Hard Disk Benches

[Flintstone]

Hey guys and gals, why not post some Atto's or Sandra benchmarks for the different hard disk subsystems out there. Describe the controller used, the OS, the hdd's, cache on the drives, and the system it's in etc.. We could come up with a real world comparison of what's available to the masses and keep track of the latest and greatest. Thoughts and or comments anyone?

Thanks,
Flint:cool:

[Tex]

6 disks. All Atlas 10k III's and an LSI Elite 1600 with 128mb cache in a 64bit slot. This was XP and ntfs with everything including indexing enabled and no tweaks

Tex

[Enverex]

Single ATA100 Western Digital 80GB 8MB SE

ATTO - 50,000 (give or take 100 either way) Hit it around 16, 32.

NS

[Flintstone]

3xX15-18 Gig U160 Cheetahs w/4Meg caches in raid 0 on an LSI Mega Raid Elite 1600 with 128 Meg cahce memory in a 64bit/66Mhz slot on an MSI K7DMaster-L with 2XXP1700' modded to MP's @1800+ speeds.

[Tex]

Turn off the cache and use direct IO and the atto's improve. You will hit about 140,000 on reads and writes that way !

Love ya Man !

tex

[Flintstone]

I'm getting a total of 3 - 8 Meg chache Cheetahs in a few days and I'll post those Atto's then. They should do better than the others. I'll tweak the latency then also and see what it brings!!

Anyone got a functioning SATA Raid arry going?

Flint

[EMT]

Well I benched my array after reading this thread and... well, see for yourself.

2x WD 1200JB @ RAID-0 (HP374, KD7-RAID) [edit-16k/16k]
Windows 2000
Cacheman and Powerstrip were not running for this test but normally are... but this much slower than original (I could pull up a pic of that too). Folding was running.


Would 3+ months without defragging cause this?

[Tex]

Folding might do it but a single drive like yours should hit 50,000 near the rim and in raid they should 90 to 100,00 on the outside but maybe only 60 to 65,000 at the very end of the disk.

Is the disk real full?

Tex

[primesuspect]

Flint: I installed a functioning SATA RAID in a customer's computer, using WD Raptor 10K SATA drives on the Promise 376 SATA RAID chipset.

No pics, but Attos were 78K in RAID 0 with no tweaks on XP. Barton 2500.

[Flintstone]

Cool! Anyone else? I want to see if it has any performance advantages vs. pata at this time or if waiting for the second generation would be the thing to do. Prime, based on those numbers, it's neck and neck with pata right now but who knows what a little tweaking might have done.
Also, it seems to me that a properly tuned pata raid 0 with 7200 rpm drives can match it.
So it seems that it's more money than it's worth atm.

Anyone?

Flint

[Tex]

Originally posted by primesuspect
Flint: I installed a functioning SATA RAID in a customer's computer, using WD Raptor 10K SATA drives on the Promise 376 SATA RAID chipset.

No pics, but Attos were 78K in RAID 0 with no tweaks on XP. Barton 2500.

Was it a problem with the promise you think? I thought the individual drives hit 55,000+ ?

My old Maxtor 740's hit 85,000 in raid-0 and teh WD's can hit above 95,000.

Here is an ATTO of a pair of WD's hung off the HPTR sata controller using converters even. This was tweaked with teh pci latency mods as I wrote in my guide. These were from mtngoats nforce rig btw...

Tex

[JPP]

KD7 Raid. Raid 0 mode 2 x 60Gb IBM 120GXP
ATTO write / read score 80700 MB/s to 93600 MB/s
JPP

[EMT]

Originally posted by Tex
Folding might do it but a single drive like yours should hit 50,000 near the rim and in raid they should 90 to 100,00 on the outside but maybe only 60 to 65,000 at the very end of the disk.

Is the disk real full?

Tex

33GB free out of capacity 223GB, feels like it's getting full... I think I'll defrag this week and try ATTO again. But can fragmentation do this or is it merely because it's getting full? Scores were definitely above 80 before.

[Tex]

More because of full but if the remaining is really fragged then it can effect it.

Sandra uses a huge test file so its really a true picture in some ways of what the real performace is. ATTO is only 33 mb so it can hit non-fragged pieces pretty easy.

Tex

[danball1976]

Lets see, these are fairly old, so

[tUx]

If I'm running on a single drive, what should I set for the Total Length value?

[Tex]

Same thing. 32mb

[Keelhaul]

"BOOTY" machine (see below)
160GB Maxtor DM9+ with 8MB cache,
80GB NTFS partition on WinXP SP1a.

[EQuito]

FAT32, more specs below...

[Tex]

EQ have you messed with the pci latency at all ? It appears teh latency for the sata controller needs bumped up maybe?

