Samsung SM951 (512GB) PCIe SSD Review
by Kristian Vättö on February 24, 2015 8:00 AM ESTAnandTech Storage Bench - Heavy
While The Destroyer focuses on sustained and worst-case performance by hammering the drive with nearly 1TB worth of writes, the Heavy trace provides a more typical enthusiast and power user workload. By writing less to the drive, the Heavy trace doesn't drive the SSD into steady-state and thus the trace gives us a good idea of peak performance combined with some basic garbage collection routines.
| AnandTech Storage Bench - Heavy | ||||||||||||
| Workload | Description | Applications Used | ||||||||||
| Photo Editing | Import images, edit, export | Adobe Photoshop | ||||||||||
| Gaming | Pllay games, load levels | Starcraft II, World of Warcraft | ||||||||||
| Content Creation | HTML editing | Dreamweaver | ||||||||||
| General Productivity | Browse the web, manage local email, document creation, application install, virus/malware scan | Chrome, IE10, Outlook, Windows 8, AxCrypt, uTorrent, AdAware | ||||||||||
| Application Development | Compile Chromium | Visual Studio 2008 | ||||||||||
The Heavy trace drops virtualization from the equation and goes a bit lighter on photo editing and gaming, making it more relevant to the majority of end-users.
| AnandTech Storage Bench - Heavy - Specs | ||||||||||||
| Reads | 2.17 million | |||||||||||
| Writes | 1.78 million | |||||||||||
| Total IO Operations | 3.99 million | |||||||||||
| Total GB Read | 48.63 GB | |||||||||||
| Total GB Written | 106.32 GB | |||||||||||
| Average Queue Depth | ~4.6 | |||||||||||
| Focus | Peak IO, basic GC routines | |||||||||||
The Heavy trace is actually more write-centric than The Destroyer is. A part of that is explained by the lack of virtualization because operating systems tend to be read-intensive, be that a local or virtual system. The total number of IOs is less than 10% of The Destroyer's IOs, so the Heavy trace is much easier for the drive and doesn't even overwrite the drive once.
| AnandTech Storage Bench - Heavy - IO Breakdown | |||||||||||
| IO Size | <4KB | 4KB | 8KB | 16KB | 32KB | 64KB | 128KB | ||||
| % of Total | 7.8% | 29.2% | 3.5% | 10.3% | 10.8% | 4.1% | 21.7% | ||||
The Heavy trace has more focus on 16KB and 32KB IO sizes, but more than half of the IOs are still either 4KB or 128KB. About 43% of the IOs are sequential with the rest being slightly more full random than pseudo-random.
| AnandTech Storage Bench - Heavy - QD Breakdown | ||||||||||||
| Queue Depth | 1 | 2 | 3 | 4-5 | 6-10 | 11-20 | 21-32 | >32 | ||||
| % of Total | 63.5% | 10.4% | 5.1% | 5.0% | 6.4% | 6.0% | 3.2% | 0.3% | ||||
In terms of queue depths the Heavy trace is even more focused on very low queue depths with three fourths happening at queue depth of one or two.
I'm reporting the same performance metrics as in The Destroyer benchmark, but I'm running the drive in both empty and full states. Some manufacturers tend to focus intensively on peak performance on an empty drive, but in reality the drive will always contain some data. Testing the drive in full state gives us valuable information whether the drive loses performance once it's filled with data.
The SM951 performs even strongly in our Heavy trace and presents nearly 100% improvement in data rate over the XP941. In full state the SM951 loses a bit of its performance, but that's normal and the drop isn't any bigger than in other drives. Despite the lack of NVMe, it's starting to be clear that the SM951 is significantly faster than its predecessor and any SATA 6Gbps SSD.
The average latency is also cut in less than half, which is actually a more substantial improvement than going from a SATA 6Gbps drive to the XP941.
The share of high latency IOs is also the lowest with only 0.06% of the IOs having a higher than 10ms service time.
Facebook
Twitter
RSS
128 Comments
View All Comments
Railgun - Wednesday, February 25, 2015 - link
Boot times are irrelevant as there, too, there are several variables. BIOS or UEFI? HW involved. Other applications involved. And in the grand scheme of things, it's a one and done thing. If someone is so concerned on booting taking 5 seconds over 30, one can assume they'd leave the thing on. It's an irrelevant metric. Installing an OS again is an irrelevant metric due to the HW involved once again. I've never understood the fascination over boot times.Edgar_in_Indy - Wednesday, February 25, 2015 - link
Boot times and OS installation times (or game installation times, if it makes you feel better) would be interesting because they would be direct reflections of how a drive's theoretical speed is manifest in real world situations.I'm not really sure what your point is, and what you're arguing for. Why would you *not* want a few basic, real world metrics added to the other measurements? *Of course* the test system isn't going to be the same as every user's system. So what? We should still be able to glean some useful information about a drive's relative performance to other drives.
