Sign in with
Sign up | Sign in
Your question
Solved

SSD Experts wanted: Low benchmark scores for OCZ Vertex 2 RAID0 setup

Last response: in Storage
Share
a c 99 G Storage
March 18, 2011 4:19:10 PM

SSD Experts wanted: Low benchmark scores for OCZ Vertex 2 RAID0 setup.

Let me apologize, as there are a few things I should’ve tried before posting a thread, but I’m not on the computer in question at this time.

I replaced 2 Intel X25-V 40GB SSDs in RAID 0 with 2 OCZ Vertex 2 60GB SSDs in RAID 0. I upgraded the firmware to 1.29 before installing them as a boot drive.

The problem is that I have been noticing over the last two weeks of usage, the system is lagging (slowing at boot up). Now I have problem with Internet Explorer (hanging/crashing). I may reinstall the OS, and then upgrade to IE9, and monitor for any further issues.

I ran the AS SSD benchmark program, and the numbers are way off from what I think they should be. Sorry I don’t have a screen shot. They are not much better than my Intel SSD drives were. I should be getting reads/write close to 570/540MBps (theoretically). But, they are only getting 384/60, the 4K and 4K-64Thrd numbers and access times are lower than the Intel drives were. This is why Intel drives are so good, right? 4K reads/writes? **See Follow UP**

When I installed the OS (Windows 7 Home), I pointed it to use the new Intel RST v10.1 drivers for the chipset (from the downloadable Floppy files to use during a <F6> install). However, I have not upgraded the chipset (X58/ICHR10) drivers otherwise, by running the Intel RST program, which is what I think I really need to do, which updates the chipset driver, and gives me a GUI to monitor or change the drive settings.

But, when I installed the OS, I did not partition only 80-85% of the array, as I did for the Intel SSD drives. (i.e. only used 64GB of my 80GB size of the Intel drives for the OS/Programs, leaving the other 16GB unallocated and unused for drive self optimization.) I thought that the new SandForce 1222 controller could handle the drive optimization itself, and the drive is nowhere near full. I also know that TRIM is enabled, but not used for SSD drives in a RAID array.

I just wanted to know if anyone has any other tips or comments, before I run the Intel RST v10.1 program tonight, and monitor any changes.
I’ll give it a week, and if these drives just don’t meet my expectations, I’ll return them, and put my Intel drives back in.

**Follow Up** I just found this webpage: AS SSD Benchmark stating the following:

SANDFORCE Disclaimer::
This benchmark uses non-compressible completely RANDOM data, so Sandforce drives won't produce "super-fast" numbers like they do in ATTO.
Also writing a whole bunch of non-compressible data to the drive quickly has reportedly pushed a few Sandforce drives into "write throttled" mode, basically Sandforce users have been warned. ;) 


I’ll try to ATTO benchmark, as well as HDTune. But I know too many banchmarks can harm the drives.

Regardless, my system is not a fast/responsive as it was with the Intel SSD drives…

P.S. I had my CPU overclocked to 3.2GHz, and memory timing @ 1604 (8-8-8-24) at install, but I has put the CPU back to stock timing (2.67GHz), but left the memory timing @ 1600 (8-8-8-24) because that is what it should be, else it defaults to lower settings of 1600 (9-9-9-24), or 1333 (7-7-7-24).
a c 99 G Storage
March 19, 2011 5:55:55 PM

Well, alot of readers, no replies!

It appears as if the "disclaimer" above IS very correct. When I ran ATTO & HDTune benchmarks, I got figures closer to what I expected!

I ran a series of benches before I installed the Intel RST v10.1 driver program, to "update" my chipset drivers, AND a series after. I saw not much difference between to two (less than 1%). Had some number actually lower after the update, but not by much.

AS SSD Benchmark - SSD Drives in RAID 0 before iRSTv10.1 install:



ATTO Disk Benchmark - SSD Drives in RAID 0 before iRSTv10.1 install:



HD Tune Benchmark - SSD Drives in RAID 0 before iRSTv10.1 install (Read ONLY):



HD TUne Benchmark - SSD Drives in RAID 0 AFTER iRSTv10.1 install (Read Only):




I still don't know if the system is faster? It is still lagging, especially in Internet Explorer. I'm gonna install IE9 this afternoon. And then run some tweaks for SSDs which I didn't do for the Intel X25-V drives.