Tex

[EQuito]

No, I haven't done any tweaks. I'm not into OC'ing or benchmarking as much as I used to... :(

[Tex]

It would take ya 5 minutes if you want to mess with it. Should boost the writes way up.

Tex

[EQuito]

Are you familiar with this board and the SATA corruption issue?
Abit had to release a bios with the option to increase latency to avoid data corruption on dives attached to the SATA ports and/or using RAID0.
Originally, was set to 30us but it had to be raised to 38us or even 1ms in order to avoid corruption.
So until NVidia, SiliconImage and Abit fix the issue I'm afraid threre is no much we can do. Decreasing latency will cause corruption but if you have a suggestion I'm all ears... :D

[Enisada]

I'll post my ATTO's results when I get home. Running 10K Raptors RAID 0 on nforce 2 SATA. TIll then.....

Enisada

[Matt]

Im thinking about getting SATA or SCSI dont know witch one, are how much it's going to be, or should I stick with my ATA66 12g hard drive?

[Enisada]

Here are my SATA Raptors on a nforce 2 Sil 3112 Controllor. ANy advice to make it faster would be appreciated....

[Enisada]

Here are my PATA WD 1200JB on a HPT374 Controllor. Any advice to make it faster would be appreciated....

[Tex]

How was it striped and formated? It should do more like this ..

This is two WD SE's using converters on a HPT sata raid card on a nforce2.

Tex

[Tex]

Oops.. First thing run the test again and change the total length to 32mb

Tex

[Enisada]

Yes going to test right now, I just noticed that....

[Enisada]

Here is the new test format is NTFS everything default. Let me know what I should change. These are PATA...

[Tex]

Thats tons better ! Now check the pci latency of the controller and see if doubling it helps.

Tex

[Enisada]

Here are my SATA scores everything default again....

[Enverex]

System - http://atomnet.co.uk?p=Systems#RedDwarf

Used to bench alot higher (50000/50000) when empty.

NS

[Enisada]

May sound like a noobish question but where to I go to check the pci latency of the controller???

[EMT]

Use PowerStrip, the program more commonly known for overclocking any video card. Right click the system tray icon, Options menu -> Adapter information...

You can cycle through all the devices with the up and down arrows to the right of the device name. The key is the latency. You can change it by unticking "Read only" and raising or lowering as desired. For me, higher latency on the RAID controller means faster writes.

[Flintstone]

Now we're getting somewhere! An Atto that shows an SATA Raid setup outperforming a PATA setup. That is just what I was looking for. Apparently there is some reason to migrate.

Let's see some more if you've got them!

Flint

[EQuito]

Originally posted by Flintstone
Now we're getting somewhere! An Atto that shows an SATA Raid setup outperforming a PATA setup.Flint, look closely, those are Raptors (native SATA/10k rpm) we're talking about...
SATA is no much faster than PATA if using regular IDE drives as demostrated by previous benchmarks.
Only when using highly tweaked WD SE's as shown by Tex you can really tell the difference.

Now, Raptors are a whole new ball game and to be honest, Enisada should be able to get much higher scores out of those than 90+k

[Tex]

Originally posted by Enisada
May sound like a noobish question but where to I go to check the pci latency of the controller???

All you ever wanted to know about pci latency tweaking with powerstrip.

Good Luck !

Tex

http://www.digitalferret.com/sections.php?name=Guides&rop=showcontent&id=3

[Enisada]

Originally posted by EQuito
Flint, look closely, those are Raptors (native SATA/10k rpm) we're talking about...
SATA is no much faster than PATA if using regular IDE drives as demostrated by previous benchmarks.
Only when using highly tweaked WD SE's as shown by Tex you can really tell the difference.

Now, Raptors are a whole new ball game and to be honest, Enisada should be able to get much higher scores out of those than 90+k

How do I go about getting these higher scores...PCI Latency Tweaking???

Enisada

[EQuito]

Originally posted by Enisada
How do I go about getting these higher scores...PCI Latency Tweaking??? Don't ask me dude, ask the master from Texas... :D
I'm just along for the ride with my poor old IBM's... :(

No seriously, I've been out of the loop for too long but Tex will fix you up real good.
I've seen Raptors hit over 100k.

[Enisada]

I keep playing with it until Tex get back to this thread....


Enisada

[EQuito]

Check it out:

[Enisada]

What kind of setup are you using???

[Enisada]

Here is the best I get with my WD setup....

[Enisada]

Here is the best I get with my SATA WD Raptor setup....

