4 x Samsung SP2504C HDTune RAID0/1/5/10 Performance results

Silencing hard drives, optical drives and other storage devices

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

Post Reply
Blizzboy
Posts: 5
Joined: Fri Sep 09, 2005 8:31 pm
Location: Van Diemens Land

4 x Samsung SP2504C HDTune RAID0/1/5/10 Performance results

Post by Blizzboy » Fri Sep 09, 2005 8:56 pm

Hi There,

I thought some people might be interested to see the results of using 4 x Samsung SP2504C SATA II 250G hard disks in various RAID configurations on a Gigabyte K8N Ultra 9 NForce 4, AMD64 3000, Win XP and the latest NVidia drivers. The mother board also had a Silicon Image SiI3114 RAID controller (but only SATAI).

I tried all combinations of 1x SATA Drive (no RAID), RAID-0 (4 drives), RAID1 (2 drives), RAID 5 (4 drives), RAID 10 (4 Drives) to see the differences in throughput. I wasn't able to use the Samsung Hdutil to control the Acoustic management settings (seemed to only work it PATA drives?), nor was I able to confirm that NCQ was or wasn't enabled... It is interesting to see that SiI3114 RAID 5 performance is pretty poor. Not sure why though.

(Images removed from website - see updated post at the end of this thread)

As for the acoustic performance of 4 of these drives.. I have them mounted in an Antec P180 lower HD chamber. Their vibrations during head seeks are very well damped from the case, and the IDLE noise is barely audible with my ear right up against the front of the case. I found it hard to tell if they were running and could only tell by the centrigual resistance force when I moved all 4 of them!! All in all a very nice drive for the cost - equal or less than most other 250G drives.
Last edited by Blizzboy on Tue Jan 24, 2006 10:55 pm, edited 1 time in total.

vine-au
Posts: 85
Joined: Sat Sep 03, 2005 6:24 pm
Location: melbourne, australia
Contact:

Post by vine-au » Fri Sep 09, 2005 10:04 pm

pictures don't load for me???

myshkin
Posts: 14
Joined: Wed Jun 01, 2005 11:48 am

Post by myshkin » Sat Sep 10, 2005 7:31 pm

Nice benchmark set there, for anyone wondering about different configs. I'm kind of wondering if you have a configuration problem tho, as far as the first pic on the page, the 4x Raid0 result. I'm running two 80 gig sata1 SP0812C in raid 0 and get this result (live disks though, running the installed os and winamp playing from the drive, as well as steam and mbm5 etc, so it's a bit wavy):

Image

That's 32k stripe size, nforce 5.10 drivers. The chipset is nforce3 250, nvraid bios 4.42. Also tried the 4.84 and 5.30 nvraid bioses, as well as the 6.66 nforce drivers, little difference.

16k stripe size nets me an extra 5-10MB/s in benchmarks, but feels slower for desktop usage, i suspect due to the longer I/O queues at 16k when multitasking. so i stick with 32k now. 64k, which i assume you were using(?), drops things by about 10MB/s, and seemed a little slower on dekstop usage, but still less 'laggy' than 16k.

Looking at your results tho, my average with 2x sata1 is higher than your average with 4x sata2, which can't be right. Your single drive benchmark is identical to my single drive results though, about 58MB/s. Looking at the 4xraid0 graph, i noticed you're getting no tail-off as the head reaches the end of the platter, like you do for the single drive. So it looks to me like something is artificially limiting the throughput way below what the 4x drives at the beginning of their platters are capable of delivering. Try some older/newer drivers maybe?

Blizzboy
Posts: 5
Joined: Fri Sep 09, 2005 8:31 pm
Location: Van Diemens Land

Post by Blizzboy » Tue Sep 13, 2005 12:35 am

I had wondered why I didn't get much of a performance boost from 4 x drives in RAID 0. I am using the latest NVidia drivers (6.66). I also tried the silicon image controller and got a similar result (ie flat) but with 20MB/s less throughput. Hopefully it is not a CPU / motherboard limitation?. I am not sure what else to tweak/test. Followups appreciated..

