AS SSD Benchmark

Toggle sidebar Toggle sidebar
N

nsa666

Junior Member
Jun 20, 2009
17
0
0
I am developing a new benchmark especially for Solid State Drives.

The benefits of my benchmark are:
* Good precision of results
* Simplicity of use
* Optimized for SSD
* Screenshot function
* Text output
* It's freeware :D

Min. sys req:
* Windows XP/Vista/7
* .NET Framework 3.0
* An SSD to test with 3gb free space on it :D

Known bugs:


Feel free to report your experiences and results here.

Screenshot1

Screenshot2

P.S.: sorry my bad english.

If you want to support this tool:
Donate

Download: ->HERE<-

German support: <a target=_blank class=ftalternatingbarlinklarge href="http://www.hardwareluxx.de/community/showthread.php?t=617226">http://www.hardwareluxx.de...howthread.php?t=617226</a>

P.S.: Use at your own risk! I'm not responsible for any damage it can cause to your data or hardware.
 
A

alcoholbob

Diamond Member
May 24, 2005
6,365
433
126
The score is heavily tilted towards 64 thread 4K performance, so the scores will skew heavily toward Intel SSDs. Intel SSDs have 10 memory channels while the Indinilix controller has 4. So while random read performance is both in the 20-30MB range, in the 64 thread 4K read test, a Vertex will score around 60MB/s, while an Intel X25-M will score around 150MB/s.

As a result you get an OCZ Vertex scoring around 230 points, and an Intel X25-M will net you around 480, more than double the score. In practice the drives aren't actually that far apart unless you are running a server. Of course if you were doing that you'd probably own an Intel drive anyway (SLC variant).
 
O

Ourasi

Junior Member
Jun 15, 2009
19
0
0
Cudos for implementing the QD#64 random test, it's a real test, and QD#1 tells only a minute fraction of the story with no relevance to the real world. Sure, real life QD shows the benefit of the NCQ enabled Intel controller, and it's tremendous speed, but why leave the truth to sites like Anandtech an PC Perspective exclusively with their IOMeter tests, after all this is whats happening in the real world, or as close to it as you can possebly come..

The I/O pattern from IOMeter "workstation" config is the test that most users will see in the real world on a daily basis, and if possible, the difference is even more substansial between the Intel and the rest of the SSD crowd: http://www.pcper.com/article.p...733&type=expert&pid=10 If you can copy the IO pattern of "Workstation" config with QD#32-64 and add it to the AS Bench, you can be presumptious enough to call it a real world test in the bench;)

AS Bench is a "simple" (no pun' intended) utility, that actually gives the average users a glims into the real world with this QD#64 test, and even if it may put a dent into their SSD's and their own ego, it still is only the unmasked truth of whats going on at OS level in real world scenarious with regards to Queue Depth..

BTW: The raid0 array of my 2x X25-M drives, put out a substansial 72448/28160 IOPS 283/110mb/s in the QD#64 Random 4kb R/W and a score of 859 points, but even that won't come close to the advantage over the rest using "Workstation" config in IOMeter... Most of the Indilinx and new Samsung MLC SSD's are very good, and good value, but the Intel MLC drives are just in a different league..
 
Rubycon

Rubycon

Madame President
Aug 10, 2005
17,768
485
126
Effect of various stripe sizes here

Hope that makes sense.

With this program 64K seems to be the best. Of course there are a lot of other parameters on the card that can be changed that will change scores. Unfortunately I don't just have the time to play with them. :(
 
N

nsa666

Junior Member
Jun 20, 2009
17
0
0
That stripsize test is interesting. Thank you. That textbox next to partiiton selection box is for your notes, so you dont need to edit the screenshot afterwards.
;)
 
I

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
Originally posted by: Rubycon
Effect of various stripe sizes here

Hope that makes sense.

With this program 64K seems to be the best. Of course there are a lot of other parameters on the card that can be changed that will change scores. Unfortunately I don't just have the time to play with them. :(
Click to expand...

What do you suspect is the cause for the nearly 10x higher read access times? Do you think that is the true "metal" access time but the write access times are being masked by caching?
 
