Vertex SSD vs. HDD RAID 0 vs. Vertex SSD RAID 0 *Updated*

[garritynet]

I have had a rough couple of months followed by three nice cash windfalls in a row. Sweet. I paid off all my bills, purchased lots of presents for my family and with the money I was supposed to save I responsibly purchased two OCZ Vertex 30GB SSD drives from Newegg @ $99 AR and a 3ware 9650se-4LPML RAID card from Gunblade for $160 on the FS/FT forums. Lets see how performance much $6 a gig will get you.

The OCZ Vertex Drive uses the Indilinx Barefoot controller and is well documented in the AnandTech Storage section. It advertises a max read of 230 mb/s and a write of 135mb/s. Also, contrary to their original announcement the 30GB and 60GB models had 64mb cache in the final specs.

The 9650se-4LPML is a hardware RAID card produced by recent LSI acquisition, 3ware. It is powered by a proprietary 3ware processor, has 256mb DDR2 533 RAM and rides a PCIe 1.1 4x slot. This particular model connects to a maximum of 4 drives. Its my first RAID card so I'm pretty exited. The manual for this thing is intimidating at over 300 pages, but is mostly redundant information. I was able to install the device, load the 9.5.3 software that fully supports Windows 7 64bit and create an array in under half an hour.

***Edit: I have decided against the Rocket 620 comparision as the results were exactly the same when compared against the MB***

Since I installed the SSD I have done no maintenance beyond letting the trim do its thing. I have installed, deleted, and reinstalled programs, moved, copied and deleted files at will and reinstalled Windows 7 multiple times without wiping. I think it is safe to say that this drive is not optimized for benchmarks. Thats what I wanted. I wrote a 620GB file to the raid and basically done my best to fragment it, but due to the limited amount of time I have been using it is probably still mostly intact. The Vertex RAID has had Windows written to it a few times so I am going to let log out overnight so the OCZ GC can do its thing. I will update the results tomorrow night to see if there has been any change.


First lets look at HD Tune.

WD RAID

Single Vertex

Vertex RAID 0

 

[garritynet]

Next up, lets compare Crystal Disk Mark.

WD RAID

Single Vertex

Vertex RAID

 

[garritynet]

Finally the ATTO test.

WD RAID

Single Vertex

Vertex RAID

 

[Zap]

It doesn't do shit.

LOL. Well, it won't until you get some SSDs (which aren't yet available) that use the faster connection.

Good job on your first (ongoing) review!

 

[garritynet]

Updated the Results, let me know what you guys think.

Would it be worth it to try 4x Vertex RAID 0?

 

[Rubycon]

Would it be worth it to try 4x Vertex RAID 0?

No because neither host you have is powerful enough to show significant advantages of additional expense and complexity.

 

[MalVeauX]

Heya,

Measuring maximum bursts and sustained reads really doesn't give you a lot of information about HDD's in the sense of actual use. They're just big numbers for benchmarks.

If you really want to see which drive(s) setups are better for actual use, look at their random read and writes. These numbers are much, much smaller. But if you compare HDD and RAID0 HDD with even a single SSD, you'll see a very big difference. It's at this end that you actually feel the difference of an SSD when using your system (unless all you do is copy gigs of large files back and forth and watch it for fun or something; and I'm not being snarky saying that, just saying).

Example:

There's max and average there for an agility SSD.

And here's random reads. If you really want to benchmark an HDD and an SSD, look at the access time and rates of these randoms. Do this on your RAID, do this on an SSD and do this on SSD in RAID and look at these numbers. When you see the difference here, you won't care about burst maximums anymore. You will be looking at the IOPS, average time for random file sizes, and the very different data speeds of these. This is more a representation of actual use and will give you a much better picture of which setup performs truly better than the other in the computing environment.

Very best,

 

[garritynet]

Wonder why its less than your single drive?

 

[Idontcare]

There is something wonky about your raidcard and your raid config tests with the vertex. You should NOT see a step-function decline in read speed on the vertex array in HDtune and look at the differences in crystaldiskmark between vertex and vertex raid0.

Until you figure out what is causing these inconsistencies in the results you can't really do much of an assessment of the performances. I mean you could ignore these issues of course and make a conclusion, but the conclusion doesn't hold much value when it is apparent the hardware setup you have there is borked somehow.

 

[Rubycon]

Yes random i/o is very important!

SSD array:

Single 7200.1 on same SAS host:

Little faster don't you think?

 

[garritynet]

There is something wonky about your raidcard and your raid config tests with the vertex. You should NOT see a step-function decline in read speed on the vertex array in HDtune and look at the differences in crystaldiskmark between vertex and vertex raid0.

Until you figure out what is causing these inconsistencies in the results you can't really do much of an assessment of the performances. I mean you could ignore these issues of course and make a conclusion, but the conclusion doesn't hold much value when it is apparent the hardware setup you have there is borked somehow.

Any idea where I should start looking? What specifically should I be looking at with the Crystal Disk Mark?

This is after running the garbage collection overnight. Could it be possible that in testing it ran into one of the 15GB or so that gets deleted immediately after installing W7 and had to do the whole read, write, read deal? Window.old, Hibernation file, ect.

MalVeauX- You do anything special there? Anything I could do? or did you just plug it in the MB and not worry about it? If not, what kind of MB are you using?

 

[garritynet]

Well, I've spent the past couple of days trying out every configuration of RAID that my card allows and running benchmarks on them.

I just wanted to see what was making my results "off". These are the settings I found best. They are my original settings. After letting the drive idle for a while and do its GC thing in a new W7 install I ran a couple of quick test. Its not really a comparision anymore. Anyway, here are my results. I am not sure what is causing my performance loss, other than the card but I will be looking into how my various bios settings affect the PCIe bus as well as giving the AMD southbridge drivers a second chance. Suggestions welcome.

 

[MalVeauX]

MalVeauX- You do anything special there? Anything I could do? or did you just plug it in the MB and not worry about it? If not, what kind of MB are you using?

Heya,

Nothing special at all. It's the difference between SSD's. They're not all equal. Some have better cells, some have better controllers, etc. They key is simply to balance price and performance if that's what you're after, otherwise, you can spend all you want and get a ton of intel SSD's that will nuke most anything else.

And on that note, Rubycon's comparison shows how SSD and HDD compare in real world differences. The max sustained reads are of course nicer, but only by a percentage that is not that insane. When you look at the IOPS though, you see a difference where the SSD is doing 683 times more speed. That's monumental.

Very best,

 

[garritynet]

MalVeauX- I get what your saying, but we both have OCZ SSDs with the same controller. My single SSD on the MB gets no where near your performance. What MB are you using? What drivers?

 

[MalVeauX]

MalVeauX- I get what your saying, but we both have OCZ SSDs with the same controller. My single SSD on the MB gets no where near your performance. What MB are you using? What drivers?

Funnily enough, that was from an old Asus M2N-SLI board. AHCI enabled in bios. Running Windows 7. No special drivers installed. Nothing crazy. The cells used in the Agility are different I believe from the Vertex though (not to be confused with the Vertex Turbo).

Very best,

 

[garritynet]

True, they are different but the Vertex is nothing special. Not to where it is twice as fast as a Vertex. Heck, its faster than my RAID. If your not doing anything differnt on your board then the problem exist on my own. I know the 9650se is not the best card for SSDs, I knew that when I purchased it, but the AMD raid utility does much worse for me. I blame it on the SB700 for now, but I am still looking for something I can fix.

 

[MalVeauX]

True, they are different but the Vertex is nothing special. Not to where it is twice as fast as a Vertex. Heck, its faster than my RAID. If your not doing anything differnt on your board then the problem exist on my own. I know the 9650se is not the best card for SSDs, I knew that when I purchased it, but the AMD raid utility does much worse for me. I blame it on the SB700 for now, but I am still looking for something I can fix.

Heya,

Really, you don't need a dedicated RAID controller card for RAID0 (especially with less than 4 drives). I would take it out completely and use the onboard unless it's significantly different (which is weird if it is). It does look like something is not right with your drives and settings. Though, your settings should have nothing to do with it honestly. You should be able to have everything set to default. Enable RAID on the SATA controllers (or use your dedicated card). And just make a array and call it a day with nothing to set or worry with. Even your single drive performance is lower than I expected.

Do you have another motherboard you can test the drive(s) on?

Very best,

 

[garritynet]

Nope. This is my only MB. I don't really want to buy another MB just to test with. The card I have no problem using with only HDDs. If I can use SSDs on it, well thats just gravy. I tried with the Asus drivers and the MS drivers in W7 and by far the W7 drivers perform better. I recently upgraded to the generic SB drivers from the AMD website and got a small boost from those.

I am wondering if I have something going on in the bios that is screwing with my PCIe bus. Gonna reset the cmos and see what happens. Also gonna run GC overnight. Maybe the drive is more fragmented than I expected. Other than that I don't know what to do. After that I'm just gonna stop benching my poor SSDs and be happy with my instant performance on everything I do. Even if its not the fastest, its still pretty fast.

 

[garritynet]

I don't get it. I boot into Vista off a HDD and run test on the RAID 0 SSDs and its every bit as fast as I want it to be. I install windows onto the array and its runs at about 1/3 the speed. Not even that, but its much slower than it was before. I can hit 380mb/s on HD Tune if I boot from another HDD and 98mb/s if I boot from the array. Even if I move the temp folders, disable prefetching and all that other stuff.

Is this just an artifical glitch in the testing? Something to do with HD Tune?

 

[Emulex]

try a real benchmark those are benchmark queen programs.

 

[garritynet]

try a real benchmark those are benchmark queen programs.

What "real benchmark" should I try?

 

[Ayah]

The random write power of the WD Raid 0 is pretty sexy I have to say.