Intel SSD (X25-E & X-25M) RAID 0 Benchmarks

Hi There.

(recommend you resize your web browser to accomodate displaying the benchmark graphs side by side)

First post, be nice :) I'm a storage noob. Got my hands on a number of different SSD units that I thought it would be fun to benchmark:

1) Qty 2 Intel X25-E (32GB, SLC)
2) Qty 3 Intel X25-M (80GB, MLC)
3) Just for kicks, included an Apple (Samsung) 128GB SSD (MLC)

I didn't want to rip apart my main rig, so all testing was done with spare parts I had lying around: Gigabyte P35-DS3R (Intel ICH9R), Q6600, 4GB DDR2-800 running XP, SP3 and the latest Intel Storage Manager drivers from Oct 2008.

The goodies:
http://i383.photobucket.com/albums/o...8/IMG_1054.jpg

First let's start with single drives. According to the well known Anandtech article, the ICH9R had a 80MB/S cap on throughput. I believe this NOT to be an issue anymore, as I saw transfer rates well above 80MB/Sec. For example, here is the read rate for the X25-E vs the X25-M. You'll see that the M has slightly highly read rates than the E, a theme I found consistent across all of my testing.

Single Drive Benchmarks
X25-E (left) vs X25-M (right)
http://i383.photobucket.com/albums/oo278/bm5-2008/1.jpghttp://i383.photobucket.com/albums/oo278/bm5-2008/4.jpg

Writes, on the other hand are a different story. The E drive blows away the M in terms of writes.

X25-E (left) vs X25-M (right)
http://i383.photobucket.com/albums/oo278/bm5-2008/2.jpghttp://i383.photobucket.com/albums/oo278/bm5-2008/5.jpg

And just for fun I included an Apple Branded Samsung SSD I have from my Macbook. This is a 128GB MLC unit that is currently shipping with the new (late Oct 2008) Macbooks. I'll compare it to the Intel X25-M.

Intel X25-M (left) vs. Apple/Samsung SSD (right)
http://i383.photobucket.com/albums/oo278/bm5-2008/4.jpghttp://i383.photobucket.com/albums/oo278/bm5-2008/8.jpg

Umm ok, it's fair to say that the reputation that the Intel SSDs have is well deserved :p:

Ok let's move on to the RAID 0 benches.

Raid 0 Benchmarks
Configuration: Raid 0, 128K stripe, ICH9R. I started with the "default" of 128K. I found that enabling the volume write cache on the Intel driver made a significant difference. Results with the volume cache on and off:

X25-E, 128k stripe, no cache (left) vs X25-E, 128k stripe, cache on (right)
http://i383.photobucket.com/albums/o...m5-2008/10.jpghttp://i383.photobucket.com/albums/o...m5-2008/11.jpg

Because of the significant improvement in performance, I left the volume cache enabled for all subsequent testing.

Intel X25-E (left) vs Intel X25-M (right) -- Read
http://i383.photobucket.com/albums/o...m5-2008/11.jpghttp://i383.photobucket.com/albums/o...m5-2008/14.jpg

As we saw with the single drives, we the X25-M performs slightly better than the X25-E in terms of read rates. Again, writes completely reverse the story:

Intel X25-E (left) vs Intel X25-M (right) -- Writes
http://i383.photobucket.com/albums/o...m5-2008/13.jpghttp://i383.photobucket.com/albums/o...m5-2008/15.jpg

File Benchmark
Intel X25-E (left) vs Intel X25-M (right)
http://i383.photobucket.com/albums/o...m5-2008/17.jpghttp://i383.photobucket.com/albums/o...m5-2008/18.jpg
You can see the dramatic difference in write speeds!

At this point I questioned if changing the stripe size would make a significant difference. The common wisdom is that using a smaller stripe size would ensure that small files get split between the two drives, at the expense of longer access times. Well, since these are SSDs would it not make sense to use the smallest stripe size?

Intel X25-E, 4k stripe
http://i383.photobucket.com/albums/o...m5-2008/19.jpg

Well, using a small stripe size does tend to smooth out the transfer rates, but check out the CPU usage! Obviously this is not optimal in terms of system usage. This small stripe size might be an option if I was using a hardware based RAID controller.

Through trial and error, I determined that a stripe size of 64K or 128K is optimal for my system. 64k gave me slightly higher CPU usage and a much better burst rate, while 128k gave me lower CPU usage and a slightly higher average transfer rate.