Rubycon

Rubycon

Madame President
Aug 10, 2005
17,768
485
126
Originally posted by: Idontcare

What do you suspect is the cause for the nearly 10x higher read access times? Do you think that is the true "metal" access time but the write access times are being masked by caching?
Click to expand...

Since writeback cache was in effect I'd certainly say so. There's another setting called read ahead (off/conservative/normal/agressive) which has little effect on access times (read) but helps out with sequential reads. (normal works best)

Too many knobs to turn. But that's nice because I've been in places where if you turn all these knobs (without a clue as to what you're doing!) you die. At least a blue screen never killed anyone! :laugh:
 
N

nsa666

Junior Member
Jun 20, 2009
17
0
0
Update:

1.0.3462.36297
- Filename bugfix at screenshot-function of copy-benchmark
 
taltamir

taltamir

Lifer
Mar 21, 2004
13,576
6
76
may i make a recommendation? fully write out what things are instead of using shorthand.. for example "programs" instead of "prog", etc. it greatly simplifies the program, and we have more than enough window and disk space today.
 
N

nsa666

Junior Member
Jun 20, 2009
17
0
0
Results are limited by cpu. 4k-64thread test seems acutaly ok(10GB/s is litle too fast, enough free space on the ramdrive?) its only some graphical bug. I will fix it.Since you have multicore cpu the 64thread test can be faster than sequentual single thread tests. You will not get stable results with this test because of background load of ram and to small test size for such fast "drive".

I dont know what happened to the access time test(cant open physical device by name?). It seems to be rare issue with windows xp. You are the second who reportes this bug. Can you test another drive so i know if it system-related or your ramdrive-drive emulation is just not very good.
 
N

nsa666

Junior Member
Jun 20, 2009
17
0
0
Than it is problem with the ramdrive-driver that not fully emulates a disk drive. But this is nothing to be concerned about, because its only used by some low level tools(and benchmarks), so its not practicaly relevant.
 
S

sxr7171

Diamond Member
Jun 21, 2002
5,079
40
91
My Samsung SLC is getting 1MB/s on the 4k writes test right now and it is taking forever. Looks like I made a bad choice of drive.
 
E

Emulex

Diamond Member
Jan 28, 2001
9,759
1
71
yeah remember the indilinx is using cache for lazy writes. not good if you lose power.

so you bought one of those samsung SLC off geeks? i was contemplating those SLC's they are cheap.
 
S

sxr7171

Diamond Member
Jun 21, 2002
5,079
40
91
Originally posted by: Emulex
yeah remember the indilinx is using cache for lazy writes. not good if you lose power.

so you bought one of those samsung SLC off geeks? i was contemplating those SLC's they are cheap.
Click to expand...

I wish I bought it off geeks, I bought it before that when they about $300 for 32GB (2.5"). At that time the X25M was $700+. In about 2-3 weeks suddenly the X25M was in the $400s. Then the geeks deal came.


I'm not 100% sure but I believe those drives have a Samsung designed controller.








 
I

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
Ah yes, its a privilege to bleed a little every now and then when handling hardware on the leading edge of a technology curve. (called the bleeding edge for a reason I suppose :laugh:)
 
N

nsa666

Junior Member
Jun 20, 2009
17
0
0
it is surely a lot faster than a hdd..

Update:
1.1.3466.29641
* Update checker
* New score system
+ Small design changes
 
Rubycon

Rubycon

Madame President
Aug 10, 2005
17,768
485
126
Wow you've updated it yet again! I need to update that page I put up above since I'm getting higher numbers across the board. I have the controller in an Asus P6T6 WS Revolution and its i/o and bandwidth is significantly better than the EVGA Classified.
 
Rubycon

Rubycon

Madame President
Aug 10, 2005
17,768
485
126
I put one in my notebook this afternoon... Text

It's OK I suppose. Nothing like running on a MP Xeon setup. Mobile chipsets just aren't optimized for SSD yet. Definitely snappier than the 7200 RPM Scorpio that was in there but not a wow-omg-holyshit-that-is-fast upgrade either. ;)
 
You must log in or register to reply here.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom