Intel D975XBX2 - Matrix vs. Marvell Controller Performance

[Matt Dralle]

Greetings,

As a new owner of an Intel D975XBX2 motherboard, I got to wondering which of the two RAID controllers on this board had the better performance. I searched the Intel web site and did a fair amount on the Google, but really came up with zero performance information on these controllers specifically relating to which one was better within the context of this particular motherboard.

In light of the lack of information, I set about getting some information of my own. I built two RAID 1 arrays with two Seagate 750GB SATA II drives on each controller. I also took two 500GB Seagate SATA II drives and put one on each of the two controllers to test the non-RAID, single-disk performance.

I used the latest version of PCMark05 and ran specifically the HDD test suite on each of the configurations. The system is running Windows Vista 32. Below is a chart I made of the raw numbers and they are certainly interesting.

In a nut shell, here's what I found. In RAID 1 configuration, the Intel Matrix is significantly better than the Marvell coming in about 10-15% better in performance overall with some tests showing nearly a 50% increase in throughput.

On the other hand, in the non-RAID configuration it was basically a draw between the two controllers. It is interesting to note the significant benefit in read performance that is afforded by the RAID 1 configuration.

At any rate, these tests aren't particularly meant to be the end-all performance measurement between these two controllers, but it is an interesting comparison. The bottom line is if you plan to use RAID 1 for your system disk, you will do well to use the Intel Matrix over the Marvell controller. If you plan to run in a non-RAID, single disk configuration, it doesn't much matter which controller you opt to use.

Best regards,

Matt Dralle

Comparsion Table:



PDF Version:

http://www.matronics.com/personal/Intel ... ay2007.pdf

JPG Version:

http://www.matronics.com/personal/Intel ... ay2007.jpg

[mickrussom]

In a nut shell, here's what I found. In RAID 1 configuration, the Intel Matrix is significantly better than the Marvell coming in about 10-15% better in performance overall with some tests showing nearly a 50% increase in throughput.

Now that you have proven to the world that the non-WHQL crap Marvell drivers are not only slower in RAID-1, maybe everyone will finally stop using this horrible embarrassing asic.

Why wouldn't they just bond two ICH7R together for more ports? Why mess with this garbage chip? Bad drivers, bad performance unreliable and not Intel.

I wish they would start putting 3ware or some other normal real-hardware-raid chips on these brutally expensive motherboards. I don't mind paying for quality.

[IRQ Conflict]

Why wouldn't they just bond two ICH7R together for more ports? Why mess with this garbage chip? Bad drivers, bad performance unreliable and not Intel.

Hmmm..... Money....naw! couldn't be?!

[JasonH.]

Matt,

As always, you rock, these test results are awesome. I am going to try a couple of benchmarks in Windows XP 32bit and Server 2003 32bit with the Marvell and Intel controller RAID and non RAID. Marvell is actually a partnership with Promise Technologies. These guys make a pretty solid product. Most people are worried about how fast it is, but in this case speed and reliability play a big role depending on how we chose to outfit the system. Can you provide some reliability results for the different controllers using RAID1? Perhaps maybe a rebuild or an array integrity check.

Anyway I am currently using the MARVELL storage controller on the INTEL-D975XBX2 with two seagate 320 gb drives (ST3320620AS SATA 3Gb/s interface) in a RAID1 array. My system is running Windows Server 2003 Enterprise and has about 6 VMware VMs running on it using VMware Server. The performance of the MARVELL controller vs. the INTEL seemed to way in favor of the MARVELL storage controller. I dont have any benchmarks, so I can't prove anything. Again, I will run some tests and let you know of my findings.

As an end note, lets face it Intel is not going to spoil there flagship desktop motherboard by putting a low-end SATA controller on it. I am confident that Intel did there homework. And yes it is all about money, so don't plan on buying an INTEL desktop motherboard with a 3Ware or an Adaptec SATA raid controller on it anytime soon.

Anyway have fun!!!

[mickrussom]

Anyway I am currently using the MARVELL storage controller on the INTEL-D975XBX2 with two seagate 320 gb drives (ST3320620AS SATA 3Gb/s interface) in a RAID1 array.

Hate to break it to you there, but #1, marvell makes bad drivers. Why dont you try and pull out a drive and test the redundancy.

#2, if you care about performance, you would get a real-hardware-raid controller, and not use a piece of crap hacked Fake-RAID controller.

Problem is, I haven't found a PCI-express SAS or SATA2 raid or hba that will run on the D975XBX2.

Anyways, you would be better off with software raid1 on windows 2003 vs fake raid, why not try benching that. its rather fast - faster than fraid.

[Matt Dralle]

Anyway I am currently using the MARVELL storage controller on the INTEL-D975XBX2 with two seagate 320 gb drives (ST3320620AS SATA 3Gb/s interface) in a RAID1 array.

Hate to break it to you there, but #1, marvell makes bad drivers. Why dont you try and pull out a drive and test the redundancy.

#2, if you care about performance, you would get a real-hardware-raid controller, and not use a piece of crap hacked Fake-RAID controller.

Problem is, I haven't found a PCI-express SAS or SATA2 raid or hba that will run on the D975XBX2.

Anyways, you would be better off with software raid1 on windows 2003 vs fake raid, why not try benching that. its rather fast - faster than fraid.

So much hate; damn that Rap music...

Seriouly, though, I would find it hard to believe that software RAID would be even marginally better or equal to even the hardware-assist RAID implimentation, especially under Windows. I have built some systems with Redhat Linux using the software RAID and it does seem to work alright, although I have to say that I have never actually had to recover from a bad drive in that implimentation.

Is there a useful (i.e. built in) way to impliment software RAID on a Windows Vista 32 Ultimate system?

If I had an extra PCI slot in my system, I would definately utilize a full-on hardware RAID implimentation such as LSI or Adaptec. But, alas, with two NVIDIA 8800 GTX cards, one 7800 GTX, and one Creative SB X-Fi card, I don't have any room (the 8800's take up two fricken slots!).

I read a review somewhere on Internet that was comparing a couple of different hardware-based RAID adapters against the Intel Matrix ICH7R on the D975XBX2. Across the board, the ICH7R was typically one-third of the performance of any of the other 3rd party hardware RAID adapters.
So I guess, if you've got the slot avaialble, that would be the way to go.

I do have to echo someone else's comment regarding the quality of parts put on these motherboards. Why not have a version of the D975XBX2 with a real, honest-to-goodness RAID adapter? Frankly, I'd pay another $200 for a motherboard with this capability. I remember having some Supermicro motherboards that actually had a little special slot just for a plug-in Adaptec RAID controller. Now that was cool. I want that.

Matt

[mickrussom]

Seriously, though, I would find it hard to believe that software RAID would be even marginally better or equal to even the hardware-assist RAID implementation, especially under Windows. I have built some systems with Redhat Linux using the software RAID and it does seem to work alright, although I have to say that I have never actually had to recover from a bad drive in that implementation.

Is there a useful (i.e. built in) way to implement software RAID on a Windows Vista 32 Ultimate system?

If I had an extra PCI slot in my system, I would definitely utilize a full-on hardware RAID implementation such as LSI or Adaptec. But, alas, with two NVIDIA 8800 GTX cards, one 7800 GTX, and one Creative SB X-Fi card, I don't have any room (the 8800's take up two fricken slots!).

I read a review somewhere on Internet that was comparing a couple of different hardware-based RAID adapters against the Intel Matrix ICH7R on the D975XBX2. Across the board, the ICH7R was typically one-third of the performance of any of the other 3rd party hardware RAID adapters.
So I guess, if you've got the slot available, that would be the way to go.

I do have to echo someone else's comment regarding the quality of parts put on these motherboards. Why not have a version of the D975XBX2 with a real, honest-to-goodness RAID adapter? Frankly, I'd pay another $200 for a motherboard with this capability. I remember having some Supermicro motherboards that actually had a little special slot just for a plug-in Adaptec RAID controller. Now that was cool. I want that.