Intel X25-E, 64k stripe (left) vs 128k stripe (right)
http://i383.photobucket.com/albums/o...m5-2008/21.jpghttp://i383.photobucket.com/albums/o...m5-2008/11.jpg

I'll probably use the 64k stripe, just because of it's high burst rate (if indeed HD Tune is accurate). HD Tach shows a similar result.

http://i383.photobucket.com/albums/o...m5-2008/23.jpg

ok moving on, if two Intel units are good, how about three? :D

I used three X25-M units to make a Raid 0 volume (I didn't have a third E unit). The results are presented below:

Read
http://i383.photobucket.com/albums/o...m5-2008/25.jpg

Write
http://i383.photobucket.com/albums/o...m5-2008/26.jpg

I believe that the ICH9R has a throughput limit of 666 MB/Sec. The throughput of the above array might (or might not) be higher, but I think this is the max throughput. The reason I believe this to be the case is because I used all 5 Intel units (mix of E and M units) to create a single volume. While I realize this is suboptimal and probably would not be created for actual use, it should show increased read speeds.

http://i383.photobucket.com/albums/o...m5-2008/28.jpg

But again we see max throughput at around 666 MB/Sec. Perhaps it's related to the mix of E and M drives, but I think it's a limitation of the on-board raid controller.

Well, that's all I have for today. Hope you found this data interesting. I'll be rebuilding my main rig with the X25-Es as the system volume soon. I'll report back if I see any signficant difference using the ICH10R vs the ICH9R.

One question to the storage experts out there, do you think going with a RAID card will signficantly improve performance? Also, do RAID cards behave properly with power management (in other words will my system be able to sleep and hibernate)? My previous experience with RAID cards were that they were designed for always on serves so power management was an issue.

Cheers! :D

Kick ass benches man. Definitely going to be picking up a few E's.

Sweet setup ! I need a price drop to happen badly! :)

Must Have.

Thanks for all that work.

More HDTach/Tune... Blah :(

Quote:

---

Originally Posted by alfaunits

More HDTach/Tune... Blah :(

---

like I said, I'm a storage noob -- recommendations on what you would like to see?

Thanks a lot for all that work! Now, go download IOmeter! :D

Thanks for the info! I'll be picking up some E's in the future.

iometer is confusing as all hell, but it is good and reputable, try a 2 io workstation pattern

can you do a benchmark of the three X-25M in raid 5? their performance might be surprisingly similar to raid 0... thanks.

Quote:

---

Originally Posted by videy

One question to the storage experts out there, do you think going with a RAID card will signficantly improve performance? Also, do RAID cards behave properly with power management (in other words will my system be able to sleep and hibernate)? My previous experience with RAID cards were that they were designed for always on serves so power management was an issue.

Cheers! :D

---

Thanks a ton for the benches! Very interesting. I think going with a RAID card would definitely improve performance. I am not sure about anyone else, but I do suffer from sleep/hibernate problems with my RAID card and Vista 64bit. I actually have to disable both sleep/hibernate to prevent my computer from rebooting/crashing. Of course, it may just depend on the user or the brand of card.

Do you have any games? How about some comparisons of game load times? :)

thanks for the reviews, all these have very impressive results in raid

Try also ATTO, but it is very sensitive to cache.

Just want to inform you that you have made a mistake in your graph images. The 8th and 10th images of HD Tune Pro are identicle and named "14.jpg" but you called the first one a test of "X25-E, 128k stripe, cache on (right)" and the second image in question the "Intel X25-M (right) -- Read" test. Both images should not be identical with the same file name. Look below and then compare above with your tests.


Image 8 test X25-E, 128k stripe, cache on (right)

http://i383.photobucket.com/albums/o...m5-2008/14.jpg


Image 10 test Intel X25-M (right) -- Read (Identical to above image)

http://i383.photobucket.com/albums/o...m5-2008/14.jpg


Is it possible to find the correct images for these test please? I am thinking of purchasing an Intel SSD and want the most accurate information. Thanks :)

SkOrPn

Quote:

---

Originally Posted by videy

One question to the storage experts out there, do you think going with a RAID card will signficantly improve performance? Also, do RAID cards behave properly with power management (in other words will my system be able to sleep and hibernate)?

---

I'm not sure about the power management thing, but I can say this for sure good quality PCIe RAID cards should use all available lanes that they were made for so in theory you should have 500 MB/s on each PCIe 2.0 lane, so if you get a x4 raid card would result in 2000MB/s throughput, and x8 4000MB/s theoretical throughput. NOW, after saying that the SATA II interface is still a maximum of 300MB/s theoretical throughput, so in order to take full advantage of all three drives you would need a PCIe RAID card of x4 minimum (since x1 does not provide enough throughtput for the bandwidth of 3 x SATA II connectors which is 900MB/s) and a processor on board capable of handling it. Now one last note, since both drives E and M seem to have a maximum of 215MB/s (on ICH anyway) read transfer you should be good to go with a high quality PCIe 2.0 x4 RAID card. Intel says that both E and M have a maximum of 250MB/s read, so I think a good dedicated raid card would provide you with about 105MB/s (above the ICH controller) of additional throughput with 3 M drives in raid 0. That is my best guess, but I'm no storage expert... :)