[EQuito]

Originally posted by Enisada
What kind of setup are you using??? No, that ain't mine. I told you I've seen Raptors hit over 100k and that's the proof but this isn't it, I just can find the other ones...

thesaucier

NF7-s 1.2 with 2700+ @ 11.5 X 220
corsiar3500 XMS dual@ 6222 on dual
OC Systems level III 9700 pro @420/730
two WD raptor serial drives raid 0

http://forum.abit-usa.com/showthread.php?s=&threadid=13433

[Hatop]

single IBM 15k rpm 18.4 gig drive (model 3z15 or something)
http://hatop.servepics.com/atto.jpg



i think something is off. This is from a brandnew install of XP Pro, FAT32, with almost no disk usage.

[Enverex]

Erm......... note to everyone, save these images as GIF files, they will be about <10k as opposed to 50-60k. Just looking out for the people on Modems.

NS

[Hatop]

it came out 40k larger for me...maybe xp gif encoding wrong or something.but alright, ill change it.

[Tex]

Originally posted by Hatop
single IBM 15k rpm 18.4 gig drive (model 3z15 or something)
http://hatop.servepics.com/atto.gif



i think something is off. This is from a brandnew install of XP Pro, FAT32, with almost no disk usage.

Also you can only really go by the reads on atto with most but not all scsi controllers right now still on XP. They made changes in the filesystems to fix bugs. The writes look artifically low but it only impacts ya in real life doing file copies to the scsi using explorer and running atto. Convert the drive to dynamic and the writes jump way up. Or they also appear to have new options regarding this with win2k3 which I run also so perhaps the next service pack will include new options in XP also.

If thats an older u160 then the scores are maybe close. What controller is it on?

Tex

[Enverex]

Something is very off with that, that GIF is 101k, use something like photoshop of paint shop pro.

NS

[Enisada]

Thats the least of his problems his HD is a hurting unit...

[Hatop]

I suppose i could try to use 2k3 or whatever, but I don't really have it handy. Ill try the dynamic drive option. Also this is running off of a adaptec aic 7891 ultra2 lvd controller. its the only scsi device on it, and i think that ultra2 lvd has a high enough bus speed to keep a single drive going. It seems to perform faster than atto says as I just tested it with HD tach and my average reads are ~45 MB. Also it beats the pants off my ide drives in loading games still, despite atto saying the ide drives are faster in every aspect. Its not new persey, but it is an IBM Ultrastar 36Z15 15k rpm disk. I have read on storagereview that it should be a decently fast drive. Any other reasons why its reads are low?

[Tex]

Just convert to dynamic you will be shocked especially at the top end of the graph. It will jump from them being like 127 to three or four thousand.

I didn't look close at the top end before That looks very familiar!. My lsi u160 did the same thing under XP. Is it empty? Do this. Comvert to dynamic and format it with a cluster size of like 64. Also check the pci latency of the controller itself. That can help but its usually just the writes that get helped.

[Hatop]

WOW did that make a difference.
http://hatop.servepics.com/atto2.jpg



now its dynamic and the reads doubled in some cases
:-D

[Tex]

And now read the pci latency guide I posted and we can kick the writes in the pants too ! (grin)

Your 15 minutes from paradise.

Tex Support at your service as always my friend.

[Enisada]

Hey Tex, so my two RAIDs are all tweak up for the performance that I can expect from them. Sorry to persist just want to get the best out of my system.

[Hatop]

:D :D :D

Tex is yay, thanks much for the help :)

[Tex]

Originally posted by Enisada
Hey Tex, so my two RAIDs are all tweak up for the performance that I can expect from them. Sorry to persist just want to get the best out of my system.

You looked good but if you wanted to tweak anything then check the pci latency. I have never user and have no experiance with that sata controller. It was the sis one wasn't it?

Do they have differant bios and driver versions for it? Tmod our killer bios guy could help ya patch in other bios versions if its on your MB and stuff but the main thing to check would of been the pci latency. Usually a symptom of the latency being set low is the writes lagging way behind the reads and the sata scores were dead on. The other hpt needed some tweaking it seems as the writes were lower. With the hpts differant bios and driver versions seemed to really help also as they have new versions all the time. Promise seems to offer way fewer updates.

Tex

[Tex]

Originally posted by Hatop
:D :D :D

Tex is yay, thanks much for the help :)

But the writes are to low. Did you check the pci latency for the controller. I linked a guide I wrote about adjusting the latency of disk controllers and used a scsi controller as an example. I bet we can kick the writes ass up another 15 to 20,000 also very possibly.

Tex

[Missileman]

Here's a set of Seagate SATA 80's

[Missileman]

Here's a good link. Seems to follow what I have been seeing playing with the SATA drives.

http://www.madshrimps.be/?action=getarticle&articID=69

[Enverex]

ph334r teh Barracuda II (It honesly didnt feel slow)

NS