Besides, I have seen some situations where the synthetic tests didn't look great for a particular drive, but in the real world tests it fared much better. This is what led me to choose the Crucial M4 when I was shopping for a 256GB SSD a couple years ago. It wasn't the darling of the synthetic tests, but in the real world scenarios it was right up there with the best of them. It did particularly well in Anand's "Light Workload" test, which seemed much more typical of real-world use than the stress-test type scenarios.
And in regards to the "fascination" with boot times, I think that almost everybody prefers a computer that starts more quickly. I've been around since the DOS days, and that was the last time that I had a computer that booted in seconds, until recently. So having SSD's that can do the same is pretty dang cool to me.
I guess you could also ask a car guy why they have a "fascination" with 0-60 times. I mean, how often will someone really need to get to 60 mph in 5.2 seconds? It's ridiculous to think that somebody would be so worried about 0-60 times, unless they're an Indy Car driver or something.
And besides, there are too many variables (weather, humidity, altitude, tire conditions, driver skill, etc. etc.) to get a definitive 0-60 times, so we may as well junk the whole idea, and just assume that the $60,000 sports car is faster than the $30,000 sports car just because it puts up better numbers on the dyno or has a larger displacement engine.
But I really can't believe I'm having to explain this...
Railgun - Thursday, February 26, 2015 - link
I can't believe you had to explain that as well. You said you want real world tests, which already exist and refer to it during your selection of the M4. How does loading an OS by itself reflect anything? A plain vanilla install is a case that everyone will have only once during the lifecycle of that particular install. You allude to the light workload being more typical, which in itself is more than fine. If you've looked at what's included in the test suites, in particular the destroyer, you'll see that what can be considered normal usage tasks are there.-Download/install games, play games,
-Copy and watch movies,
-Browse the web, manage local email, copy files, encrypt/decrypt files, backup system, download content, virus/malware scan.
How are those not real world tests? They're not synthetic tests. What was your real world scenario that showed the M4 was better than whatever you were comparing it to? Why is that worse, or different than a boot speed test? I too don't hold a lot of value in the synthetic marks. As you mention, it's more for bragging rights than anything else.
I too remember the DOS days. Compared to that, there is no comparison as DOS, compared to Win7 is like comparing a Yugo to a Ferrari. They're both cars and get you from point A to B, but one is so much more than the other. Yes, they're both operating systems, but one has so much more to it and is more complicated than another. DOS 1.0 was about 4000 lines of code. Win7 is around 40 million. What about a nice striped down Linux build? That will load faster. What about Mac OS? Throw in a RAID controller and boot times get tossed out the window.
I don't think anyone has been missing any boot time metrics in the history of testing drives, whether SSD or otherwise. I've not seen one single review anywhere that shows boot times. The ONLY time I've ever seen it was initial comparison between an HDD and SSD. It's a moot point. You know it will be quick. Kristian's point is dead on. We're in the realm of possible milliseconds here. There's no point for the metric.
BTW, I am a car guy, and while 0-60 is all great, I'm more interested in the whole package. Handling, build quality, design, etc. :)
Edgar_in_Indy - Thursday, February 26, 2015 - link
I agree with most everything you said, but I would go back to my original gripe that there was no stopwatch involved. Data rates are great, and let's definitely keep them coming, but I would simply like to see some timed tests too. And even if the timed tests show little or no difference, then that is also a very valuable piece of information.My basic gripe with the article is that it does not clearly answer the question "Should I or shouldn't I?" Sure, some of the graphs are dramatic, but how much will they be manifest in the real world? I think the answer to "Should I buy it?" should be the payoff for reading a big review like this.
And if we can now say that we've reached the point where the real-world difference between SSD's for 99.99% of users is negligible, then I guess it begs the question whether these types of in-depth articles are worth writing, and worth reading, for very much farther into the future.
Kind of like how in-depth sound card reviews have mostly gone away, since we've reached the point where they just work without drawing attention to themselves. Unless you are in the tiny percentile where your occupation relies on having the best soundcard with very specific features, then you don't have to worry about it. Like I said, I came from the DOS days, and for many years soundcards were one of the hottest topics in PC hardware. Now they're pretty much a non-issue.