Thanks for reading. I'll keep ya'll updated.


m
0
l
a c 353 G Storage
March 20, 2011 3:24:48 AM

Think you still have a problem - just not sure what it is.

Your Raid0 beats my single Pheonix Pro in only one cat ( Seq read). Both the vertex and the Pheonix Pro use the SF1200 controller and have simular specs.

Raido, as you know, only increasest the seq read/writes and does very little, if any improvement, in small file random read writes. it it offten better to get one larg SSD as opposed to two smaller ones placed in raid0. PS Before SSDs, ALL my rigs have been Raid0 since around 2000.

My specs (Have them in Word and to lazy to do screen shot)
From AS SSD (Yes about non-compressible data)

Seq ............. 201.74 ......133.74
4k Random .... 22.79 .........80.33
4k-64............125.24.........119.26
Acc Time ........0.115 ..........0.206

Score .............168 ............ 213
................................473
m
0
l
Related resources
a b G Storage
March 20, 2011 7:18:43 AM

I would go to the OCZ Sandforce forums and check out the 25nm nand thread. You may have drives that are mismatched(ie: 25nm and 34nm and/or 64Gbit and 32Gbit). You may be eligible for free replacements.

60MB/s on the sequential writes is absolutely horrid. If I had to guess, I would say that write caching is not active on that array. If you're using the RST app set it there(is called write-back cache). If you don't have the RST app you can also activate it in device mgr/drive properties as well. Many don't know that ALL Intel raids are inactive by default DESPITE being shown as enabled in device mgr and you must untick the caching and click "ok" followed by a retick/"ok". Then rerun the Atto for the caching to ram to be seen in all its glory.

and an FYI.. the only thing that doesn't increase with raids using Sandforce is the 4k random reads.

Sorry again Chief.. not tryin' to follow you around and step on your knowledgable toes. Is just because I beta test for OCZ and have a bit of time built up with the Sandforce controller so far. The Vertex 3 is a heck of a drive and should stir up some healthy competition.
m
0
l
a c 353 G Storage
March 21, 2011 1:10:38 AM

The increase in access time is very little (same is raid0 on HDDs). And it is the random 4 k reads that primarily results in the hugh boost in performance. Close to half of the files on the boot/program drive are under 16 K and a pretty fair share are 4 K (= 1 cluster). An increase of 10 % to 15 %, in benchmarks, may or may not be noticeable in daily performance.

From what I've read, In my Opinion I see little benefit in putting SSDs in raid0, (unlike HDDs) and losing trim - may change my mind later!

The Seq's work wonders on large files that are sequentials. These are typically large databases, Cad/Cam type files, Jpeg/bitmap photos, Video files. With a 120 gig (or smaller SSD) But most would tend to put these type files on a HDD.

As I said, I'm a big fan of Raid0 for HDDs. The big diff is Size/cost. With Mech HHDs for < 200 Bucks I can Raid0 two 1 TB drives, short stroke them, Which does improve access time and 4 K random file work (Ex cut access time on WD Blks from around 12 mSec to 9.5 mSec). This allows for setting up partition for working with these large files.

Will look for supporting links, as I've seen several.
Here is one based on Corsair Performance 3 Series 128 Gig (shows Raid0 abd single drive performance)
http://www.legitreviews.com/article/1567/1/
Note the outstanding Seq performance ( As I indicated), but also the 4K Random performaqnce, file times and program loads are all Lower in RAID0). This seem prevalant in reviews that look at the over all performance, note just Sequencial performance - and as mentioned currently loose trim suport.
m
0
l
a c 99 G Storage
March 21, 2011 4:51:06 PM

Well, I still have not installed IE9.

I was working with some Excel files, and (still) noticed a lag at times.

It occurred to me that maybe my memory timings are off. So I changed the settings for memory in BIOS back to “Auto”, and still set the frequency to DDR3-1600. I ran CPU-Z, and the memory is set to 1600MHz, with the timings @ 9-9-9-24 when my sticks should be DDR3-1600 @ 8-8-8-24.

All seems better!! So much for overclocking! But my settings were not outrageous: 3.2GHz CPU, DDR3 @ 1604 8-8-8-24. It’s probably the OCZ Gold memory sticks.

BTW, the main reason I RAID’d the SSDs was for size, as I had got the Intel X25-V which are only 40GB. I was impressed with the results, so I did the same for the new OCZ Vertex 2 60GB drives.

The “Sandforce Disclaimer” I posted was VERY correct! Numbers were much better when I ran ATTO and HDTune, vs AS SSD.

