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!

---

Quote:

---

Originally Posted by videy

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

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

---

1. Are you using the latest version (2.34)?
2. You should put Total Length to 32mb and I/O to 'neither'. The latter is essential.

Please redo the test with these settings.

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. 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.

---

Although I do agree with you and I was not aware that these cards are not already PCI-e 2.0 I think with that said Im sure all raid chip manufacturers are hard at work making sure they are up to speed soon. SSD's are obviously going to saturate the SATA II bus by the end of next year and if you have say 32 SSD's on a controller your going to want every MB/s and IOP you can get for your money. Im sure Adaptec would agree with me... I bet we will be seeing x16 pci-e 2.0 cards by 2010 easy... :)

2GB/s IOPs, however cruious in the text, aren't really useful in bulk of the cases... except file copy essentially. With that much disk I/O, the processor and memory will become the big bottleneck :D
It's getting very fast random read/write performance that most want now and will want in the near future.

videy, any luck with IOMeter?

Quote:

---

Originally Posted by alfaunits

2GB/s IOPs, however cruious in the text, aren't really useful in bulk of the cases... except file copy essentially. With that much disk I/O, the processor and memory will become the big bottleneck :D

---

Usually chipset becomes the limit first.

You said the controllers maximum throughput is 666MB/s, but the burst rate is around 3000MB/s. Do controllers have a burst rate to?

Could someone explain?

This thread convinced me to go out and get a pair of X25-M SSD's. I was going to wait until the X25-E price drop, but I don't think that 64 gb from two SSD's would be enough. The M's are going to replace my slow My Seagate 7200.11. I am hoping for some similar benchmarks!

I wish videy would post the IOmeter results ;)

Quote:

---

Originally Posted by imperium

You said the controllers maximum throughput is 666MB/s, but the burst rate is around 3000MB/s. Do controllers have a burst rate to?

Could someone explain?

---

If the data requested is in the controller cache, then you'll get burst speeds reading it across the bus into main memory. Otherwise, the speed of the onboard processor will determine the limit on how much data can be read sequentially off-disk. If it's having to do parity calculations, there is a limit on how many can be done per second. Bigger cards use beefier (or dual) processors, clocked higher, not just more cache.

So what about having a ARC-1231ML card with 2gb cache with these two Intel ssds (X25-E)? Is that the most uber setup in this case or can you get a better controller card? :)

Quote:

---

Originally Posted by IanB

If the data requested is in the controller cache, then you'll get burst speeds reading it across the bus into main memory. Otherwise, the speed of the onboard processor will determine the limit on how much data can be read sequentially off-disk. If it's having to do parity calculations, there is a limit on how many can be done per second. Bigger cards use beefier (or dual) processors, clocked higher, not just more cache.

---

I was going to answer this question earlier, but then it occurred to me that the ICH9R does not have any dedicated controller cache, this burst looks to be normal write caching and/or Advanced Performance settings as provided in the Intel Matrix Storage Manager software or Windows Device Manager. If you uncheck these options you will see the true burst of the ICH9R and storage drives, I think... If I have them checked I get similar results on my Raptors...

You are correct. On ICH, the "burst" is essentially a memory copy ;)

I would guess that the ICH driver reserves some system memory for cache.

Quote:

---

Originally Posted by IanB

I would guess that the ICH driver reserves some system memory for cache.

---

Yeah I was thinking along the same lines...

Hey guys, I just received my pair of X-25M SSD's from newegg. I ran HD Tune read test and got max 539 mb/s, min 397.6 mb/s, and 514.2 mb/s average. 1996.1 Mb/s burst rate.

This was on a ICH10R on a Gigabyte X58-UD5. I could not seem to use the write test. When I tried to use it, HD Tune would tell me that "Writing is disabled. To enable writing please remove all partitions. Please check the manual for more information" Does any one know what the issue is?

Edit. used 64k blocks and windows vista ultimate.

Quote:

---

Originally Posted by FuriousSalesman

To enable writing please remove all partitions. Please check the manual for more information" Does any one know what the issue is?

---

Yes, if you read the error message carefully, you will see that the answer is included in the error message: you need to remove any partitions from the RAID0 array.
If you already installed your OS on it, you cannot run the write tests.
If your OS is still on another HDD, then go to Disk Management, delete the partition(s)/volume(s) from the RAID0 array and run the HDTune write benchmark.

Quote:

---

Originally Posted by Chosen.

Yes, if you read the error message carefully, you will see that the answer is included in the error message: you need to remove any partitions from the RAID0 array.
If you already installed your OS on it, you cannot run the write tests.
If your OS is still on another HDD, then go to Disk Management, delete the partition(s)/volume(s) from the RAID0 array and run the HDTune write benchmark.

---

bear in mind you'll loose whats on the drives, so make backups if you need

Quote:

---

Originally Posted by Levish

bear in mind you'll loose whats on the drives, so make backups if you need

---

I installed a fresh copy of vista, I haven't accumulated anything important.

Quote:

---

Originally Posted by Chosen.

