Write will depend on what the benchmarking program specification (File size, block size, number of threads, queue depth, API, buffering & caching) are for the test.

According to Samsung Magician 6 installation guide:

Performance Benchmark

1) Benchmarking results are for a Queue Depth of 32.
2) The numbers of threads used for performance benchmark are: 1 for Sequential and 4 for Random for NVMe device; 1 for Sequential and 1 for Random for SATA device.
...

PerformanceTest use a IO queue length of 20 for the sequential tests with block size of 32 KB. You'll need to find out the what parameters is used by Samsung Magician for testing, then you can recreate their test using Advance Disk Test option in PerformanceTest.

[David (PassMark)]

Also Samsung Magician is a bit of a scam.
https://www.passmark.com/forum/perfo...msung-evo-ssds

[DedPixlz]

Also Samsung Magician is a bit of a scam.
https://www.passmark.com/forum/perfo...msung-evo-ssds

Both CrystalDiskMark and Samsung Magician gave me results that reflect the average speeds for my disk tho (when compared to results from other users with the same NVMe drive). Only Passmark puts my disk way below average for the same model. Do you have an explanation why my disk would benchmark over 30% below average according to your own user benchmark database?

Your explanation for Samsung Magician being a scam pertains to the usage of RAPID. However, the 970 EVO NVMe drives do not support RAPID mode, so this is not the issue here.

I'm not here to bash your benchmark software or claim that it's inferior to others. I am just trying to figure out if there's something wrong with my NVMe drive or if there's simply something off with the benchmark. (When I claimed my results should be XXXX amount, this number is based off of the average Passmark results for the same drive, not advertised speeds or whatever).

[David (PassMark)]

Sequential Read/Write will depend on what the benchmarking program specification (File size, block size, number of threads, queue depth, API, buffering & caching) are used for the test.
PerformanceTest use a IO queue length of 20 for the sequential tests with block size of 32 KB. You'll need to find out the what parameters is used by Samsung Magician for testing, then you can recreate their test using Advanced Disk Test option in PerformanceTest.

You can get a massive difference in results by changing the test conditions slightly. For example writing 512bytes of data at a time is a valid test. So is writing 4MB blocks of data at a time. The throughput in each case will be massively different. But both are still valid tests.

Of course the vendors spend many hours working out the test scenario that gives the best result for each drive. They don't really care if the scenario is realistic or not, as long as it gives the biggest number.

[DedPixlz]

Sequential Read/Write will depend on what the benchmarking program specification (File size, block size, number of threads, queue depth, API, buffering & caching) are used for the test.
PerformanceTest use a IO queue length of 20 for the sequential tests with block size of 32 KB. You'll need to find out the what parameters is used by Samsung Magician for testing, then you can recreate their test using Advance Disk Test option in PerformanceTest.

You can get a massive difference in results by changing the test conditions slightly. For example writing 512bytes of data at a time is a valid test. So is writing 4MB blocks of data at a time. The throughput in each case will be massively different. But both are still valid tests.

Of course the vendors spend many hours working out the test scenario that gives the best result for each drive. They don't really care if the scenario is realistic or not, as long as it gives the biggest number.

I understand the way the test methodology is different between different benchmarking software, but for the average user score in Passmark to be much higher than mine according to your explanation, that would mean that almost every user benchmarked their 970 EVO using custom settings in your Advance Disk Test? It doesn't make any sense...

[David (PassMark)]

Only results from the standard test are displayed on the web site.

So if you are significantly below the average, then there is some issue with your system.

But remember not everyone can be above the average (that wouldn't make any sense). So being slightly below average isn't of any great concern.

[DedPixlz]

Only results from the standard test are displayed on the web site.

So if you are significantly below the average, then there is some issue with your system.

But remember not everyone can be above the average (that wouldn't make any sense). So being slightly below average isn't of any great concern.

Ok, thanks for the insight. I will perform several more tests and check my configuration again to be sure my disk is fine. I am about 30% below average which is a significant margin IMO.