Of course I’ll keep a close eye on the system performance, and see if lack of TRIM is an issue down the road. But I think the SandForce controllers should keep up.
m
0
l
a c 353 G Storage
March 21, 2011 6:53:18 PM

I think that the "GC" has improved in the newer SSD, so you may be fine. Driving force for improving GC is for systems that do no have trim, ie XP, Vista and macs. Also for SSDs used in raid, but as you said time will tell.

If you find you want to ditch the Raid(0) - with two 60 gig SSDs you could put your OS plus some programs on one (C:)  and use the 2nd as D: and install remaining programs to it.

On memory, I'm using the GShkill ripjaws DDR3 1600 CL 7,8,7,24. Currently have 2 x 4 Plus 2 x 2 for 12 gigs. Just got a 2nd set of 2 x 4 and Am going to install them now. Did your origninal timings for your Ram pass Prime95 testing.
m
0
l
a c 99 G Storage
March 22, 2011 3:17:20 PM

I have not run Prime95, but I will.

Off topic, but these memory timings are screwy! If I let my BIOS set them Auto, they will be 1333 @ 7-7-7-24 (I think, they may even be 6-6-6-24), but If I OC them to 1600 (which is what they are rated at), they run @ 9-9-9-24 (when they should be 8-8-8-24). 1333 @ CL7 is faster than 1600 @ CL9, isn't it?

Right now, today, I'm thinking of returning the 2 OCZ Vertex 2 60GB drives, and getting an OCZ RevoDrive 120GB PCI-e, for $20 more. I'm reaseach/looking at threads now. Number are (theoretically) lower, but not by much (5%: 285x2=570Mbps for Vertex 2 in RAID 0, 540Mbps for RevoDrive), but the IOPS are higher. I'll check 4K reads/writes, and access times.
m
0
l
a c 143 G Storage
March 22, 2011 3:32:47 PM

Read through this site... http://www.ocztechnologyforum.com/forum/showthread.php?...

I think from the testing you are seeing the drive throttle, which I can say from first hand experience is accurate. I've been doing some testing on SSD and noticed this effects on my whole system (when testing from the OS). In discussions with OCZ, their recommendation is to wait a week between testing and to do very little testing in the first place. Each one of those test are writing roughly 18G of data in a very short time (how it was explained to me :D  ) This quickly eats up your OP and cause the drive to throttle. So I would expect you would see lag in your system afterwards or in the next day. With no testing, the system should be back at normal over time (week or two).
m
0
l
a c 353 G Storage
March 22, 2011 4:48:55 PM

On memory timings. In my system (P55-UM4P) what I did was to go into BIOS and specify use XMP Profile 1 and it set Timings and Voltages to correct values. Also have My I5-750 set to 3.2 GHz

One cavet - It also changed my Bus speed from 133 -> 166 and decreased the multiplier from 20x to 16X end result is no change in CPU, still at 2.67 Ghz. I then reset the Bus to 200 Mhz and put the multiplier to 16 for 3.2 GHz. Had to up the vcore just slightly for stable operation (@ work so can not say what vcore is)

Added: On my going from 12 Gigs Ram -> 16 Gigs of ram - went great, removed the 2 x 2 sticks and popped in the 2 x 4, booted and ran Prime95, Passed. End added

tecmo34 is probably correct. Since you loose Trim you are relying solely on internal CG.
m
0
l
a c 99 G Storage
March 22, 2011 6:34:33 PM

Thanks tecmo34. I will wait for a while....great reads on your link.

Thanks RetiredCheif. I'll try the XMP in BIOS for my memory settings.

** EDIT** XMP did not work, it defaulted to 1066 @ CL9. I put everything back to Auto, but set the DRAM timing to 1600, and leave it alone.
m
0
l

Best solution

a b G Storage
March 23, 2011 5:00:39 PM

I've been using a 6 drive array of OCZ Vertex 2 drives for about 1 year now. I also beta test the new Vertex 3 as well. Point being.. I have major experience in the use of these controllers and actually helped form the way in which maintenance/recovery is done with many 5000 word posts over on the OCZ forums. Although admittedly, much of which was useless blabbering while I tried to tell others what was going on with these controllers through my testing. lol

Anywho, here's the deal foscooter.

