hard drives & raid - benchmark and compare!

[Angry_Games]

but I still saw the x850 XT PE as a very tough cookie to crack, since there was a review that had it beating dual 6800GT's in HL2

 

HL2 was really optimized (no matter if Valve or ATI will admit it) for ATI architecture (hence all the free HL2 coupons that came with ATI cards). Drivers play a big big role in whether a game runs so-so or #$^$#@ kick arse!

 

76.41 drivers is what you want to try, and load the profile for the game in NV Control Panel for the vid card before you play it.

 

71.84 = the most awful drivers I have ever tried, especially for SLI.

 

AMD has the superior cpu design AGAIN. I wonder what the heck Intel is smoking these last two years? Ah well.

 

 

the FX-57 will exist, dont worry. My FX-55 must be a new E0 stepping as it is only 1.4v (tho still 130nm) and has SSE3...but I wont get to really fool with it until she goes to work for the summer on Monday (she got the FX-55 + X850XT PE).

 

btw, there's no reason why SLI 6800GT's shouldnt freakin annihilate a single X850XT with good drivers and the right profile for SLI.

 

i dont sell cars, and I'm really not TOO biased...I just get to test a lot, I read a lot, get to talk to a lot of mfg's (and filter out their bull crap marketing) and then make my own decisions.

 

the fact that I've worked on a couple of mod teams with the two main game engines (Quake3/Unreal) helps a lot too as I know a lot about how game code works on certain hardware

 

I dont know it all though, far far from it. I just am passionate about it!

[Vista]

I finally got a copy of ATTO so I can submit some results on my Raptors.

 

ATTO bench of 3GB C: pagefile/boot partition, 16k stripe/4k NTFS clusters:

[bIMG]http://vistadfi.home.comcast.net/bench/attoc200505211402.png[/bIMG]

 

ATTO bench of 100GB E: data partition, 16k stripe/32k NTFS clusters:

[bIMG]http://vistadfi.home.comcast.net/bench/attoe200505211407.png[/bIMG]

 

HD Tach 32MB bench:

[bIMG]http://vistadfi.home.comcast.net/bench/hdtach200505211501.png[/bIMG]

 

It is interesting how close the ATTO read/write speeds are for block sizes over 32k. I expected writes to always be slower than reads. I also expected the 16k stripe and 32k NTFS clusters on the 100GB partition would hurt performance more than it does (about 8% slower).

 

Overall I'm very happy with the performance and noise level... we will see about reliability. I usually get one HDD failure every 2-3 years of 24/7 use, usually because of a cooling problem that cooked the drive. This Lian-Li case has excellent HDD bay cooling, so I'm hoping my array will last at least 2 years.

[Angry_Games]

got some X2 4400+ benches coming...but in WindowsXP 64-bit (RC1433, 6.39 NForce beta, 71.84 Nvidia 6800GT x64)

here...

 

 

 

oops forgot the long version...384.1MB/s!

 

[Sorrento]

Vista, you got better results with the 16/4k setting than with the 16/32k one. Have you tried 32k stripe size?

 

Angry, arghh... 64 bit... hehe. Question, do I need to load every nvidia profile before starting a game??? Isn't that automatic since it has which exe corresponds to each profile?

 

I am trying 76.41, maybe tomorrow cause its quite late now. With 71.89, FSAA 4x and Aniso 8x (at config file), highest quality on everything (all maxed), Doom3 timedemo scored 71+ fps... and cards running stock 350 GPU and also stock memory speed. But OGL is not what I am looking to beat, it is D3D.

[Angry_Games]

Question, do I need to load every nvidia profile before starting a game??? Isn't that automatic since it has which exe corresponds to each profile?

 

i dont trust auto anything too much

[Sorrento]

LOL, well said

[ncsa]

Here are some results for 4 x 200Gb DM 10 HDD RAID 0 run on 3500+ @ 285x11

stripe 16K / cluster 64K

I am not sure why the CPU utilization is so high 24%, seems a bit excessive?

[Vista]

ncsa, you are getting some sweet RAID 0x4 results! If you aren't sure, just ship those drives to me and I'll test them for you LOL Really, I'd just back up that array often and call it a good day.

[ncsa]

Yeah no worries... and they appear to be working quickly - loading of apps / XP etc but my CPU is about twice what others are reporting...?

 

Maybe some other setting needs to be made or a HDD firmware upgrade.. have been searching but no easy access to these...any thoughts?

[soundx98]

Very, very impressive. Big thumbs up.

[red930]

Make that, 16K cluster with a 64K stripe. The cluster can never be larger than the stripe.

 

Your CPU Utilization is higher due to the tracking across four drives.

 

Here are some results for 4 x 200Gb DM 10 HDD RAID 0 run on 3500+ @ 285x11

stripe 16K / cluster 64K

I am not sure why the CPU utilization is so high 24%, seems a bit excessive?

[ncsa]

Make that, 16K cluster with a 64K stripe. The cluster can never be larger than the stripe.

 

Your CPU Utilization is higher due to the tracking across four drives.

 

It appears that I have confused my ratios C:S... and re-reading back this thread I am still missing an element about this 4:1 ratio, the logic behind it and it maybe due to how the stripes are established across the Array.

 

Here are my assumptions and I would be greatful if you could help clarify my confusion :confused:

 

Stripe: I chose 16K as being a balance across the whole array and the partitions that would be created within it, given that you can only have one stripe size for any physical array.

 

Stripe Width: In this case is 4, given that there are 4 physical drives which provides a logical stripe size of 64K.

 

Cluster: 64K being the smallest file block chosen. This allows the RAID controller to slice the file into 4 x 16K blocks across all drives in the array.

 

e.g. A 242Kb file would be allocated 4x64K clusters to be written in 8x16K stripes on the HDD for the given setup.

 

If the ratio was the other way around (as is suggested) 16K clusters : 64K stripes then the same file would be allocated 8x16K clusters and written in 4x64K stripes

 

Does this then increase performance due to the fewer reads / writes required by the HDD, reduce head thrashing and help drop the CPU utilization?

 

Why is it that the cluster can never be larger than the stripe?

 

Thanks again

 

EDIT: I think I have answered my own question and have been reading more from other sources.. will come back and post the revised performance charts.