At least I won't try the RAID5 option with the Silicon Image SI3114 controller.. seriously poor performance, but perhaps that is the cost of parity.

myshkin
Posts: 14
Joined: Wed Jun 01, 2005 11:48 am

Post by myshkin » Tue Sep 13, 2005 3:36 am

From what i've heard, the silicon images raid controller has some bandwidth issues. I'll try to explain as much as i know about the issue. My own board is a gigabyte K8NS s754 nforce3 250 chipset, and only provides 2 sata ports. They're attached to the internal nforce chipset raid controller, and the available bandwidth is the full Athlon64 HT bus speed (2x800Mhz). The ports are located quite near the AGP slot. Most nforce based A64 motherboards, especially the nforce4 boards, include a second pair of sata ports, usually located near the bottom of the motherboard. These two ports are connected to the silicon images controller. From what i gather, the silicon images controller is not connected into the HT bus directly, but communicates via the pci bus, essentially like a hardwired pci card. That gives you the limitations of the pci bus speed, which is 33mhz, severely limiting the throughput of SATA drives:

"The normal PCI bus (whether operating at 3.3V or 5V) is 32 bits wide and runs at 33.3MHz (normally quoted as 33MHz), offering a maximum bandwidth of 133MB/second."

Actually i've just looked up your K8N-Ultra-9 and it does offer 4 ports on the nvidia controller, as well as 4 more on the silicon images. I don't mean to patronise but it'd be worth making sure all 4 drives are really on the nvidia controller only, just to rule out the silicon images pci bus issue.

My next guess would be the 6.66 nforce drivers. If you visit www.nforcershq.com forums, apparently the sata and raid drivers were rewritten for nforce4 boards in the 6.66 set, and the 6.66 drivers require an nvraid bios 4.84 or greater. I know gigabyte are really slow updating the nvraid component of their bioses, so that may be the problem. Take a look at the screen after your bios posts, where the nvraid bios runs and declares your stripe healthy, it should give a version number in the first line on the screen. It might be 4.81. Apparently bad things will happen with the 6.66 ide drivers on anything below 4.84, whether it's corruption or poor performance, i'm not sure.

If your nvraid bios version is 4.81 or less than 4.84, try installing the ide component of these drivers - http://www.nvidia.com/object/nforce_nf4_winxp_6.39.html and rerun the hdtune test after rebooting.

If that turns out to be the problem, and you want the latest nvraid bios so you can run the 6.66 driver set, i'll update the latest K8N-Ultra-9 bios with the nvraid 4.84 rom for you and post it here. I found out how to do that recently because gigabyte kindly haven't updated the nvraid for my board since version 4.42!

But If you're still getting poor (~100MB/s) throughput from 4xraid0 though, and a driver change makes no difference, and your nvraid bios is up to date, i'd suggest testing smaller arrays and then all disk/port combinations. Connect only two drives to sata ports 1&2, set up a raid 0 with 32k block, and run hdtune. Then swap both drives out for the other 2, again on ports 1&2, and see what graph they give. Then try a 3 disk array. Basically you'd need to trial and error all 4 ports and disk combinations to rule out them being the problem.

myshkin
Posts: 14
Joined: Wed Jun 01, 2005 11:48 am

Post by myshkin » Wed Sep 14, 2005 12:31 pm

I've updated your K8NUltra-9 bios with the nvraid 4.84 bios, if you just want to download and flash it - http://www.vapulus.com/K8NU-9.F6e.zip . Also i've written up how it's done, as you wanted to know.

Ok here's a howto on replacing the nvraid bios within the main motherboard bios. Posted here so it's shows up in google eventually, should anyone else need to do this. Award bioses are modular, meaning they're made up of seperate modules, eg, the mainboard bios, the graphic logos (epa etc), and several small bioses for the onboard components like soundcard, network, and raid controllers.

to view and edit the modules within an award bios, you can use a program called cbrom, which operates on the bios rom's downloadable at your motherboard makers website. Cbrom can be downloaded at www.biosmods.com which also contains some sparse details on how to use it.

cbrom207 - http://www.biosmods.com/download/award/cbrom207.zip

We'll look at using cbrom to replace the nvraid bios component of the F6 gigabyte K8N-Ultra-9 bios with an updated version.

