RAID0 ICH10R SSD scaling comparison

[lowfat]

Copying & pasting from XS since a lot of you don't frequent it.


All tasks were run on the following:
Core i7 920 @ 4.2GHz w/ HT off
6GB Aenon PC3-12800 @ 1600MHz 9-9-9-18
Gigabyte EX58-UD5
2 XFX HD 4980s
Asus Xonar Essense STX
1000W Antec Quattro
1, 2, and 4 X25-M's.
A system image was created. Then before each install on each setup, an HDDerase of all the drives was completed. I performed a reboot between every test. Superfetch & page file were disabled. All startup items were disabled through msconfig.

Results:






As was thought my many people here, RAID0 on ICH*R just doesn't benefit a lot of tasks out there. The best bet would to be to buy the fastest single SSD you can find. Buying multiple slower SSDs just shows no real benefits.

[Perineum]

Interesting reading, thanks!

[Squeetard]

I bet your antivirus is bottlenecking your system or there is something wrong with your OS and driver setup. I did my own tests with 1,2 and 3 60gb vertexes. and got a lot better scores and a lot better scaling with more drives. Virtually identical computers too. I was running Win7 x64 RTM. My boot times were 11, 22 and 30 seconds. Just looking at your iometer tests tells me something is up. Your sequential read scores should be at least twice that and damn near saturating the 600mbs peak of the ich10r chipset.

One tweak most users are missing is to go into the intel matrix storage manager, look at the properties of your array and enable write caching. Doing that via the device manager does nothing for an array.

[lowfat]

I'm not a n00b. I know there was nothing wrong w/ the system. If I would to run CrystalDiskMark I could be maxing out ICH10R's throughput. 250MB/s @ 20,000 IOPS is HUGE. Besides 4 X25-E's I don't think it is really possible to get higher than that on ICH10R. That was the point of this. Benches show more of a gain than there actually is with real world tasks. The only places it should a benefit were w/ boot up and copy/pasting files as you can see. Games had zero increase and the 'gains' in Photoshop was pathetic.

And write caching on ICH*R isn't all that helpful. Once again it really only helps with benches.

Forgot to add that the iometer results were from the OCZ fulltest.icf config. W/ a 'Number Of Outstanding I/Os' @ 64.

[Squeetard]

That is where I disagree. On my system (3-60gb Vertex in Raid0). Photoshop CS4 opens in a blink, under a second. saves files in a blink(20mb was the largest I tried). Game loading is a few seconds(wolfenstein and Batman are the 2 I installed and tested).

[Oversized Rooster]

It is true that the Intel drives achieve the greatest I/O per second but they are not impressive at all in terms of transfer rates.

Just 2 OCZ Vertex 60GB 1.3 achieve:

[Phobia]

Interesting results. Makes me re think getting 2 x25m's.

[Squeetard]

Rooster, what lowfat was getting at, and he is correct, is that the iometer benches are not comparable to atto or crystalmark or hdtune etc.. as they are high io sequential transfers.
I get different results than him in real world scaling.

[Putz]

from what Ive seen in reviews the ich10r scales well up to 3 drives (raid 0) but the 4th doesn't add much in terms of speed

and if you want raid with parity (5/6) then you'd be using a real controller anyways


the intel drives still crush the random IO competition, so what we see/feel the intel drives excel at, they don't excel so much in benchmark apps anymore

[Squeetard]

Again, what I think lowfat is saying and his scores reflect it, is that they do scale well in synthetic benchmarks but gain you little in real world performance.. I'm showing better improvements than him with more striped drives added in real world testing.