Matt

Legit just did a review on a promising RAID card, Adaptec RHR(Real Hardware Raid) 3405 / 3805.

http://legitreviews.com/article/508/1/

Funny that the bios shot is from a Dell BIOS:
http://www.legitreviews.com/images/revi ... 40x480.JPG

Windows Server only supports RAID1 and RAID5. XP and Vista Ultimate (for whatever stupid ass reason, **** microsoft for screwing ultimate) only support striping. When you use software raid you have the unpleasant requirement to make the disks "dynamic" which makes mounting them in case of emergency a hell of a lot harder. I traditionally want a raid card where you can break a mirror and put the disk in another machine, being about to "ghost" or dupe disks with the card. Most raid cards SCREW this up. This industry is a sad joke, the point of RAID is to save data, not pepper it with proprietary crap like dynamic disks, signatures, etc.

Vista 32 and 64 got the shaft with RAID1 and RAID5, it only does raid0/stripes.

You want to use that 3rd PCI-express slot for a RHR HBA, but a lot of people have trouble with this board with add-in HBAs from LSI, Areca, Promise. PCI is pretty much a dead thing, so I pretend it doesn't exist anymore. I could see PCI-X/133 slots being useful, but PCI-32-33Mhz is pathetic, and express is cleaner due to the lanes beind dedicated to the chipset.

I think I may have made that comment. I would pay supermicro prices any day for a no-compromise design. Do not lust after that dedicated raid slot, ZCR or zero channel raid, those cards have traditionally sucked hard.

You want an asic that is equivalent to a retail boxed PCI-express HBA with real hardware raid - no corners cut. ZCRs cut some corners.

Best of luck in the holy war against Marvell and ****-raid.

[JasonH.]

mickrussom,

Thanks for educating us. We would have never figured this out. Lets re-wind and take a journey to the first posting that Matt published. See the first two sentences?, here let me point them out.

As a new owner of an Intel D975XBX2 motherboard, I got to wondering which of the two RAID controllers on this board had the better performance. I searched the Intel web site and did a fair amount on the Google, but really came up with zero performance information on these controllers specifically relating to which one was better within the context of this particular motherboard.

Clearly you missed the point, we all know that the built-in RAID controller or controllers for that matter have their limitations and that if you want a real RAID controller you would need to buy one. We are trying to determine which of the two built-in RAID controllers on the Intel 975XBX2 desktop motherboard performed the best, and offer the most reliable solution, since we are not able to find any of this information on the internets. Got it!

Stay with me now. You mentioned that I would be better off using software RAID instead of um...fraid? Thanks chief.

Anyways, you would be better off with software raid1 on windows 2003 vs fake raid, why not try benching that. its rather fast - faster than fraid.

With that being said it leads me to believe you have little or no real world experience setting up, managing, or mantaining any form of RAID for redundancy or performance purposes. You would know better than to use software RAID vs. two built-in RAID controllers or even suggest it for that matter.

Anyway Thanks for the review links atleast they were helpful.

P.S. Does your Mommy know you cuss like a sailor in these forums.

Windows Server only supports RAID1 and RAID5. XP and Vista Ultimate (for whatever stupid ass reason, **** microsoft for screwing ultimate) only support striping.

Maybe I could provide her a couple of good links with reviews to a great bar of soap.

-Jason

[mickrussom]

With that being said it leads me to believe you have little or no real world experience setting up, managing, or maintaining any form of RAID for redundancy or performance purposes. You would know better than to use software RAID vs. two built-in RAID controllers or even suggest it for that matter.

Interesting. I've setup software raid on production servers running Solaris (RAID "10" is a favorite using Solaris Volume Manager/DiskSuite/meta* commands), used Software RAID on NT 4, 2000 and 2003, and have a bit of experience with Linux software raid (md and LVM), and I've also tinkered with Sun's ZFS.

I guess Veritas, and all "those" volume managers are all "crap" based on what you are saying here, that a good volume manager is worse than fake-raid chips.

All I asked is that a comparison would be nice to see with Windows 2003 setup with dynamic disks in a mirror. I think you might be surprised. It has been shown in the Linux world md beating "fake raid" controllers time and time again.

See: http://linux-ata.org/faq-sata-raid.html for a start.

Oh, and I'll go over to one of these systems I didn't setup or know anything about and dump a config for you:



These mirror and mirrored stripe arrays are so fast that many people mount these volumes over NFS and the performance of these volumes over NFS is comparable (most of the time faster) than the local SATA disk.

