150GB Raptor or 2 150GB raptors Raided

Here are 4x150 Raptors
http://www.gamepc.com/labs/print_content.asp?id=raptor1...

Well more drives = *marginally* slower seek time, MUCH faster transfer rate.

Also of course, more space  


The only issue is that alot of RAID controllers found on home PCs and onboard motherboards are software based and offload most of the work to the CPU, meaning you pay a slight CPU penalty for the increased access speed....

XFX did some nice fully hardware-based RAID 0/1/3 controllers recently, might want to look at them. Because they do all the work themselves they are even able to present themselves to Windows as a standard IDE controller, eliminating the need for f6 driver installs while installing windows etc.

I dont remember if they were PCI-E or PCI, if they were PCI then 2 150GB Raptors is likely to saturate the PCI bus...

EDIT: They were the XFX Revo64 series, here: http://www.xfxforce.com/web/product/listConfigurationDe...

They come in 3 Port and 5 Port versions, all using the Netcell processor and with 64mb of cache. They are all PCI 32bit/66MHz compatible, although they will run @ 33MHz you'll hit the PCI bottleneck easily. If you have a mobo that supports 66MHz PCI you'll be ok (thats 266MB/s, enough for 2 drives easily)

Thanks for that link! Excellent.

Why buy 1 when you can have 2 at twice the price..? Makes sense to me.....

When I bought my mobo (Asus A8N32SLI) the sales guy told me that the RIAD processing was done by the mobo NVIDIA chipset. Is this incorrect?

Yes, he's right. However the A8N32-SLI also has a 2nd on board RAID controller manufactured by Silicon Image.

However, the benefits outweigh the negatives. I've seen benchmarks that show the thoroughput is much higher on the "software" based NVRAID solution (this is due to the fact that the controller has direct access to the hypertransport. the silicon image controller has to go thru the PCI-E bus to get to to HT).

IMO, the Silicon Image controller is good for RAID 1 and 5 only. RAID 0 is most beneficial using NVRAID.

When I bought my mobo (Asus A8N32SLI) the sales guy told me that the RIAD processing was done by the mobo NVIDIA chipset. Is this incorrect?

I know the nVraid found on the nForce 4 chipsets has a reputation for being faster and having a slightly lower CPU loading than most onboard RAID controllers (for example the Silicon Image sencondary RAID controller found on your board) due to it being integrated into the chipset, but I'm pretty sure its still at least partially software based.

nVRaid supports RAID5, the XOR calculations for which can be pretty hefty, and companies like Adaptec charge hundreds for products like this that normally need onboard cache etc too (which the nForce 4 does not, but it is quite close to the system RAM so could be using a small amount of that I suppose)

I'm not 100% sure on the specifics of the nForce4 RAID solution, I'm mostly going on speculation and reviews I have read on it, so dont take this post as gospel  

I just noticed something in that review... they're using a $550 Areca PCI-E RAID controller in the benchmarks! I thought the benchmarks looked a little bit off. Areca makes the best RAID controllers... the one they used is like top of the line. 128MB of on board RAM.

No wonder the thoroughput in RAID 0 was so high...

While I know that the nVRAID controller does 0,1, and 5, I find it hard to believe that it does full RAID 5 support like a dedicated XOR card would. Anyone got benches comparing the nV RAID 5 to other RAID 5 options?

Yes, I knew of the true Software based RAID systems with the OS (my JBOD storage array runs on this so I can add additional drives at will and transport it to a new computer independant of controller), but I didnt remember the 'host-RAID' term; if you look above I invented 'Hardware Acellerated' RAID for it instead heh...

The nVraid system is not true Hardware RAID though as you have said, and even with a Dual Core cpu I still want it doing only CPU-related things and not RAID5 XOR calculations  

In a perfect world I'd have a vagina and penis. Tits too...

Geez, from your pic, I always assumed you had a dick. Sorry.  

So are you saying the results of that test was somewhat inflated vs what most of us would see?

I also have (on the way) an asus AN32 SLI with build in raid support. I also have (on the way) the same 150G raptor. Would getting a second raptor for raid zero actually hender my performance?

I have a 3 drive RAID 0 with 3 Hitachi 80 GB SATA II's. In HD Tach Here's what I get:
Random Access Time: 12.4 ms
CPU utilization: 10%
Average read: 135.8 MB/s
Burst Speed: 336.1 MB/s
Keep in mind, I had a few things running when I did this test but it didn't score a lot higher/lower running the bares. The CPU utilization was a little lower though. This is on a Nforce 4 MB.

here's good article on raid 0 with graphs

http://overclockers.com/articles1297/index03.asp

it helps your gaming load times for sure.......id'e sure like to get into a map faster.