The question now? Is spending a few hundred more for a descent raid card worth an additional 100MB/s or so? There is also other benefits to having a dedicated RAID card, so my answer is YES it is worth it. lol

Man this is such a nice review, thanks!
I wish I had some $ for the E xD

Quote:

---

Originally Posted by SkOrPn

I'm not sure about the power management thing, but I can say this for sure good quality PCIe RAID cards should use all available lanes that they were made for so in theory you should have 500 MB/s on each PCIe 2.0 lane, so if you get a x4 raid card would result in 2000MB/s throughput, and x8 4000MB/s theoretical throughput.

---

I don't think you'll find ANY PCIe cards that are NOT videocards able to use PCIE 2.0 bandwidth. This was specifically introduced for video. Given the glacially slow rate of uptake of PCIe by hardware controller manufacturers changing from PCI, and the probability that they are going into boards that can only support PCIe 1.0 anyway (and the chip issue mentioned below), I bet there's little chance of a PCIE 2.0 controller card soon.

Halve those numbers and you might be closer. And even then the throughput is limited by the IOPs chip on the card for wider bandwidth slots, not the slot width, so IIRC PCIe 1.0 x4 is about the best bandwidth you'll get whatever card you have and even if it's designed for/plugged into x8.

Quote:

---

Originally Posted by IanB

I don't think you'll find ANY PCIe cards that are NOT videocards able to use PCIE 2.0 bandwidth.

---

http://www.mellanox.com/content/page...n=41#tab-three

videy, any chance you will run IOMeter on these? If you have problems setting it up (i.e. figuring which test to run), let us know.

Quote:

---

Originally Posted by SkOrPn

Just want to inform you that you have made a mistake in your graph images. The 8th and 10th images of HD Tune Pro are identicle and named "14.jpg" but you called the first one a test of "X25-E, 128k stripe, cache on (right)" and the second image in question the "Intel X25-M (right) -- Read" test. Both images should not be identical with the same file name. Look below and then compare above with your tests.

SkOrPn

---

oops you're right. The original post has been corrected with the correct information. thanks for catching the mistake.

Great and informative review! There isn't a lot of testing on the X-25E on the internet. This is probably the most informative example I have seen.

I am having trouble deciding which SSD's to get. It seems like better drives are coming out all the time. I was thinking about getting a pair of these, but Mtron claims that they will have drives with read/write speed of 260/240MB/s . If price isn't an issue, what do you get?

http://www.mtron.net/English/PressRo...&sb=&category=

Quote:

---

Originally Posted by alfaunits

videy, any chance you will run IOMeter on these? If you have problems setting it up (i.e. figuring which test to run), let us know.

---

sorry, the thanksgiving (U.S. Holiday) festivities kept me busy recently. I'll try to run the IOmeter benchies over the next day or two.

Quote:

---

Originally Posted by m^2

Originally Posted by IanB:
I don't think you'll find ANY PCIe cards that are NOT videocards able to use PCIE 2.0 bandwidth.

http://www.mellanox.com/content/page...n=41#tab-three

---

Apologies, although Infiniband is hardly "consumer" hardware that 99.9% of users here will use. :p:

It goes to show though that smaller niche companies are better at moving with the leading edge. I remember trying to find a PCIe SCSI card card a few years ago and seeing Adaptec's site stating specifically that there would be NO PCIe interfaces from them on anything under SAS cards, ie. nothing "consumer" grade: even at a time when Areca was making a big splash they didn't see a market. :shakes:

need some help configuring IOmeter. What types of operations do you want to see in the benchmark? I see that there are "workstation" and "file server" profiles somewhere, but can't find them.

How many worker threads is appropriate?

thanks!

Quote:

---

Originally Posted by RealTelstar

Try also ATTO, but it is very sensitive to cache.

---

per request, ATTO results (X25-E, Raid 0)

http://i383.photobucket.com/albums/o...008/atto-1.jpg

Here's a nice IOMeter config file that you can use.
Just choose the target disk, and select Random Write (4K) access pattern.
Let it run...
(it will run for a long time, it first fills the drive completely by making the largest possible iorw.dat file)

Quote:

---

Originally Posted by videy

need some help configuring IOmeter. What types of operations do you want to see in the benchmark? I see that there are "workstation" and "file server" profiles somewhere, but can't find them.

How many worker threads is appropriate?

thanks!

---