Performance consistency tells us a lot about the architecture of these SSDs and how they handle internal fragmentation. The reason we do not have consistent IO latency with SSDs is because inevitably all controllers have to do some amount of defragmentation or garbage collection in order to continue operating at high speeds. When and how an SSD decides to run its defrag or cleanup routines directly impacts the user experience as inconsistent performance results in application slowdowns.
To test IO consistency, we fill a secure erased SSD with sequential data to ensure that all user accessible LBAs (Logical Block Addresses) have data associated with them. Next we kick off a 4KB random write workload across all LBAs at a queue depth of 32 using incompressible data. The test is run for just over half an hour and we record instantaneous IOPS every second.
We are also testing drives with added over-provisioning by limiting the LBA range. This gives us a look into the drive’s behavior with varying levels of empty space, which is frankly a more realistic approach for client workloads.
Each of the three graphs has its own purpose. The first one is of the whole duration of the test in log scale. The second and third one zoom into the beginning of steady-state operation (t=1400s) but on different scales: the second one uses log scale for easy comparison whereas the third one uses linear scale for better visualization of differences between drives. Click the dropdown selections below each graph to switch the source data.
For more detailed description of the test and why performance consistency matters, read our original Intel SSD DC S3700 article.
Default
25% Over-Provisioning
Despite the custom Transcend firmware, performance consistency is an exact match with ADATA's SP610. I'm suspecting that the reason for low steady-state performance might be the hardware because the SM2246EN is a single-core design. Most controller designs today are multicore because today's NAND requires a lot of management and with multiple cores the NAND management can be dedicated to one or more cores, which leaves the rest of the cores available for host IO processing. In Silicon Motion's case, the one core has to take care of everything from host IOs to NAND management, which translates to lower overall performance as the controller can't keep up with everything that needs to be done.
Default
25% Over-Provisioning
Default
25% Over-Provisioning
TRIM Validation
To test TRIM, I filled a 128GB SSD370 with sequential 128KB data and proceeded with a 30-minute random 4KB write (QD32) workload to put the drive into steady-state. After that I TRIM'ed the drive by issuing a quick format in Windows and ran HD Tach to produce the graph below.
Barring catastrophic failures, endurance testing a drive to destruction takes many months. Tech Report started torturing a set of 256GB drives in late 2013; as of last month 2 of the 6 drives were still running.
I guess you don't read Anandtech much. Generally they run down the drive enough to move the counter down a few percent, then make a good estimate of endurance based on those numbers. I think it's very interesting and pretty much only Anandtech does it. Or used to do it.
Maybe it's you that doesn't read AT that much ;) (Haha, I had to)
They typically will only do that when they are testing drives with a new type of NAND that we haven't seen yet before, or testing some weird scenario, or something like that. Micron 20nm NAND is a well known entity at this point, and even though they are using custom firmware on this controller, it's performance and behavior seems very similar to the stock SMI firmware -- so basically there is nothing remarkable here. I am sure the endurance will be similar to most other drives with this type of NAND.
Right on the spot. I only do endurance testing with new NAND generations that we haven't seen before to figure out the P/E cycle rating. It takes days, possibly weeks, so there is no point in testing that with every drive. After all, the manufacturers' ratings still matter because once those are reached the warranty will be voided anyway, so my endurance tests aim to tell more about the NAND than the drive itself.
But wait. Does it take days or weeks to run down the counter 1 or 2 percent? That's all you need to get an estimate on actual endurance right? And isn't there variation in the NAND that each manufacturer buys for each line of drives? I'm talking about the binning and how endurance can vary for the same process.
The problem is finding the exact spot where the counter changes by 1 percent, so it usually takes at least a few percent worth of cycling to find that. Generally it takes a couple of days for client drives, but even then that time is away from testing other drives.
You are correct that not every die is equal, but the P/E cycle rating is usually conservative to guarantee that all SSD-grade dies comply that spec. With binning and parameter trimming it's possible to get much more out of the dies though.
I've fried 3 brand new SSDs in my torrent box over the past year. Granted two were 64GB, one 120GB, and they were pretty meh to begin with performance wise.
I knew the SSD wouldn't last, but didn't expect it to fail that quickly. Currently have an OLD 64GB Intel SLC (X25 I think) in for past few months...no issue yet.
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.
Facebook
Twitter
RSS
44 Comments
View All Comments
Hulk - Tuesday, January 27, 2015 - link
Perhaps I missed it but no mention/testing of endurance? All I see are manufacturer quoted numbers in the table.DanNeely - Tuesday, January 27, 2015 - link
Barring catastrophic failures, endurance testing a drive to destruction takes many months. Tech Report started torturing a set of 256GB drives in late 2013; as of last month 2 of the 6 drives were still running.http://techreport.com/review/27436/the-ssd-enduran...
Hulk - Tuesday, January 27, 2015 - link
I guess you don't read Anandtech much. Generally they run down the drive enough to move the counter down a few percent, then make a good estimate of endurance based on those numbers. I think it's very interesting and pretty much only Anandtech does it. Or used to do it.extide - Tuesday, January 27, 2015 - link
Maybe it's you that doesn't read AT that much ;) (Haha, I had to)They typically will only do that when they are testing drives with a new type of NAND that we haven't seen yet before, or testing some weird scenario, or something like that. Micron 20nm NAND is a well known entity at this point, and even though they are using custom firmware on this controller, it's performance and behavior seems very similar to the stock SMI firmware -- so basically there is nothing remarkable here. I am sure the endurance will be similar to most other drives with this type of NAND.
Kristian Vättö - Tuesday, January 27, 2015 - link
Right on the spot. I only do endurance testing with new NAND generations that we haven't seen before to figure out the P/E cycle rating. It takes days, possibly weeks, so there is no point in testing that with every drive. After all, the manufacturers' ratings still matter because once those are reached the warranty will be voided anyway, so my endurance tests aim to tell more about the NAND than the drive itself.Hulk - Tuesday, January 27, 2015 - link
Okay makes sense.I can admit when I'm wrong.
Hulk - Tuesday, January 27, 2015 - link
But wait. Does it take days or weeks to run down the counter 1 or 2 percent? That's all you need to get an estimate on actual endurance right?And isn't there variation in the NAND that each manufacturer buys for each line of drives? I'm talking about the binning and how endurance can vary for the same process.
Kristian Vättö - Tuesday, January 27, 2015 - link
The problem is finding the exact spot where the counter changes by 1 percent, so it usually takes at least a few percent worth of cycling to find that. Generally it takes a couple of days for client drives, but even then that time is away from testing other drives.You are correct that not every die is equal, but the P/E cycle rating is usually conservative to guarantee that all SSD-grade dies comply that spec. With binning and parameter trimming it's possible to get much more out of the dies though.
Hulk - Tuesday, January 27, 2015 - link
Wow. Thanks for the specific reply.Souka - Wednesday, January 28, 2015 - link
Just thought to share:I've fried 3 brand new SSDs in my torrent box over the past year. Granted two were 64GB, one 120GB, and they were pretty meh to begin with performance wise.
I knew the SSD wouldn't last, but didn't expect it to fail that quickly. Currently have an OLD 64GB Intel SLC (X25 I think) in for past few months...no issue yet.