Yes, those numbers are somewhat inflated. They're not fake or anything, but don't expect those kind of numbers using a motherboard RAID solution.

-mpjesse

How the hell can you have a 3 drive RAID 0 setup? Do you mean RAID 5?

I can send ya a screenshot if you like. No hardware controller here. And yes, you can have 3 drives in RAID 0. I've been running RAID 0 since the Celery 300A days, but with a controller card.
Forgot the link: http://www.pcguide.com/ref/hdd/perf/raid/levels/singleL...

Yup, RAID 0 = 2 or more drives, odd or even numbers doesnt matter.

RAID 1 = 2 drives

RAID 0/1 = 2+ drives, always even numbers of drives however.

RAID 3 and RAID 5 = 3+ drives, odd or even numbers

RAID 6 = 4+ drives, odd or even numbers.

8O

True, except that alot of systems would be limited by the PCI/PCI-E bus before they hit the level of performance there (unless 4x PCI-E or PCI -X)

The 4 Raptors have a sustained Xfer Rate of 303MiB/s, more even than 66MHz 32bit PCI at 266MiB/s or 1x PCI-E at 238MiB/s.

Even most onboard controllers hang off one ot other of these, although to be honest I'm not sure about the nVraid on the nForce4

With 2 drives however, the difference between the Areca and normal controllers wouldnt be much

http://www.barefeats.com/hard51.html
They compare SATA, firewire with 1, 2 and 4 drive configurations with RAID.

You need to remember that, assuming enough read-ahead, while the first drive is finishing the seek, the 2nd drive is already 50% finished its seek, and so on and so on.

RAID-0 seek times are not as slow as people claim so long as the stripe-size is set correctly for the task at hand.
EDIT: For RAID-5 the write seek time, assuming no caches on the HDDs themselves or RAID controller which is wrong anyway, would be affected.  

For gaming I'd suggest 32 KB, 64 KB or 128 KB stripe.

The RAID-0 will be heaps faster HDD wise.

However, most game loads times are CPU bottlenecked. To prove this overclock +10% and time how much faster BF2 loads - If it is a linear 10% improvement in load time then it is CPU bottlenecked.

Most game resources are compressed, and need to be decompressed to memory when loaded, thus HDD performance is less important that ppl claim and game load times will usually scale with CPU performance.

Not that I can talk  


However the array is 'almost idle' while loading BF2, indicating the bottleneck is still the processor. (Game load isn't running isolated threads).

Going from 1 GB to 2 GB also improves BF2 load times substantially as it doesn't try to page to disk while also trying to read more resouces into to load into memory. Memory that isn't available and thus gets paged to disk. (The cycle continued until finally loaded). But 2 GB fixes that, and 3 GB ensures the whole game + OS, etc + all game related resource files sit in the OS disk cache eventually.

I assume we are talking BF2 load times, because frankly they suck.

Exactly. You need to configure your RAID for what you are going to be using it for. For some, 2 HD's are best, others,3, 4 or more. On this PC, data isn't that important as I backup often but accessing that data is. I was even surprised by the speed of a few of the tests. This also confirms the benchmarks I got from my setup, although my read speeds were a bit higher.

Get 3 HDDs and set them up as RAID-0 striping.
Uncommon yes, but it works just like 2 or 4 HDDs in RAID-0, as in if any fail the RAID-0 array will break.



If you are running a dual-core the 'nvraidservice.exe' process will just get scheduled to run on the more idle core. When/If it is doing XOR work a 2 GHz CPU is core going to do a pretty good job compared to a 300-400 MHz XOR processor. Bearing in mind likely under 20% of your disk I/O is going to be writes. Reads will not suffer this 'penalty' in CPU time.

The hardware solution is more 'portable' from machine to machine though, in that you don't need to break the array to port it to a new platform. They also have other similar benefits (like some have 128 MB cache on the RAID controller cards, but with 4 x 16 MB caches on the HDDs themselves, each with command queuing the difference is not as large as it once was, back in the SCSI glory days before IDE/SATA added large caches and NCQ.).

you would be better off using the raptor 150 for your OS and games and get a fast 250gb for storage. The real world difference in raid 0 is not really worth the added cost and added unreliability. TH did an article on this awhile ago. It showed the load time on a game between the 74gig and 2x74gig raid 0 the difference was 1 second. They did several other benchmarks as well with roughly the same results.

That's exactly what I have done. I got a 150 raptor for microsoft stuff and games, and a 250 WD for storage.