Sun Microsystems Inc. SunOS 5.9 Generic January 2003

#uptime
2:26am up 497 day(s), 14:36, 1 user, load average: 0.12, 0.16, 0.16

# metastat
d60: Mirror
Submirror 0: d40
State: Okay
Submirror 1: d50
State: Okay
Pass: 1
Read option: roundrobin (default)
Write option: parallel (default)
Size: 573235200 blocks (273 GB)
d40: Submirror of d60
State: Okay
Size: 573235200 blocks (273 GB)
Stripe 0: (interlace: 64 blocks)
Device Start Block Dbase State Reloc Hot Spare
c0t2d0s0 0 No Okay Yes
c0t3d0s0 0 No Okay Yes
d50: Submirror of d60
State: Okay
Size: 573235200 blocks (273 GB)
Stripe 0: (interlace: 64 blocks)
Device Start Block Dbase State Reloc Hot Spare
c0t4d0s0 0 No Okay Yes
c0t5d0s0 0 No Okay Yes
d37: Mirror
Submirror 0: d17
State: Okay
Submirror 1: d27
State: Okay
Pass: 1
Read option: roundrobin (default)
Write option: parallel (default)
Size: 53136000 blocks (25 GB)
d17: Submirror of d37
State: Okay
Size: 53136000 blocks (25 GB)
Stripe 0:
Device Start Block Dbase State Reloc Hot Spare
c0t0d0s7 0 No Okay Yes
d27: Submirror of d37
State: Okay
Size: 53136000 blocks (25 GB)
Stripe 0:
Device Start Block Dbase State Reloc Hot Spare
c0t1d0s7 0 No Okay Yes
d35: Mirror
Submirror 0: d15
State: Okay
Submirror 1: d25
State: Okay
Pass: 1
Read option: roundrobin (default)
Write option: parallel (default)
Size: 40962240 blocks (19 GB)
d15: Submirror of d35
State: Okay
Size: 40962240 blocks (19 GB)
Stripe 0:
Device Start Block Dbase State Reloc Hot Spare
c0t0d0s5 0 No Okay Yes
d25: Submirror of d35
State: Okay
Size: 40962240 blocks (19 GB)
Stripe 0:
Device Start Block Dbase State Reloc Hot Spare
c0t1d0s5 0 No Okay Yes
d34: Mirror
Submirror 0: d14
State: Okay
Submirror 1: d24
State: Okay
Pass: 1
Read option: roundrobin (default)
Write option: parallel (default)
Size: 16384320 blocks (7.8 GB)
d14: Submirror of d34
State: Okay
Size: 16384320 blocks (7.8 GB)
Stripe 0:
Device Start Block Dbase State Reloc Hot Spare
c0t0d0s4 0 No Okay Yes
d24: Submirror of d34
State: Okay
Size: 16384320 blocks (7.8 GB)
Stripe 0:
Device Start Block Dbase State Reloc Hot Spare
c0t1d0s4 0 No Okay Yes
d33: Mirror
Submirror 0: d13
State: Okay
Submirror 1: d23
State: Okay
Pass: 1
Read option: roundrobin (default)
Write option: parallel (default)
Size: 4098240 blocks (2.0 GB)
d13: Submirror of d33
State: Okay
Size: 4098240 blocks (2.0 GB)
Stripe 0:
Device Start Block Dbase State Reloc Hot Spare
c0t0d0s3 0 No Okay Yes
d23: Submirror of d33
State: Okay
Size: 4098240 blocks (2.0 GB)
Stripe 0:
Device Start Block Dbase State Reloc Hot Spare
c0t1d0s3 0 No Okay Yes
d30: Mirror
Submirror 0: d10
State: Okay
Submirror 1: d20
State: Okay
Pass: 1
Read option: roundrobin (default)
Write option: parallel (default)
Size: 24577920 blocks (11 GB)
d10: Submirror of d30
State: Okay
Size: 24577920 blocks (11 GB)
Stripe 0:
Device Start Block Dbase State Reloc Hot Spare
c0t0d0s0 0 No Okay Yes
d20: Submirror of d30
State: Okay
Size: 24577920 blocks (11 GB)
Stripe 0:
Device Start Block Dbase State Reloc Hot Spare
c0t1d0s0 0 No Okay Yes
d31: Mirror
Submirror 0: d11
State: Okay
Submirror 1: d21
State: Okay
Pass: 1
Read option: roundrobin (default)
Write option: parallel (default)
Size: 4098240 blocks (2.0 GB)
d11: Submirror of d31
State: Okay
Size: 4098240 blocks (2.0 GB)
Stripe 0:
Device Start Block Dbase State Reloc Hot Spare
c0t0d0s1 0 No Okay Yes
d21: Submirror of d31
State: Okay
Size: 4098240 blocks (2.0 GB)
Stripe 0:
Device Start Block Dbase State Reloc Hot Spare
c0t1d0s1 0 No Okay Yes
Device Relocation Information:
Device Reloc Device ID
c0t4d0 Yes id1,sd@SSEAGATE_ST3146807LC
c0t5d0 Yes id1,sd@SSEAGATE_ST3146807LC
c0t2d0 Yes id1,sd@SSEAGATE_ST3146807LC
c0t3d0 Yes id1,sd@SSEAGATE_ST3146807LC
c0t1d0 Yes id1,sd@SSEAGATE_ST373307LC
c0t0d0 Yes id1,sd@SSEAGATE_ST373307LC

I like the uptime of over a year on a busy box. Yeah, I don't know much about this stuff.

(I've also done extensive testing with bonnie++, iozone, lftest, etc).

[Matt Dralle]

Ah, well, I'm not sure about all of that, but I do know that my clients report being far more satisfied when I let them use all of my 3.5" hard drive. What size do you give them?

Yeah, this sounds a lot like that...

Matt

[JasonH.]

Mick,

Stick to the context. We are talking about an INTEL975XBX2 motherboard. Marvell SATA controller vs. Intel SATA controller nothing more, nothing less. We are not trying to prove anything other than which SATA controller on the motherboard performs the best and offers the most reliable solution. Clearly in the given situation software RAID is not the favorable solution for many reasons. Be helpful not hostile.

[stevefal]

Sorry to take this off topic, but it seems folks on this thread may have an answer to this.

I'd like to use an Intel 975XBX2 motherboard with Windows Server 2003 (32 bit). Intel's site says the board doesn't support the OS - "drivers not available"

But people here seem to be using it anyway. Is this just Intel avoiding the support cost wrt this desktop board, or are there known issues?

Do you see any problems with this low-end server?

Windows Server 2003
975XBX2 motherboard
E6600 Core 2 Duo
4GB 667
(2) SATA RAID 1
SATA DVD

[kenc51]

Sorry to take this off topic, but it seems folks on this thread may have an answer to this.

I'd like to use an Intel 975XBX2 motherboard with Windows Server 2003 (32 bit). Intel's site says the board doesn't support the OS - "drivers not available"

But people here seem to be using it anyway. Is this just Intel avoiding the support cost wrt this desktop board, or are there known issues?

Do you see any problems with this low-end server?

Windows Server 2003
975XBX2 motherboard
E6600 Core 2 Duo
4GB 667
(2) SATA RAID 1
SATA DVD

Should be no problem, I've used 2003 server for a while and if I couldn't find a driver, I'd just use the XP one instead...they're basically the same OS...
I haven't got a XBX2 board though.

[mickrussom]

Do you see any problems with this low-end server?

The only problems I noticed after slipstreaming the RAID/AHCI Intel Storage Matrix drivers with Nlite and SP2 with Windows 2003 was the audio didn't work well with all games.

If you aren't gaming, I cant think of anything that would be an issue.

[mickrussom]

they're basically the same OS...

I disagree with that statement.

Windows 2003 SP1 = Windows XP x64. Meaning, Windows XP x64 is the same OS as Windows 2003 x64.

However, windows 2003 and XP 32 bit are very, very different.

Now, if you say, DirectX 9 is the same API and works the same in both, thats ok, and you could say they have the same driver model, and that's ok too.

But the OS is not the same. I find Windows 2003 to be vastly superior to XP is just about every department - including speed. Microsoft was busy doing something in the years between the two OSes. Mostly refactoring and making stuff work, bug fixes.