Samsung SSD 840 EVO Review: 120GB, 250GB, 500GB, 750GB & 1TB Models Tested
by Anand Lal Shimpi on July 25, 2013 1:53 PM EST- Posted in
- Storage
- SSDs
- Samsung
- TLC
- Samsung SSD 840
Random Read/Write Speed
The four corners of SSD performance are as follows: random read, random write, sequential read and sequential write speed. Random accesses are generally small in size, while sequential accesses tend to be larger and thus we have the four Iometer tests we use in all of our reviews.
Our first test writes 4KB in a completely random pattern over an 8GB space of the drive to simulate the sort of random access that you'd see on an OS drive (even this is more stressful than a normal desktop user would see). I perform three concurrent IOs and run the test for 3 minutes. The results reported are in average MB/s over the entire time. We use both standard pseudo randomly generated data for each write as well as fully random data to show you both the maximum and minimum performance offered by SandForce based drives in these tests. The average performance of SF drives will likely be somewhere in between the two values for each drive you see in the graphs. For an understanding of why this matters, read our original SandForce article.
Random read speed is very close to that of the 840 Pro. The EVO doesn't look like a mainstream drive here at all.
Even peak random write performance is dangerously close to the 840 Pro. Only the 120GB drive shows up behind the pack. I should add that I'll have to redo the way we test 4KB random writes given how optimized current firmwares/architectures have become. The data here is interesting but honestly the performance consistency data from earlier is a better look at what happens to 4KB random write performance over time.
The relatively small difference between QD3 and QD32 random write performance shows you just how good of a job Samsung's controller is doing at write combining. At high queue depths the EVO is just as fast as the 840 Pro here. So much for TLC being slow.
Sequential Read/Write Speed
To measure sequential performance I ran a 1 minute long 128KB sequential test over the entire span of the drive at a queue depth of 1. The results reported are in average MB/s over the entire test length.
Sequential read and write performance, even at low queue depths is very good on the EVO. You may notice lower M500 numbers here than elsewhere, the explanation is pretty simple. We run all of our read tests after valid data has been written to the drive. Unfortunately the M500 attempts to aggressively GC data on the drive, so even though we fill the drive and then immediately start reading back the M500 is already working in the background which reduces overall performance here.
AS-SSD Incompressible Sequential Read/Write Performance
The AS-SSD sequential benchmark uses incompressible data for all of its transfers. The result is a pretty big reduction in sequential write speed on SandForce based controllers.
Facebook
Twitter
RSS
137 Comments
View All Comments
Grim0013 - Sunday, July 28, 2013 - link
I wonder what, if anything, the impact of Turbo Write is on drive endurance, as in, does the SLC buffer have the effect of "shielding" the TLC from some amount of write amplification (WA)? More specifically, I was thinking that in the case of small random writes (high WA), many of them would be going to the SLC first, then when the data is transferred to the TLC, I wonder if the buffering affords the controller the opportunity to write the data is such a way as to reduce WA on the TLC?In fact, I wonder if that is something that is done...if the controller is able to characterize certain types of files as being likely the be frequently modified then just keep them in the SLC semi-permanently. Stuff like the page file and other OS stuff that is constantly modified...I'm not very well-versed on this stuff so I'm just guessing. It just seems like taking advantage of SLCs crazy p/e endurance in addition to it's speed could really help make these things bulletproof.
shodanshok - Sunday, July 28, 2013 - link
Yea, I was thinking the same thing. After all, Sandisk already did it on the Ultra Plus and Ultra II SSDs: they have a small pseudo-SLC zone used both for greater performance and reducing WA.shodanshok - Sunday, July 28, 2013 - link
I am not so exited about RAPID: data integrity is a delicate thing, so I am not so happy to trust Samsung (or others) replacing the key well-tested caching algorithm natively built into the OS.Anyway, Windows' write caching is not so quick because the OS, by default, flush its in-memory cache each second. Moreover, it normally issue a barrier event to flush the disk's DRAM cache. This last thing can be disabled, but the flush of the in-memory cache can not be changed, as far I know.
Linux, on the other side, use much aggressive caching policy: it issue an in-memory cache flush (pagecache) ever 30 seconds, and it aggressively try to coalesce multiple writes into a single transactions. This parameter is configurable using the /proc interface. Moreover, if you have a BBU or power-tolerant disk subsystem, you can even disable the barrier instruction normally issued to the disk's DRAM cache.
Timur Born - Sunday, July 28, 2013 - link
My Windows 8 setup uses quite exactly 1 gb RAM for write caching, regardless of whether it's writing to a 5400 rpm 2.5" HD, 5400 rpm 3.5" HD or Crucial M4 256 gb SSD. That's exactly the size of the RAPID cache. The "flush its cache each second" part becomes a problem when the source and destination are on the same drive, because once Windows starts writing the disk queue starts to climb.But even then it should mostly be a problem for spinning HDs that don't really like higher queue numbers. Even more so when you copy multiple files via Windows Explorer, which reads and write files concurrently even on spinning HDs.
So I wonder if RAPID's only real advantage is its feature to coalesce multiple small writes into single big ones for durations longer than one second?!
Timur Born - Sunday, July 28, 2013 - link
By the way, my personal experience is that CPU power saving features, as set up in both in the default "Balanced" and the "High Performance" power-profiles, have far more of an impact on SSD performance than caching stuff. I can up my M4' 4K random performance by 60% and more just by messing with CPU power savings to be less aggressive (or off).shodanshok - Monday, July 29, 2013 - link
If I correctly remember, Windows use at most 1/8 of total RAM size for write caching. How much RAM did you have?Timur Born - Tuesday, July 30, 2013 - link
8 gb, so you may be correct. Or you may mix it up with the 1/8 part of dirty cache that is being flushed by the Windows cache every second. Or both may be 1/8. ;-)zzz777 - Monday, July 29, 2013 - link
I'm interested in caching writes to a ram disk then to storage. This reminds me if the concept of a write-back cache: for almost everyone The possibility of data corruption is so low that there's no reason not to enable it: can this ssd ramdisk write quickly enough that home users also don't have to worry about using it? Beyond that I'm not a normal home user, I want to see benchmarks for virtualization, I want the quickest way to create, modify and test a vm before putting it on front life hardwareWwhat - Monday, July 29, 2013 - link
I for me still say: I rather go for the pro version.andreaciri - Thursday, August 1, 2013 - link
i have to decide if buy an 840 now, or an EVO when it will be available, for my macbook. considering that RAPID technology is only supported under windows, and that i'm more interested in read performance than write, is 840 a good choice?