First we need a copy of a more recent nvraid bios to insert into the F6 bios, which if you try to google for, is a hard thing to find. The best method is to find a bios for a different motherboard that uses the same chipset, and is updated and maintained much better than the gigabyte bios. A bit of googling reveals the asus a8n-e nforce4 ultra board has the nvraid 4.84 bios included in it's version 1006 bios, so let's download that:

http://dlsvr02.asus.com/pub/ASUS/mb/soc ... NE1006.zip

also we need to download the F6 bios for the gigabyte k8n-ultra-9.

extract all three zip files to c:\cbrom207 then delete everything except cbrom.exe, K8NU-9.F6 and A8NE1006.BIN. You don't need to do that, but it saves on clutter in the directory.

get the command prompt up from start - all program - accessories.

Code: Select all

cd c:\cbrom207
First let's look at the gigabyte bios.

Code: Select all

cbrom.exe K8NU-9.F6 /d > gig.txt
this is telling cbrom to operate on the F6 bios file with the /d flag, to display it's contents, and send (>) the output to a file called gig.txt, as the output is larger than the command prompt window can display.

Open gig.txt, we'll see the nvraid bios comes under pci device drivers, as pci driver A, module 8. on the list.

To be honest, i haven't had time to look into whether the order of the modules within the bios are important. It's possible the bios calls modules based on name, but it may also call modules based on their byte offset within the rom. So although it might be unneccesary, i always replace modules in the same position as they were removed, which adds a few more steps. Anyone willing to test position changes effects on their bios, let us know how it goes, personally i like to keep things as they were rather than risk a bad bios flash. To replace the nvraid bios and keep it at position 8, we need to remove all following modules in the bios in order to add them after the new nvraid bios module later. We could 'extract' (this flag extracts a module to disk) the nvraid bios first then 'release' it (the release flag removes a module from the bios), then add our new one. However this would move module 9 to module 8, 10 to 9, etc, and our nvraid bios would become module 14.

So, let's look at gig.txt, position 8 onward:
8. PCI driver[A] 0C000h(48.00K)064B4h(25.18K)NVRAID.ROM
9. PCI driver 10000h(64.00K)07AFFh(30.75K)NVPXE.NIC
10. PCI driver[C] 0F000h(60.00K)07264h(28.60K)5139.BIN
11. PCI driver[D] 0DE00h(55.50K)08768h(33.85K)yukpxe.lom
12. LOGO1 ROM 00B64h(2.85K)00520h(1.28K)DBIOS.BMP
13. OEM4 CODE 06DC6h(27.44K)03264h(12.60K)FINER.BIN
14. OEM0 CODE 02549h(9.32K)01B13h(6.77K)dbf.bin


The order things are extracted doesn't matter. But we need a copy of each on the disk. I'm going to extract each module and release it (delete it from the bios) in the module order. For the pci modules, you'll be given a choice A to D which module you want. Press enter after each line.

Code: Select all

cbrom.exe K8NU-9.F6 /PCI extract
A
cbrom.exe K8NU-9.F6 /PCI extract
B
cbrom.exe K8NU-9.F6 /PCI extract
C
cbrom.exe K8NU-9.F6 /PCI extract
D
cbrom.exe K8NU-9.F6 /LOGO1 extract
cbrom.exe K8NU-9.F6 /OEM4 extract
cbrom.exe K8NU-9.F6 /OEM0 extract
cbrom.exe K8NU-9.F6 /PCI release
A
cbrom.exe K8NU-9.F6 /PCI release
B
cbrom.exe K8NU-9.F6 /PCI release
C
cbrom.exe K8NU-9.F6 /PCI release
cbrom.exe K8NU-9.F6 /LOGO1 release
cbrom.exe K8NU-9.F6 /OEM4 release
cbrom.exe K8NU-9.F6 /OEM0 release
rename NVRAID.ROM NVRAID.ROM.4.81
That last line renames nvraid.rom to show it is the 4.81 version. Next we need to get the nvraid 4.84 module out of the asus bios.

