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
The 850 EVO presents a healthy increase in IO consistency. The 840 EVO wasn't exactly inconsistent in the first place, but the 850 EVO takes the steady-state IOPS from ~3,000-5,000 IOPS to 5,000-8,000 IOPS, which is actually nearly on par with the 850 Pro. The 850 EVO has without a doubt one of the highest performance consistencies out of the value/mainstream drives we have tested.
Default
25% Over-Provisioning
Default
25% Over-Provisioning
TRIM Validation
To test TRIM, I filled a 120GB 850 EVO 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.
I'm in complete agreement. I'm not going to buy another 2.5" SSD before I can get an 850 Pro (512 GB) equivalent with an M.2 interface. I have two motherboards with empty M.2 slots waiting for the market to catch up. I know about the XP941, but the pricing isn't great. M.2 drives should cost almost the same as a 2.5" drive...especially after eliminating the worthless metal/plastic box.
Seems to me like TLC will become the standard mainstream for 3D/VNAND, where as MLC will be pretty much only for high end/enterprise, somewhat like SLC was back in the day.
Last page, fifth paragraph, last line: "If I was" Should be: "If I were"
I know I'm a cunt for pointing it out, but I only do so because I generally think Anandtech offers fairly decently written articles, and I care too much about this sort of piss... Sorry.
How is data retention with this type of TLC NAND? Can the drive be powered off for a week or a month before data gets corrupted? While the drive is powered on, I assume there are refresh features for stale (a.k.a. infrequently accessed) data.
But in reality what are they going to be like? I don't think I'll buy another Samsung TLC drive after owning the 120 GB 840 EVO for the last while! It performs erratically, especially when resuming from hibernate. It can take anywhere from 10 secs to boot, up to 5 minutes, and I have applied the latest firmware and run the speed recovery app from Samsung. I have an mSata Crucial m500 240GB which is slower on paper, but in reality is much quicker and the performance is 100% consistent, it does the same thing, every time at the same speed!!
MLC all the way for me without any turbowrite nonsense, just straight forward advertised speeds across the whole drive all the time, without loosing data due to poor charge retention along the way!
Less nand equals slower performance, not more erratic performance. But it should perform at it's given speed consistentantly. I have over 30% of the drive free and 10% set as over provisioning on the 840 EVO. There is no excuse for a 5 minute boot time with an SSD. I don't trust these Samsung drives after my experience!
So I'm confused as its not very clear...Is Turbowrite turned on for all your benchmarks?
Is turbowrite needed to hit the the 540/520 read/write times? I saw the chart with TurboWrite on and off. So with turbowrite off this drive only gets 100mb/s ?
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
97 Comments
View All Comments
TEAMSWITCHER - Monday, December 8, 2014 - link
I'm in complete agreement. I'm not going to buy another 2.5" SSD before I can get an 850 Pro (512 GB) equivalent with an M.2 interface. I have two motherboards with empty M.2 slots waiting for the market to catch up. I know about the XP941, but the pricing isn't great. M.2 drives should cost almost the same as a 2.5" drive...especially after eliminating the worthless metal/plastic box.extide - Monday, December 8, 2014 - link
Seems to me like TLC will become the standard mainstream for 3D/VNAND, where as MLC will be pretty much only for high end/enterprise, somewhat like SLC was back in the day.MadDuffy - Monday, December 8, 2014 - link
Newegg prices (USD) are up:120 GB - 90
250 GB - 140
500 GB - 250
1 TB - 470
Email I received indicates these are promotional prices available through Dec 14th
casperes1996 - Monday, December 8, 2014 - link
Last page, fifth paragraph, last line:"If I was"
Should be: "If I were"
I know I'm a cunt for pointing it out, but I only do so because I generally think Anandtech offers fairly decently written articles, and I care too much about this sort of piss...
Sorry.
apudapus - Monday, December 8, 2014 - link
How is data retention with this type of TLC NAND? Can the drive be powered off for a week or a month before data gets corrupted? While the drive is powered on, I assume there are refresh features for stale (a.k.a. infrequently accessed) data.hojnikb - Monday, December 8, 2014 - link
If its up to jedec (i imagine it is) then its good for atleast a year.kgh00007 - Monday, December 8, 2014 - link
But in reality what are they going to be like? I don't think I'll buy another Samsung TLC drive after owning the 120 GB 840 EVO for the last while! It performs erratically, especially when resuming from hibernate. It can take anywhere from 10 secs to boot, up to 5 minutes, and I have applied the latest firmware and run the speed recovery app from Samsung. I have an mSata Crucial m500 240GB which is slower on paper, but in reality is much quicker and the performance is 100% consistent, it does the same thing, every time at the same speed!!MLC all the way for me without any turbowrite nonsense, just straight forward advertised speeds across the whole drive all the time, without loosing data due to poor charge retention along the way!
Lolimaster - Monday, December 8, 2014 - link
Your EVO is 120 and your other drive is 240, less nand, more erratic, simple. Look at the chart, you get consistent writes/read with 500-1TB models.kgh00007 - Saturday, December 13, 2014 - link
Less nand equals slower performance, not more erratic performance. But it should perform at it's given speed consistentantly. I have over 30% of the drive free and 10% set as over provisioning on the 840 EVO. There is no excuse for a 5 minute boot time with an SSD. I don't trust these Samsung drives after my experience!mlkmade - Monday, December 8, 2014 - link
So I'm confused as its not very clear...Is Turbowrite turned on for all your benchmarks?Is turbowrite needed to hit the the 540/520 read/write times? I saw the chart with TurboWrite on and off. So with turbowrite off this drive only gets 100mb/s ?
This article is very vague in regards to that.