[Kramy]

Hello all,

I took a few HDTune Pro benchmarks of my 640GB Black, on my Asus M4A87TD Evo. (SB850)

I've got all sorts of short-stroke benchmarks. Thought it might be interesting to see how well different short-strokes and partition sizes scale. I've posted all of them, so hopefully that saves someone else some work.

Keep in mind that on a decent controller (ICH10R!) the access times would be even lower. I've seen 640GB blacks score 11.8ms on one of those.

Other notes:

• My OS partition (XP) is 16GB.
• More interesting benches: Caviar Black 640GB Benches (on different SATA controllers)

And now for IOPS and access times:

Past a certain point, random chance seems to play a bigger effect.

I like how a 10% short-stroke increases random read performance by ~50%.

 

[AyeYo]

Inb4shortstrokehaters

 

[beers]

In all honesty it would have saved you a lot of time to just look at the bench for the entire drive and section it off based on the capacity/performance you want to run.

Based on the initial graph I'd probably short stroke it to 128 GB, as there isn't much falloff yet, and you can comfortably live with an OS on that kind of size.

 

[TheDreadedGMan]

Awesome detailed benchmarks... I could almost make a graph from the numbers, and find the "Sweet" spot...

although I guess it's already a kinda graph... but I want more lines

 

[AyeYo]

Quote:

Originally Posted by beers In all honesty it would have saved you a lot of time to just look at the bench for the entire drive and section it off based on the capacity/performance you want to run. Based on the initial graph I'd probably short stroke it to 128 GB, as there isn't much falloff yet, and you can comfortably live with an OS on that kind of size.

That'd leave him with slower random access times though. It was nice to see each section benched separately so that the average seek time difference could be seen.

 

[Kramy]

Quote:

Originally Posted by beers In all honesty it would have saved you a lot of time to just look at the bench for the entire drive and section it off based on the capacity/performance you want to run.

True. But I also wanted to see the IOPS. Those RandomAccess benchmarks show that partitioning only 10% of a drive gives a pretty huge speed increase. (50%!)

I've already decided that ~40% of the drive is a good size. Eventually, if I get an SSD, I'll use this HDD exclusively for games.

Quote:

Originally Posted by beers Based on the initial graph I'd probably short stroke it to 128 GB, as there isn't much falloff yet, and you can comfortably live with an OS on that kind of size.

Right now 16GB is working fine.

Quote:

Originally Posted by TheDreadedGMan Awesome detailed benchmarks... I could almost make a graph from the numbers, and find the "Sweet" spot... although I guess it's already a kinda graph... but I want more lines

I was considering making a chart in OpenOffice... but these had already been sitting around for a few days. It was late at night, so I just posted them.

Make a chart if you want to.

Quote:

Originally Posted by AyeYo That'd leave him with slower random access times though. It was nice to see each section benched separately so that the average seek time difference could be seen.

What amazes me is how a small access time drop improves speeds so dramatically.

Entire drive:
http://img69.imageshack.us/img69/893...100922ahci.png
http://img688.imageshack.us/img688/7...22ahcirand.png

10% of drive:
http://img80.imageshack.us/img80/625...17ahciss64.png
http://img822.imageshack.us/img822/7...22ahcirand.png

50% faster random read! 40% faster random 1MB read! That 12.4ms vs 8ms makes a huge difference in how much time the drive spends reading rather than seeking.

 

[TheDreadedGMan]

Well I fired up Excel... and after doing some charts I realized that you're missing a 128GB point, you jump straight from 40% to 10%...

EDIT: Chart updated, 128GB data point added

In any case here are the charts, and I've attached the actual Excel sheet for everyone.

I did the 640 to 16 GB range separately since it's the biggest curve, then also a full curve just for completeness

 

[The Master Chief]

I still dont think short strokes are worth it... In all honesty, I've done it before, but couldnt feel the difference.

 

[TheDreadedGMan]

Quote:

Originally Posted by The Master Chief I still dont think short strokes are worth it... In all honesty, I've done it before, but couldnt feel the difference.

Well in Kramy's extremely short 16GB stroke, he'll be getting just under twice as many I/Os per second...

It's not like on SSD where you go from 90 IOPS to 8000 or 50000 IOPS tho.

 

[The Master Chief]

Quote:

Originally Posted by TheDreadedGMan Well in Kramy's extremely short 16GB stroke, he'll be getting just under twice as many I/Os per second... It's not like on SSD where you go from 90 IOPS to 8000 or 50000 IOPS tho.

I went

160

100

64

32

all it did was nothing for me lol. So I said, why am I doing this?

but if you like what it does, more power to you.

 

[chatch15117]

I'd like to see benchies between the 00L3B2 and 00J7B1(two and 3 platters, respectively).

I get 10.2ms delay with 3 in RAID 0. No short stroke.

 

[Kramy]

Quote:

Originally Posted by TheDreadedGMan Well I fired up Excel... and after doing some charts I realized that you're missing a 128GB point, you jump straight from 40% to 10%...

Yeah, I did. I was short on time, and more interested in smaller amounts.

I may get around to it sometime this weekend. I'll post again if I add them.

Quote:

Originally Posted by TheDreadedGMan It's not like on SSD where you go from 90 IOPS to 8000 or 50000 IOPS tho.

Hehe... so true!

Quote:

Originally Posted by chatch15117 I'd like to see benchies between the 00L3B2 and 00J7B1(two and 3 platters, respectively). I get 10.2ms delay with 3 in RAID 0. No short stroke.

That's very low!

I wonder what sort of access times you'd get with a short-stroke?

 

[AyeYo]

Quote:

Originally Posted by TheDreadedGMan Well I fired up Excel... and after doing some charts I realized that you're missing a 128GB point, you jump straight from 40% to 10%... In any case here are the charts, and I've attached the actual Excel sheet for everyone. I did the 640 to 16 GB range separately since it's the biggest curve, then also a full curve just for completeness

That's a really great way to visualize it. That's a tremendous gain. I wonder if you could pick up a 1TB and give it the same 2.5% short stroke, you'd be up to 25gb which is much more reasonable, and should be at the same performance.

 

[Kramy]

Added 128GB short-stroke benchmarks.

 

[F0110111111]

Quote:

Originally Posted by Kramy That's very low! I wonder what sort of access times you'd get with a short-stroke?

NO ITS NOT LOL.

My WD AAKS 1TB black drives (RAID 0) did 7.2ms with no short stroke , short stroke did like 6.5ms

 

[Kramy]

Prove it.

Most of the HDTune benches I've seen are nowhere near that - even for RAID arrays.

 

[shadman]

Short stroke to ~100GB with my WD Black 1TB got me 7.6ms access times

Proof:

 

[TheDreadedGMan]

Quote:

Originally Posted by AyeYo That's a really great way to visualize it. That's a tremendous gain. I wonder if you could pick up a 1TB and give it the same 2.5% short stroke, you'd be up to 25gb which is much more reasonable, and should be at the same performance.

Sounds like someone needs to do a few benchmarks