Code: Select all

cbrom.exe A8NE1006.BIN /d > asus.txt
Open up asus.txt, we can see the nvraid bios is module 13, pci driver A.
13. PCI driver[A] 0C000h(48.00K)064EDh(25.23K)NVRAID.ROM
Let's extract it.

Code: Select all

cbrom.exe A8NE1006.BIN /PCI extract
A
Now the file NVRAID.ROM in our cbrom207 directory is the 4.84 version. All we need to do is put it in the K8NU-9.F6 bios and replace the modules that come after it. Refer to the gig.txt output we generated from the original K8NU-9.F6 for the module order and type. Notice when we extracted YUKPXE.LOM (pci driver D) it's name became uppercase, we'll correct that back to the original lowercase before reinserting it. Same for DBF.BIN.

Code: Select all

cbrom.exe K8NU-9.F6 /PCI NVRAID.ROM
cbrom.exe K8NU-9.F6 /PCI NVPXE.NIC
cbrom.exe K8NU-9.F6 /PCI 5139.BIN
rename YUKPXE.LOM yukpxe.lom
cbrom.exe K8NU-9.F6 /PCI yukpxe.lom
cbrom.exe K8NU-9.F6 /LOGO1 DBIOS.BMP
cbrom.exe K8NU-9.F6 /OEM4 FINER.BIN
rename DBF.BIN dbf.bin
cbrom.exe K8NU-9.F6 /OEM0 dbf.bin
cbrom.exe K8NU-9.F6 /d > gig2.txt
rename K8NU-9.F6 K8NU-9.F6e
All done, check gig2.txt is in the same order and filenames as gig.txt. The bios is renamed to .F6e indicate it is the edited version (we don't want to lose it and do all that again ;) ). Put it on a floppy and flash from dos, or use the gigabyte window bios update app. Reboot, check the nvraid bios version that is displayed after the bios post screen.

Blizzboy
Posts: 5
Joined: Fri Sep 09, 2005 8:31 pm
Location: Van Diemens Land

Thanks!

Post by Blizzboy » Thu Sep 15, 2005 8:37 pm

Hi There,

Thanks heaps for the information and links!! I would have spent ages trying to find that out myself.

I tried the modified bios with Nvraid 4.84 in my GA-K8N-Ultra-9 MB and it works, ie boots, and the O/S runs fine, but there was no improvement in the 4 x Samsung SATA striping throughput so I'll have to double check my drivers, and experiment some more with other combinations of drives/chunk sizes to see if I have something really wrong.

Thanks again.

_MarcoM_
Posts: 294
Joined: Thu Jul 21, 2005 11:58 am

Post by _MarcoM_ » Fri Sep 16, 2005 3:37 pm

Hi all, i'm new here :)

Very nice tests Blizzboy, great job. I have an MSI K8n neo4 platinum, same controller onboard (nvraid and sil3114). For the poor performance of the nvraid, try to disable NCQ and read caching under the windows hardware options, i heard this will increase performance in certain systems.

I have a question about sil3114: i want to configure a raid 10 with 4 Seagate 7200.7 200GB, it's better than the raid 01 under nvraid (10 is a little more secure). For you, there are problems with multimedia applications or gaming using the silicon image pci controller? The bandwith is a bottleneck, ok, but there are other issues about the use of a pci controller under heavy read/write conditions?

bshor
Posts: 19
Joined: Sun Aug 21, 2005 8:53 pm

Post by bshor » Sat Sep 24, 2005 3:49 pm

I'm planning to implement software Raid 5 with 3 of the same 250Gb SpinPoint drives. It's processor-intensive, but I figure with a AMD64 X2 3800, I'll have some processing power to spare. And the price point is nice.