To draw a non-computer parallel, I'm sure an engineer could also write a 7,000-word review of a particular garbage disposal, going into great detail on every aspect of how the unit is built, but it would be total overkill, because people basically just want to know if it works or not. If SSD's are reaching a level of near-parity, then how many people will want to wade through all the background information in minute detail?
This has been a very informative discussion for me, and in a way it's refreshing to know that I no longer need to sweat about choosing an SSD in the future. That also means that I will be very unlikely to click articles or visit sites that are focused on SSD performance.
Railgun - Thursday, February 26, 2015 - link
I think you and will find, and Kristian, correct me if I'm wrong, that native nvme drives will increase perceived responsiveness as it allows for full on simultaneous read/write IOPS as opposed to unidirectional operations.That should show a nice increase in some scenarios.
Kristian Vättö - Wednesday, February 25, 2015 - link
I find that it's waste of time to run tests that show the obvious, which in this case is that boot and application launch times are the same for all drives. Like I said, it's starting to become common knowledge that for basic workloads there's no difference between SSDs and I've never argued against that.If I did real world testing, I would like to do it right. This means more than timing the boot time and how many tenths of a second it takes to launch a certain app. Frankly that has no value when you consider a power user's workload with dozens of apps already open, of which some might be rather IO intensive (like running a VM).
In such scenarios it can be hard to time the absolute difference because we are talking about stuttering and not seconds long wait times, but it's something that many certainly don't want to experience.
That said, I will probably craft something basic (boot, app and installation times) to show whether PCIe/NVMe has any relevance in basic IO workloads, but it's not something that I'm looking to make a part of our regular test suite since I don't think it gives an accurate picture of actual real world performance under multitasking workloads.
Edgar_in_Indy - Wednesday, February 25, 2015 - link
So would you say we're reaching the point where having the "fastest" SSD is really mostly about bragging rights?If that's the case, then it seems like the two most important specifications of an SSD would be size and price (much like it is for platter HDD's now). It would certainly make shopping for an SSD much simpler, if relative speed is no longer a meaningful factor.
Kristian Vättö - Wednesday, February 25, 2015 - link
Yes, and I don't think I've been trying shovel the high-end SSDs down people's throat.I think the SSD market mainly consists of two segments now, which are the mainstream and enthusiast/professional segments. For the mainstream segment, any modern SSD is good enough, which is why $/GB has been the dominating factor when I recommend drives for that market (and that's why the MX100 has been my recommendation for quite some time now if you've seen our "Best SSDs" articles).
The high-end sector is different in the sense that the users tend to want the best performance they can get. In some cases it's just for the bragging rights, but there are also workloads where SSD performance really matters (multiple VMs, photo/video/audio editing, etc). Some of our tests are more geared towards these users and I think we've been pretty clear about that, but as you said the Light workload test does a good job of illustrating average consumer usage and frankly the difference between drives in that test is rather small.
My goal has never been to push people to buy "faster" drives than they need and if some of my writings have come across as that then please, give me some examples and I'll try to learn from those.
Edgar_in_Indy - Wednesday, February 25, 2015 - link
No, I'm not trying to say that you've been pushing people to faster or more expensive SSD's. And even if you were, I probably wouldn't know, since I don't read all the SSD articles on here and follow all the developments. I mostly just jump in every year or two when I'm shopping for upgrades, and I try to play catch-up at that point in order to make sure I'm spending my money as wisely as possible.So for someone like me, who doesn't follow this stuff religiously, it's good to know I don't need to worry about missing out on big speed gains by not getting the hottest SSD of the moment next time I want to upgrade.
That being said, I'm still a little bit of a performance enthusiast, so I can't help but be curious when something like this comes along and shows the potential for big improvements over previous designs. That's why I was a little disappointed to not find much in the way of real-world results.
Anyway, it's obvious you put a lot of time and effort into this review, and the some of the performance results really were dramatic, so this is some good work.
Redstorm - Tuesday, February 24, 2015 - link
Why when updating the storage bench system did you pick a motherboard without a m.2 x 4 PCIe 3.0 slot. The Asus Z97 Delux is only providing 2 x pcie 2.0 lanes to the onboard M.2 slot. seams a bit short sited with the impending avalanche of x4 PCIe 3.0 SSD controllers coming out. Your new bench system is obsolete before it began. Using PCIe adapters is old school..