Fastest Hard Drive Setup for <$200

[xollox]

I recently built a new system and have generally been happy with performance (see the link in my sig for details.) The one thing that has been bothering me is those slow read/writes from the Hard Drive (Seagate 7200.10 320gb 7200RPM perpendicular 16mb.) To remedy this I'm thinking about building a RAID0 array out of up to 4 SATA ports on my mobo. This will be used for the OS install and games install, but no "mission-critical" data. Redundancy is meaningless to me for this. Size also isn't really an issue.

Originally I was looking to RAID0 2 Raptors, but I'm thinking that I can get better performance for a better price. 2x Raptor 36.7gb drives would be $210 on newegg.

~$180 will get me 4x 80GB 7200RPM drives on newegg. I'm thinking that a 4x drive RAID0 will be significantly faster than a 2x Raptor raid.

I'll be using the onboard raid on the nforce570 for the raiding.

Any thoughts? Any one have experience with these two setups?

TIA

 

[Boyo]

I think the Raptors will still be faster.

 

[Roguestar]

I think the raptors would be faster too.

 

[xollox]

According to This Tom's article, the average read performance on the Raptor 74gb is 75.3MB/s. The average read performance on the Cavier SE 160gb is 64MB/s. In a 2x Raptor RAID0, that would be 150.6MB/s minus overhead. On a 4x Cavier that would be 256MB/s minus overhead. According to that math, the 4x cavier would be 58% better than the 2x Raptor.

Am I missing something?

 

[Matthias99]

Originally posted by: xollox
According to This Tom's article, the average read performance on the Raptor 74gb is 75.3MB/s. The average read performance on the Cavier SE 160gb is 64MB/s. In a 2x Raptor RAID0, that would be 150.6MB/s minus overhead. On a 4x Cavier that would be 256MB/s minus overhead. According to that math, the 4x cavier would be 58% better than the 2x Raptor.

Am I missing something?

Only that STR generally means very, very little for normal desktop work.

The Raptor has a lower seek time, and that will probably impact you more. Unless you spend a lot of time moving big files around (video editing, for example.)

 

[xollox]

My main complaint is that windows takes too long to start and games take too long to load.

I'm guessing that the slowness here is reading from the hard drive and putting stuff into memory. I would think that the average read speed would be the important number here. Am I wrong?

 

[gramboh]

Seek time is important for fast booting/game loading, if you look at Anand's recent review of the 74GB 16mb cache Raptor it is the fastest in load times.

 

[xollox]

I just checked it out. I'm not sure if that's totally relevant because they are testing single drives and not RAIDed drives.

 

[trOver]

jeesh... this makes me want to upgrade my 250gb 7200rpm sata drive...

 

[Matthias99]

Originally posted by: xollox
I just checked it out. I'm not sure if that's totally relevant because they are testing single drives and not RAIDed drives.

...because increasing the STR won't help. Games load hundreds of little files (or read around in big archives and load hundreds or thousands of small art assets). The limiting factor is seek time.

RAID0 will make Windows boot faster, since they do a lot of sequential reads in parallel to get all the OS data loaded. But an easier solution is to just not reboot your machine as much, and/or have fewer things loading at startup.

 

[xollox]

OK, that makes sense, thanks for the technical explanation.

As a note, it's a pretty clean install of Vista, only a few days old. There isn't really tons of stuff loading up.

To confirm:
From what I've seen, the more drives you add to a RAID, the longer your seek time gets. That means the 4x drive raid could be hurting game-loading performance (because the Sustained Transfer Rate isn't as relevant as seek time.)

The best option sounds like it would be the 2x Raptor because the seek times would still be below average (to make games load faster) and the STR would be well above average (to make windows load faster.)

Unfortunately that doesn't fit well into my <$200 budget so I may have to hold off for a little bit until I get really annoyed at waiting for stuff.

EDIT: Articles for reference:
http://www.anandtech.com/showdoc.aspx?i=2101&p=5
http://www.anandtech.com/storage/showdoc.aspx?i=2922

 

[f4phantom2500]

You could get 4 36GB raptor drives, I'm pretty confident you could grab 'em for $50/piece in the fs/ft forum. I think that would be the fastest, and have a similar capacity to 2x74GB raptor drives, but if you go that route you really can't care about noise, heat, power consumption, or capacity (obviously).

 

[Madwand1]

It's hard to come up with theoretical explanations that are correct for such matters. And even when you do actual benchmarks, there are so many different configurations and options that whatever you've come up with may not be accurate for another case.

Few RAID benchmarks get even the most basic variable right -- the stripe size. Most make no effort whatsoever to even try a different stripe sizes, and many misstate their behavior. When you have benchmarks that purport to represent RAID, but don't even understand or vary this parameter, what good are they?

E.g. Here's a thread showing some measurements which have the opposite conclusion -- RAID 0 helps game loading performance significantly.

http://www.ocforums.com/showthread.php?t=445800

But to tell you the truth, I'm not even going to look at that article in detail to think about where they got things right and where they might have gotten them wrong. I don't believe that theory-based general rules apply here. Do a dozen benchmarks, and you can get a dozen different results. Look at storage review's hard drive benchmarks for example -- they'll show vast difference between different hard drives on the same tests, with one trouncing another, and on the next benchmark, the opposite. How exactly should we deal with this amount of variability in general theoretical guidelines for every application? I won't even try.

If you've done configurations and measurements for actual applications that matter to you using hardware that's relevant to you -- your results are valid. Beyond that, there's a good chance that the generalization isn't worth squat.

 

[Matthias99]

It's hard to come up with theoretical explanations that are correct for such matters. And even when you do actual benchmarks, there are so many different configurations and options that whatever you've come up with may not be accurate for another case.

I do agree with this. I may have been a little one-sided earlier -- improved STR will help, just (in general) not anywhere near a linear improvement for desktop applications. But there may be exceptions.

If your application optimizes the layout of its files on disk so that much of the time it can do big sequential reads -- increasing STR will decrease load times significantly. Most games and desktop applications don't work like this. If your program has to access 1000 little files or pieces of data on disk (like many games when they are loading art assets), high STR is going to be of limited benefit. If it reads huge blocks of data (1MB+) sequentially all the time, you'll see more benefit. Programs that read/write enormous amounts of linear data (like video editing, or photo editing on really large multilayer images) will see significant improvement in disk-bound performance.

 

[xollox]

Been doing a bit more reading.

Another important thing to take into consideration is the controller. It seems like controllers based on the PCI interface (either on board or in a card) are not seeing good results. The more modern chipsets with built in RAID seem to do a whole lot better. The ocforums article you linked was using the built-in RAID in the nforce4 chipset. The other anti-raid0 sticked post on ocforums and the anandtech article were both written in 2004. I wonder if the raid controllers just weren't up to snuff then.

That said, it sounds like the 2x raptor solution is still going to be better due to significantly lower seek times.

Also, 2x74gb seems better than 2x36gb because NCQ and the lack of a SATA->PATA converter in the 74gb version.

 

[f4phantom2500]

Originally posted by: xollox
That said, it sounds like the 2x raptor solution is still going to be better due to significantly lower seek times.

Also, 2x74gb seems better than 2x36gb because NCQ and the lack of a SATA->PATA converter in the 74gb version.

Yes, but I'm pretty sure they came out with an updated version of the 36GB one that doesn't have the converter. Also, I believe that 4x36GB drives would beat 2x74GB drives.

EDIT: Yeah, you're thinking about the original 36GB model (WD360GD), but they have since come out with the WD360ADFD, which, judging by the letter scheme, has the same features as the 150GB and updated 74GB drive. Also according to wikipedia it has NCQ and is native SATA, so the only differences are those inherent upon the drives due to the differences in capacities (number of platters, heat, power consumption, etc)