Blizzboy
Posts: 5
Joined: Fri Sep 09, 2005 8:31 pm
Location: Van Diemens Land

Post by Blizzboy » Tue Jan 24, 2006 10:44 pm

Hi There,

I thought I'd follow up on my original post with an update on the use of 4 x Samsung SP2504C 250G SataII drives in a Nvidia RAID configuration. I finally managed to get the optimum settings :

My setup is

* Gigabyte K8N Ultra 9, Socket 939 Nividia NForce 4 chipset
* AMD 3000+ Venice SKT 939 CPU, standard non-tweaked speed/BIOS settings
* WIndoze XP SP2 with all patches
* Nvidia Raid BIOS 4.84 (using the latest BIOS download from Gigabyte which is K8 version). Note - I tried the previous patching of the K6 BIOS as described in previous posts to this thread, but it didn't have the desired results.. so I waited for the official Gigabyte updates).
* NVidia Forceware WHQL Drivers (81.98_forceware_winxp2k_english_whql.exe)
* 4 x Samsung SP2504C 250G SATAII NCQ, mounted in a Antec P180 (bottom hard drive cage). Drives connected to the NVIDIA SATA ports, I didn't use the PCI based Silicon Image RAID controller. Not sure how usefull NCQ is..
* I used the Hitachi SATA drive utility to configure the drives to use AAM, value 128 (minimum noise?). The samsung utility doesn't recognise these SATA drives on the Gigabyte MB (possible SATA compatibility issues?)

The optimum NVIDIA RAID0 4 disk array needs to have the chunk size seto to 4K or 8K. (16K and above (including the 'optimum!') incur a large performance hit. RAID1 and RAID0/1 all pretty much revert back to the throughput of a single Hard disk. Raid 5 performance is pretty poor - averaging less than 30MB/sec?? (20 - 30% of RAID0).

So here is a disk test result with HD Tune.. not bad results at 190MB/sec burst, but high CPU usage and average access time (overheads of the NVIDIA Raid drivers I expect?).

A word of warning .. I tried to upgrade my MB Bios and then 'upgrade' the NVIDIA RAID drivers on an existing windows XP system but it went pear shaped :cry: so I had to re-install WIndows XP from scratch with the desired drivers.. the possible moral is start from scratch when updating raid drivers/bios combinations ... The upgrade didn't work because the 4 SATA drives only appeared as single drives and I couldn't get the NVIDIA RAID utility (MediaShield) to recognise them as available for RAID array even though the BIOS setup utility said all was ok.

Oh, and a note on the most important aspect - Noise! - the 4x Samsung SP2504C's with AAM enabled are very quiet ... but as I have no test equipment I can't reference it to anything else. If I was running them outside of a P180 suspended hard drive cage then I suspect I would be able to hear them. Not seeking they are very quiet, but you can hear the seeking when you put your ear up against them....

Image :?
Last edited by Blizzboy on Sat Jan 28, 2006 1:59 am, edited 1 time in total.

midiman
Posts: 33
Joined: Fri Jun 25, 2004 3:39 am

Quiet?

Post by midiman » Fri Jan 27, 2006 5:42 am

Just bought four 2504c's - looking to just use them standalone right now, but was planning on a raid once i move to sata2, 3.0gb/s.

I'm a bit disapointed with the seek noise this drive produces. I have had seagate's all my life - mostly 7200.7's, and recently a 7200.8. I currently have my case open with the two 7200.7's and the new 2504c, and the 2504c is incredibly loud on seeks. I can't quiet get an idea on idle noise however - I would say its quieter on idle.

I'm running the HUTIL now to enable AAM......seek noise has *definitely* been extremely reduced, but its still louder than the Seagates.


I think I read someplace that these get quieter after about a week. Can anyone else confirm?

Thanks for the RAID results. Very helpful.

nici
Posts: 3011
Joined: Thu Dec 16, 2004 8:49 am
Location: Suomi Finland Perkele

Post by nici » Fri Jan 27, 2006 9:16 am

I have two SP2504C drives im trying to set up in RAID1 as a storage partition, i get as far as enabling it in bios and getting to the raid configuration tool and choosing the drives and set it to mirroring, boot into windows and i cant see the drive. They are formatted to NTFS.