Sandforce controllers work like no other controller and have 2 distinct throttles built into the firmware. First one(which you seem to understand already) is the "settled state" where all nand has been written at least once and a data map is fully formed to allow Durawrite to be implemented from that point on. At this point the controller is throttled for life. Period, exclaimation point, end of story. There is absolutley no way to recover fresh speed from this state UNLESS the drives data map is wiped through secure erasure(SE) and is specifically the way Durawrite is intended to function. Latency is also affected as well, which you seem to have noticed already.

You can idle the drive forever and it will NEVER regain fresh speed. I've tried 5+ straight days of idle while logged off(vacation) and "settled state" is as high as it goes. This state is intended to be the normal operational speed of the drive and cannot be circumvented without an SE/reinstall. It's purposely written into the firmware and is called Durawrite. Again, SE is the ONLY way to wipe the Durawrite process through elimination of its required data map.

Second and final stage is what's called the "hammered state". This is Durawrites "keep you from killing it" algorithm implementation and is there until the data flow settles down and the drive is allowed to recover/lift the throttle with some low activity powered on time(usually logoff idle is best). The drive will NEVER drop below this state and can maintain this state indefinately if you keep pounding it with writes(especially incompressible which hits more nand due to no further compression allowance) as the controller has the strength to maintain it through great "on the fly" recovery. Latency is also at its worst at this point as you could imagine, although it should be mentioned that even in this state, the Sandforce controller can best many other drives on the market.

I've done well over 10TB worth of testing on Sandforce based drives with highly specific degradation/recovery testing as the main focus through most of it so trust me when I say,.. "I have a clue". Is also common knowledge for those who practically live on the OCZ forums that settled states are non-recoverable without secure erases. I don't prefer "settled states" regardless of how much extra time they may add to my drives lifespan through the addition of Durawrite and always secure erase/reimage when it hits.

I'm so in tune to this controller/array, that I can actually tell the moment it hits. Logoffs/shutdowns and logins/boots take many seconds longer to occur. I can SE/reimage all 6 drives back to fresh desktop in 13 mins flat now so why deal with over 100MB/s losses in incompressible write speed when it's avoidable, right? I don't need my drives to last 5 years and would rather have more speed on the frontend of the deal.

PS. If it were my system, I would SE/reimage and avoid benchmarks from here on out. If somethings up or the drive has throttled again.. you'll know. Keep stored data as minimal as possible(that;s what HDD is for) and never exceed 75% of available capacity. You can also allow extra unallocated space to give greater stamina and recovery(idle time required), also called "OP". I do it at the creation of the riad and this eliminates the OS from babysitting and adding layers of translation between physical and logical space. Not like your ever going to fill it up to max capacity all at once anyways, right?

Hope that helps clarify some things for you.
Share
a c 143 G Storage
March 23, 2011 5:07:17 PM

^+1...

Great Post... This is the feedback I received from "Tony OCZ" over Skype, which is why I linked his thread on the subject matter. :) 
m
0
l
a b G Storage
March 23, 2011 5:13:56 PM

I know.. I taught him most of what he knows about this controller. LOL

kidding(mostly) hehe
m
0
l
a c 99 G Storage
March 24, 2011 6:36:07 PM

Thank You groberts101.

I wish I would have read your post before.......I returned the drives this morning. I ran HDDErase (just like Secure Erase, but in DOS, a whole other issue) before I repacked them for return.

I understand the issues with SandForce controllers, but....my Intel X25-V 40GB in RAID0 were just more......stable.

I'm going a different route now. I was thinking about an OCZ RevoDrive 120GB x4 PCI-e, but I know that the Vertex 3, on SATA III (which I don't have) already come close in reads/writes without RAID. Maybe the next series of RevoDrives will push there numbers up, and get a better RAID controller than SiS (as Intels ICHR is better).

Instead I got another 1TB AND 2TB HDD for RAIDing my data/media AND backup drives. I was filling the 1TB and 2TB drives up. Adding drives was cheaper than a BD-R and the disks. I already have a DVD-R, but have several files (movies) over 4.71GB in size. May still get a BD-R later, with a new i5 2500K build, when the new Z68 chipset mobos come out!!Then what'll I do, Revo on PCI-e or SATA III?

Anyway, how does one become a beta tester for OCZ? I've seen your posts on OCZ Forums. I have/had OCZ SSD drives, OCZ memory sticks, and OCZ PSUs.
m
0
l
a c 99 G Storage
March 24, 2011 6:39:23 PM

Best answer selected by foscooter.
m
0
l
!