[Samsung 830 128 GB] Is this benchmark normal?

[Riscilla]

Hi guys,

I have just bought a Samsung 830 128 GB SSD. This is a SATA 3 SSD.

My OS is Windows 7 SP1 x86. My mainboard is ASUS P8H61-M LE (This mainboard only supports SATA 2). I have enabled AHCI in my system. I have also disabled super fetch, defragmentation & hibernation.

Here are some benchmark results:

Samsung SSD Magician Benchmark

CrystalDiskMark Benchmark
​

I know the limit of SATA 2 on sequence read/write speeds, so I have nothing to complain about these benchmarks. However, the IOPS and random read/write speeds seem to be quite slow, don't they? Is there anything else I can do to optimize my SSD's performance?

Thank you very much.

 

[ShintaiDK]

Looks fine to me.

Remember a raptor HD will do around 200KB/sec in the same bench.

 

[Cerb]

However, the IOPS and random read/write speeds seem to be quite slow, don't they?

3Gbps will also limit random access results a bit, in synthetic tests. On an Intel 6Gbps, the random read aught to be around 25MB/s. Figure you have an older, less optimized controller, and 3Gbps slowing it down some. Meanwhile, most reviewers are using newer/better SATA controllers on newer/costlier mobos.

So, your figures don't look bad, and you should probably not worry about it. 99.999% of your time on the computer should be spent doing things other than running benchmarks. 45,000 IOPS is fast, and so is 15,000, no matter what someone with a perfect SSD test rig might be able to get.

Is there anything else I can do to optimize my SSD's performance?

Replace your mobo, or get a $150+ new LSI controller card. Realistically, if you're using it pre-formatted, or formatted it in Win 7 (I forget if they come formatted NTFS, but I seem to recall them being that way), there's nothing worthwhile to do but use it.

If you have a HDD installed, you aught to turn defragging back on. Windows 7's defragger will only defrag the HDD, so there is no reason to turn it off.

 

[aviator78]

Looks ok, like you said, its on SATA2.
And you can put hibernation back on as well if you want. It doesn't make problems.

 

[jhansman]

Yeah, I put an M4 in my wife's SATA 2 laptop and it performed about the same.

 

[Riscilla]

Thank you very much, guys. Especially thank you Cerb for your detail and helpful answers.

I feel safe now

 

[_Rick_]

You're actually lucky, I just found this: http://marc.merlins.org/perso/linux...slowness_-and-settling-with-OCZ-Vertex-4.html

Which is somewhat worrying.

 

[Cerb]

It's hard to find this stuff searching, just due to keyword collisions and commonness, but someone posted this in another thread, so I'm ripping it off:
http://www.tomshardware.com/reviews/sata-6gbps-performance-sata-3gbps,3110-7.html

I've seen a couple more in depth reviews of that nature, but finding them again proved to be a challenge. The results of such testing, including more steady state testing, IIRC, ended up basically the same as the few tests Tom's ran, just with more steady-state info. There's enough of a difference you should be able to feel it when under a storage IO load, but the differences between the drives themselves still dominate.

--

You're actually lucky, I just found this: http://marc.merlins.org/perso/linux...slowness_-and-settling-with-OCZ-Vertex-4.html

Which is somewhat worrying.

The C300 definitely had some GC issues, and Intel's baby enterprise firmware and controller that did mostly incremental GC was a better way to go (humans are usually slower than the accesses to SSDs, so worry about those corner cases!). Idle can get the best WA and performance (more total information to work with at a time), when it has the opportunity, but incremental needs to be there and provide only minor performance penalties. Crucial's C300, 1st-gen SF, Barefoot, and Samsung's 470, off the top of my head, all had similar problems, though Crucial firmware and the Barefoot were surely the worst, often resulting in drives dropping out, sometimes corrupting themselves when the user would reboot the system.

Mainstream SSD tech is only just making its way out of the beta testing phase, IMO.

The 830 is much more interesting, IMO, but still pointless for the OP. There seem to be some kind of issues either in the drive or the kernel that surface when it's using dm-crypt. Others appear to have lesser speed issues w/ dm-crypt and the 830, and that not using dm-crypt doesn't fix it certainly points to an issue with the drive. Intel SSDs don't have to go on sale, even when they aren't the fastest at benchmarks. Even they got bit, withe the 8MB bug, but they have performance corner cases pretty much taken care of.

 

[beginner99]

And you can put hibernation back on as well if you want. It doesn't make problems.

But it uses a lot of space, namely as much as you have RAM. Thats why I have it turned of.

 

[beginner99]

You're actually lucky, I just found this: http://marc.merlins.org/perso/linux...slowness_-and-settling-with-OCZ-Vertex-4.html

Which is somewhat worrying.

Well he is using it on linux. I mean you have just have to be prepared for issues if you use very new tech on a free OS for which the new hardware probably was never tested or validated against and on top of that use some encryption software.

And now I'm waiting for his next post afters his 10th Vertex 4 RMA. The difference between the warranty is that with OCZ you sure will need to make use of it.

 

[_Rick_]

Well he is using it on linux.

It's a SATA device, which should function independently of the OS. The fact that it got into the described situation on multiple machines, on Windows and Linux, for two different drives, means that either he got two bad drives in a row, or something bad can happen to the drive that will kill random reads.

 

[Cerb]

It's a SATA device, which should function independently of the OS. The fact that it got into the described situation on multiple machines, on Windows and Linux, for two different drives, means that either he got two bad drives in a row, or something bad can happen to the drive that will kill random reads.

Something likely involving having used them as dm-crypt devices, since most users have no problems. It should work independently of how it is accessed (OS doesn't matter, but for Samsung, FS can), but clearly there is some pathology in its firmware (or worse, HW FTL!) that dm-crypt brings out.