An hour of googling didnt turn up anything else than a broken link to some nForce 5.06(i think) RAID WHQL thing or something.

The mobo is a DFI nF4 Ultra-D.

The cd included with the mobo only includes SI3114 drivers, that chip exists only on DR models though.

If i format the drives then setup them in the raid utility thingie, they turn up in Disk Management as "Unallocated" and both drives visible as a separate normal drive. I´ve tried both formatting so they have a drive letter and the way they dont..

Any help? :)

Tzeb
Posts: 144
Joined: Sat Apr 30, 2005 4:29 pm

Post by Tzeb » Fri Jan 27, 2006 4:34 pm

Just got 2 sp2504p in raid 0.
OMG....i had to check 2 times to see if AAM is disabled (it is)... i can't belive how silent the seeks are. 8) One is inside a driveaway and the other in a novibes and both are built in 01/2006 :D
The problem is i can't read any temp from them in windws and SMART info is also missing. With their utility Hutil i can read the temps from dos and i also checked that SMART is enabled...but still nothing in windows.
Is there a "fix" for this, or no temp can be read from windows in raid 0?
Image Image

Blizzboy
Posts: 5
Joined: Fri Sep 09, 2005 8:31 pm
Location: Van Diemens Land

Post by Blizzboy » Sat Jan 28, 2006 2:13 am

Hi There - some input for previous posts..

I think the reading of the drive temperatures is not supported when the drives are in RAID configurations - a function of the driver S/W I expect... (I cannot see the RAID drive temperatures with any tools I have used).

NICI : Which drivers are you using for your RAID setup on the DFI NF4 Ultra D? The latest WHQL drivers from the NVIDIA website are worth the download - assuming you have done that?

CHeers
Blizzboy.

nici
Posts: 3011
Joined: Thu Dec 16, 2004 8:49 am
Location: Suomi Finland Perkele

Post by nici » Sat Jan 28, 2006 5:46 am

Oopsie... :oops: I did check the nVidia driver download page yesterday, but i somehow managed to miss the windows download stuff and only noticed the Linux, FreeBSD, Solaris and BETA Drivers wich are links so they are blue.. :lol: Ill try using those drivers.. Thanks for pointing out my stupidity :roll:

bshor
Posts: 19
Joined: Sun Aug 21, 2005 8:53 pm

Post by bshor » Sat Feb 11, 2006 12:37 pm

Here are the results of my software RAID5 (free, built into Windows XP and exposed with the hack from Tom's Hardware). I am using 3 Samsung 2504C (250Gb each). Not sure why the image was corrupted, but here it is. CPU usage is high at 55% but not sure if this really impacts me with a AMD 64 X2 3800+. And the price is perfect!

Image

AZBrandon
Friend of SPCR
Posts: 867
Joined: Sun Mar 21, 2004 5:47 pm
Location: Phoenix, AZ

Post by AZBrandon » Sun Feb 12, 2006 4:00 pm

Here's my RAID 0+1 array of Seagate 7200.9 160gb (ST3160812AS) single-platter SATA2 drives, using NVIDIA's SATA RAID controller built into the NF4.

Image

I ran the test with the options set with the slider turned all the way to maximum accuracy and it's still all over the place for the first 50% or so and not really smooth until 70%. Does anyone know what causes that? Also the average of 84.3mb/sec is a lot lower than SiSandra which reports an average of 110mb/sec. I'm thinking the benchmark software is getting really confused somehow.

In other news, I thought even my 5.8% CPU load was high compared to thinking it should only be 2-3% or so, but then I'm seeing several of you posting with anywhere from 20 to 55% CPU load! Holy cow!

midiman
Posts: 33
Joined: Fri Jun 25, 2004 3:39 am

Re: 4 x Samsung SP2504C HDTune RAID0/1/5/10 Performance resu

Post by midiman » Wed May 02, 2012 1:54 pm

Replacing a set of four of these in the coming weeks and was always curious how they might perform in a RAID setup. Looks like they're not worth holding onto!

Post Reply