Any chance the claims will be tested? 750k IOPS sounds nice. At 4k it translates to 3GB/s. Hard to believe. What QD is that, something completely unrealistic (and useless in client systems), like 64? 128?
Would be nice to have just 1 but useful metric for the systems, like random 64k reads at QD2. Because 1) smaller reads don't matter, NTFS runs are 64KiB (16 clusters) each, and with smaller files performance is fast enough as it is. 2) longer sequential reads do not matter much because if 64KiB reads are fast enough, longer sequential reads will be fast enough too, or processing of such large and sequentially placed files (like video) will be the bottleneck and not their reading. 3) writes don't matter for internal drives because with write-back caching the system is usable before physical writes are finished, and client writes almost never exceed file cache sizes - so the user does not actually waits for the writes to finish 4) Higher QDs do not matter because they are relatively rare on client systems.
Client performance index = transfer rate with 64kiB random reads at QD2.
Obscurity and vague claims help sales, particularly when products are actually weaker than companies claim to be. You are asking SSD vendors to be clear and reasonable in the performance metrics they announce. That's quite a big ask..
Are you guys going to review the PCIe 4.0 SSD you probably got in your review kit? I would really like to see how these Phison based PCIe 4.0 SSDs do with your custom benchmarks. I'm thinking they will be fantastic for light workloads and a little bit better than the rest at the heavy workload & the destroyer.
When you have to start putting fancy expensive heatsinks on a SSD its time to rethink the SSD. I really hope people are not buying systems with 4.0 just for something like this that a person won't even notice a difference.
Excellent! I did some testing with an m.2 stick versus the intel card (both PCI3 high end), and the m.2 would clearly throttle when paired with a ramdisk (lol 128GB ram is fun). I'd love you guys to do tests on when this thing throttles versus other m.2 sticks.
Unlikely. There will be no PCIe 5 consumer boards for more than a year. Samsung isn't going to let other NVMe drives run laps around the 970 Pro for that long. Even when PCIe 5.0 is available the issues with power, cost, and lane lengths make it unclear if/when we will see it outside of the datacenter. PCIe 6.0? nobody is waiting that long to refresh their products.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
14 Comments
Back to Article
peevee - Tuesday, July 9, 2019 - link
Any chance the claims will be tested? 750k IOPS sounds nice. At 4k it translates to 3GB/s. Hard to believe. What QD is that, something completely unrealistic (and useless in client systems), like 64? 128?Would be nice to have just 1 but useful metric for the systems, like random 64k reads at QD2.
Because
1) smaller reads don't matter, NTFS runs are 64KiB (16 clusters) each, and with smaller files performance is fast enough as it is.
2) longer sequential reads do not matter much because if 64KiB reads are fast enough, longer sequential reads will be fast enough too, or processing of such large and sequentially placed files (like video) will be the bottleneck and not their reading.
3) writes don't matter for internal drives because with write-back caching the system is usable before physical writes are finished, and client writes almost never exceed file cache sizes - so the user does not actually waits for the writes to finish
4) Higher QDs do not matter because they are relatively rare on client systems.
Client performance index = transfer rate with 64kiB random reads at QD2.
boeush - Tuesday, July 9, 2019 - link
+1Though another important key metric might be read latency: how much time elapses, on average, between a request for data and arrival of that data.
brunis.dk - Wednesday, July 10, 2019 - link
You'll have to wait for the review then :)Santoval - Wednesday, July 10, 2019 - link
Obscurity and vague claims help sales, particularly when products are actually weaker than companies claim to be. You are asking SSD vendors to be clear and reasonable in the performance metrics they announce. That's quite a big ask..Someguyperson - Tuesday, July 9, 2019 - link
Are you guys going to review the PCIe 4.0 SSD you probably got in your review kit? I would really like to see how these Phison based PCIe 4.0 SSDs do with your custom benchmarks. I'm thinking they will be fantastic for light workloads and a little bit better than the rest at the heavy workload & the destroyer.imaheadcase - Tuesday, July 9, 2019 - link
When you have to start putting fancy expensive heatsinks on a SSD its time to rethink the SSD. I really hope people are not buying systems with 4.0 just for something like this that a person won't even notice a difference.Qthulu - Tuesday, July 9, 2019 - link
Whats wven crazier.....ive seen box art for a galax radeon 5700xtPinn - Wednesday, July 10, 2019 - link
Excellent! I did some testing with an m.2 stick versus the intel card (both PCI3 high end), and the m.2 would clearly throttle when paired with a ramdisk (lol 128GB ram is fun). I'd love you guys to do tests on when this thing throttles versus other m.2 sticks.SydneyBlue120d - Wednesday, July 10, 2019 - link
Who is Galax?Is it a reliable brand?
Never heard of it before.
When can we expect PCIe 4.0 SSDs from Samsung?
haukionkannel - Wednesday, July 10, 2019 - link
Who knows... maybe Samsung go directly to pci 5.0? Or 6.0...TheUnhandledException - Wednesday, July 10, 2019 - link
Unlikely. There will be no PCIe 5 consumer boards for more than a year. Samsung isn't going to let other NVMe drives run laps around the 970 Pro for that long. Even when PCIe 5.0 is available the issues with power, cost, and lane lengths make it unclear if/when we will see it outside of the datacenter. PCIe 6.0? nobody is waiting that long to refresh their products.Santoval - Wednesday, July 10, 2019 - link
No word about the QD level of these 4K numbers I see. Which means it must be 64+ big..sofiawatson - Friday, July 12, 2019 - link
Those who come to read your article will find lots of helpful and informative tips. https://quickassignmenthelp.co.uk/michaeljames14 - Friday, July 12, 2019 - link
I just want to thank you to share your information on your blog; this is a simple and useful. https://www.1clickdissertation.co.uk/