Benchmark Performance SSD Testing: AHCI vs IDE

Articles - Featured Guides
Written by Olin Coles
Friday, 09 April 2010
Benchmark Performance SSD Testing: AHCI vs IDE NOTICE: This article has been archived - a replacement exists. The results of this article did not properly illustrate the difference between AHCI and IDE mode SSD performance on the Intel ICH10 SATA controller. Please refer to: SSD Benchmark Tests: SATA IDE vs AHCI Mode for the updated version of this article with new SSD test results and conclusion. We apologize for the inconvenience. Related Articles: SOYO 26-Inch Widescreen LCD Monitor DYLM26E6 XFX Radeon HD 5850 HD-585A-ZNFC 240GB OCZ Vertex 3.20 Solid State Drive SandForce SF-1200 SSD Processor HIS Radeon HD 6870 IceQ Video Card NVIDIA GeForce GTX-460 1GB Video Card PNY XLR8 PRO Solid State Drive MSI R6950 Twin Frozr III Power Edition OC Foxconn GeForce 9800 GTX OC 512MB Video Card Use PlayStation3 Gamepad with PC Games

Comments

# RE: Benchmark Performance SSD Testing: AHCI vs IDE — kripz 2010-04-12 15:00

Do you think these will results will be the same for a motherboard with a JMB363 chip?

Report Comment

# Intel ICH10 Only — Olin Coles 2010-04-12 15:31

I cannot attest to the performance of another SATA controller, and will only go so far as to suggest these results apply to the Intel ICH10 controller on X58/P55 motherboards.

Report Comment

# RE: Benchmark Performance SSD Testing: AHCI vs IDE — Daan 2010-04-12 16:27

Why didnt you include any of intels ssd's in this test ? seems a bit of a waiste to leave the market leader out.

Report Comment

# Intel SSDs — Olin Coles 2010-04-12 16:34

I didn't feel that I needed to test every SSD controller available in order to demonstrate the difference between AHCI and IDE. It's like saying I need to use a specific brand of memory to test speed vs latency. Since you think that Intel is the only brand worth mentioning, feel free to write an article yourself and let us know if your tests turn out differently.

Report Comment

# RE: Benchmark Performance SSD Testing: AHCI vs IDE — Hell Racer 2010-04-13 22:51

AHCI shows best results in multithreaded random operations, so with HD Tune it's just impossible to see real difference. Grab CrystalDiskMark and you'll easily see 4 times faster random operations in AHCI mode. Grab IOMeter and run 100% random 100% read test with 4KB blocks and 4 or more threads and you'll see the difference too. So in my opinion HD Tune is just not that tool to see the real difference between AHCI and IDE.

Report Comment

# Finally... constructive criticism — Olin Coles 2010-04-13 22:54

To think... I had to wait for a guy named Hell Racer to finally deliver some decent feedback. I'm generally not a huge fan of CDM, and IOMeter adds the additional step of creating a chart for the lack of graphics. Test and production time are key. What about AS-SSD benchmark?

Report Comment

# RE: Finally... constructive criticism — Hell Racer 2010-04-14 03:01

... and I am not a fan of AS-SSD :), because for some reason it shows lower results then any other benchmark, I don't know why (maybe because it is coded in .NET and it just runs slower). But I think you will see the difference in "4K-64 Thrd" test.

Report Comment

# another data point — eaw718 2010-04-15 06:52

out of curiosity, what storage drivers were you using? intel matrix/rst, or the standard microsoft drivers?

i tested my 160gb intel g2 last night with hd tune pro. random read iops were consistently about 12% higher in ahci mode vs ide (couldn't test writes as i had data on the drive). i'm running 32-bit windows 7 ultimate with the standard microsoft drivers and my motherboard is a gigabyte p35 with ich9.

under crystaldiskmark, as hell racer mentioned, the intels do come alive in ahci mode, as the 4k qd32 random reads jump from about 20mb/s to 120mb/s. as-ssd also shows similar gains.

however, i do have to say that in normal use (i.e. not benchmarking), i cannot see a bit of difference between ahci and ide modes. except that ahci takes a bit longer to boot because the motherboard stalls for an additional 10 seconds as the drives are enumerated. not all motherboards do this, though.

Report Comment

# New Article Published — Olin Coles 2010-04-15 15:50

This article has been archived. The new article has been published to replace it.

Report Comment

# SATA HD — Boldi 2010-04-16 03:16

Waiting for SATA HD Testing: AHCI vs IDE!!

Report Comment

# Where did my comment go? Annyways, here's the reply Olin Coles — GullLars 2010-04-16 12:47

So it seems my post here vanished, here is the reply to:
"@ GullLars: Why is it you automatically assume I've done something wrong in my tests? Unless you've got the same ADATA S599 and have tested it with HD-Tune 4.01 for yourself, you're really just making baseless claims against my work. I have no idea where you found a SF-1200 performing at 35,000 IOPS on HD-Tune

Also, have used Iometer before for yourself? So I'm sure you know that there are so many different settings "
I didn't say 35,000 IOPS in HD-Tune. HDtune only tests at Queue Depth (QD)=1, so that would be way off. This is also the reason why HDtune is the wrong benchmark for testing IDE vs AHCI. You need to scale QD to see the difference, and an SSD that supports NCQ!

I use IOmeter frequently myself, and adjusting "# of outstanding IO's" is easy, it's on the front page...

Report Comment

# Reply part 2 — GullLars 2010-04-16 13:07

Here's part 2 to my reply.
You can quickly demonstrate the difference between IDE and AHCI by usign a SSD that supports NCQ, and running CrystalDiskMark _3.0_ and ATTO. Look at 4KB QD32 in crystal and test ATTO in QD 2, 4, 10 and compare the numbers. In AHCI mode they will be multiples of IDE mode.
Also, i accept your "challenge" regarding IOmeter, and i've made you a config where you can simply select target (partition) and hit run. It should give you data to set up nice graphs to illustrate the difference between AHCI and IDE quite nicely. AHCI will scale with queue depth, IDE will not.
IOmeter config comes in reply part 3.

Report Comment

# Reply part 3 — GullLars 2010-04-16 13:10

Part 3:
The config goes as follows: 10GB testfile, 1 sec ramp time, 10 sec run time pr access spec, exponential QD stepping (2^n) from 1-32, access specifications; 100% random and 100% sequential for 100% read and 100% write at block sizes 4KB-32KB. This totals 120 runs of 10+1 sec each, and will take up ca 20-25 minutes to run pr drive.
Link to config zip file: ##diskusjon.no/index.php?app=core&module=attach§ion=attach&attach_id=362070
If you don't want to go through the hazzle of making the raw data into graphs, send me the raw csv files in an e-mail and i'll mail you back graphs in a nice excell spreadsheet. I'll dedicate my weekend to it if you want.

Report Comment

# Ouch! — Parsec 2010-12-03 21:23

Hmm, I'm somewhat surprised at your reply, Olin. Although Daan was wrong to say the test was a waste (waiste?) without using Intel SSDs, and your response stating that was unnecessary is correct, suggesting he should write his own review was a bit much, IMO. I realize it is difficult reading criticism of your hard work, particularly when it is unjustified, but in this case you would have appeared more professional if you left out the last remark. Thanks for performing this test, which I have not ever seen elsewhere, I certainly appreciated it.

Report Comment

Refresh comments list

Comments have been disabled by the administrator.

Related Articles:

Comments