Hi, Can someone (probably from guru3D) tell me a bit more about the details of these tests? They run two or three: - File Copy: either a 100 GB compressed file (compressed how?) or else 38 GB compressed MKV file There's only a small image of the test without any axis details. Perhaps Guru3D can post the actual full details of the tests? In Dutch or Polish is OK. - Performance ISO Creation: cannot see any details I'm trying to compare results from the various benchmarks that are run by Reviewers. When I see how to upload an image, I can give ran example of where I've plotted average file size vs normalised transfer speed. You can see that when average file size exceeds about 10MB, the rate tends to jump, but also the spread in the rates goes up a lot more. As in there are other factors that dramatically affect transfer speed. But by comparison, below 10MB, most drives behave much more similarly. What I need to do my analysis is: (1) Test Files Used: - Total Size of the Test Load (Total of all Files) - Number of files - Ideally file size distribution but an average and standard deviation would be OK - Type of files used (2) Procedure: - Run from what disk, and enough detail to see any bottleneck - System used to run it (3) Time to Run: how long it runs so I can see if things like pSLC size or thermal throttling affects the results (4) Details on SSD: - How full it is before the test - How it was formatted (like how much additional over provisioning there was) - Age of SSD (new or used?) I think that many Reviewers have one test SSD that started out new, but gets used quite intensively in some tests, and maybe it is affected by age in some cases. It's something that should be reported for each test: total cumulative data written to the SSD. Thanks very much, Alan
Thanks for the speedy reply. There's one that they've done for quite a few years called "File Copy". https://www.guru3d.com/articles_pages/toshiba_rc500_500gb_nvme_m_2_ssd_review,7.html It has changed a lot, and it even looks like different Reviewers run it differently, but maybe that's not the case. I suspect they use an "off-the-shelf" benchmark program, and just run their own data file in it. But they don't report many details. All you can tell is that they use a "compressed" file of a certain size: which has changed from 38GB to 68GB to 100GB to 110GB. And OK, for some of them they say it's an MKV file, which is indeed a compressed file to some degree. But that's not a very exact description of of the degree of compression. Especially not if you look at the other things that have been varied. I haven't checked the correlation yet, but they've also varied the SSD drive they use to transfer the file from. And I think the CPU/motherboard that it's run on has changed too. And I cannot see where they get their values from that they use to describe what happens. All they show is one or two screenshots of the test, partway through. And then some fairly vague, and frequently repeated comments about what has happened. And for the ones I've checked, the values seem wrong. And the overall analysis seems to miss some interesting things that the drives do during the test. Things that a Reviewer should pick up on, and look into. A few examples: https://www.guru3d.com/articles_pages/samsung_850_evo_m_2_and_msata_ssd_review,9.html These tests show a very big difference between two SSD's that only vary due to size. I think a Reviewer who wants to show they actually thought about what happened should have mentioned what the big drop in the 120GB drive was caused by. And then why the other drive did not see it. https://www.guru3d.com/articles_pages/ocz_vector_180_480_and_960gb_ssd_review,10.html And this, seriously??? "That number will fluctuate a bit here and there during the write process, but was pretty consistent. Yeah -- you can't complain here." Hmmm.....so THIS is "consistent"?? Actually I suspect that the test was not actually run on both drives, but rather just one the one, and then duplicated. Again, I think someone who wants to say they have some expertise in computing hardware needs to at least notice things like this. https://www.guru3d.com/articles_pages/sandisk_ultra_ii_240_gb_ssd_review,9.html Another one. Like what was the Reviewer inhaling when they wrote this: "As you can see the result starts at ~ 450 MB/sec writes, yet after a few secondss drops to 220 MB/sec. This is the culprit for TLC NAND. Read performance however is exellent." None of these graphs have any values on the axes, but I don't think the change from 450MB/sec to 220MB/sec happened "after a few seconds". It's about 1/3 of the way down through the 66% completed screen dump of a 38 GB data write at about 220-240MB/sec. So that seems to be closer to 60 seconds than a "a few". And 60 seconds is a time that some drives suffer thermal throttling. Is this what occurred? Anyways, there's a lot more. But without any actual details of what was done, it's hard to get much value from these tests.