Yes, if you read the error message carefully, you will see that the answer is included in the error message: you need to remove any partitions from the RAID0 array.
If you already installed your OS on it, you cannot run the write tests.
If your OS is still on another HDD, then go to Disk Management, delete the partition(s)/volume(s) from the RAID0 array and run the HDTune write benchmark.

---

I did read the error message, but I didn't understand the part about the partitions. This is also the first time I have ever done RAID. I made the array and just installed on it. I was not aware that there were any partitions.

Quote:

---

Originally Posted by FuriousSalesman

I installed a fresh copy of vista, I haven't accumulated anything important.
I did read the error message, but I didn't understand the part about the partitions. This is also the first time I have ever done RAID. I made the array and just installed on it. I was not aware that there were any partitions.

---

Well, if you really want to test the writes with HDTune, the only solution would be to install your OS on another disck, boot from that disk and remove the partitions on the RAID0 array. As previously stated, all data on the array will be gone after that and then after your tests you will have to re-install the OS.

Quote:

---

Originally Posted by Chosen.

Well, if you really want to test the writes with HDTune, the only solution would be to install your OS on another disck, boot from that disk and remove the partitions on the RAID0 array. As previously stated, all data on the array will be gone after that and then after your tests you will have to re-install the OS.

---

So, if I understand this right, I cannot do a write test if there is an OS on the array?


However, I think the write performance for the X25M is predictable. I don't think I have seen it higher than 75MB/s for 1 and 150MB/s for 2. I probably don't need to run a write test.

Have any of you seen OCZ's latest announcement? The new High-Performance Vertex series SSD drives with completely new controller and up to 64mb of cache. Nearly as fast as the Intel drives but at 1/2 the cost or less.

http://www.ocztechnology.com/aboutocz/press/2008/320
http://www.ocztechnology.com/product...ata_ii_2_5-ssd

Sure hope to see some in the wild soon...

There are already two topics on the OCZ drives, keep this X25-M/E related.

Quote:

---

Originally Posted by alfaunits

There are already two topics on the OCZ drives, keep this X25-M/E related.

---

Ok, thanks for the info. I thought this was just a SSD topic, not specific to Intel. But thanks for pointing this out to me.

The only ATTO bench is in raid so here it is in non raid =)
Sorry i could not resist!!!
http://71.108.249.186/sc02.jpg
I tried to sqeeze as much info in the screen shot but i know some details are still left out like... no page file. Well if you guys need other x25-e benches in non raid just reply or let me know. Ok thats all.

btw, does onboard intel raid have long initialization times?

Quote:

---

Originally Posted by Boogerlad

btw, does onboard intel raid have long initialization times?

---

No, it's extremely fast. A non-issue.

Quote:

---

Originally Posted by Boogerlad

btw, does onboard intel raid have long initialization times?

---

NO, and that's why some of us wish to have drives that work like they are supposed to on it, and not add another piece of hardware just for Raid0, to compensate for bad products:)

so the only disadvantage with intel raid is no cache and uses cpu for raid processing?

There is cache on the Intel RAID also. It just uses the System RAM instead of the card RAM. This is actually faster cache as no PCI-e device can transfer to memory faster than a memory copy can :D (and if the driver can do some MDL optimizations, there is no need for a copy even)
CPU usage is no concern really today. The reliability of integrated solutions isn't noteworthy, though I have never had problems with them, several members have reported dropped arrays.

Areca is worse for me though in every regard except number of ports :(

I've been using R0 via the ICHxR controllers for many years with 2-3 HDDs/SSDs and NEVER had a single drop, ever. Probably 10-15 different mobos, multiple systems, multiple different drives (mostly Raptors, VRaptors and SSDs.

ICH has been 100% reliable for me.

For 4 x X25-E 64G in Raid 0, I suppose we need a dedicated raid card?

Quote:

---

Originally Posted by sniper_sung

For 4 x X25-E 64G in Raid 0, I suppose we need a dedicated raid card?

---

Yeh mate, a LSI card with fasthpath.. preferably the LSI 9260.. Quick question u said the iCH9 has a limit of 660mb/s, What about the ICH10?:shrug:

Quote:

---

Originally Posted by Johnny87au

Yeh mate, a LSI card with fasthpath.. preferably the LSI 9260.. Quick question u said the iCH9 has a limit of 660mb/s, What about the ICH10?:shrug:

---

Same 660MB/s.

Quote:

---

Originally Posted by sniper_sung

For 4 x X25-E 64G in Raid 0, I suppose we need a dedicated raid card?

---

If I were building a system _today_ I would use 3-4 x25-e's on ich10. Unless you have some specific need for higher sequentials the ich10r is a beast.

The 9260+key is great but for most everything other than a few niche cases (and synthetic benchmarks) the ich10r+some intel slcs will give you an extreme high end storage subsystem.

The 1880 may change things, however what info has leaked out isn't mind blowing.

If you're like me you want all your apps to open instantly, and they will, an increased as ssd score won't open anything any faster. :)

+1 with Mr mbreslin - all true

Quote:

---

Originally Posted by F@32

Same 660MB/s.

---

Thanks mate :up: