I'm a bad person (UDMA33 vs. 66)

Slackmaster2K

Slackmaster2K

Gone
CRIPES. I just lost this entire message.

Anyway, here 'tis again:

I did some tests with my new Maxtor 7200 RPM drive. "I thought you hate Maxtor!" I did.

I've always claimed in the past that HD RPM's were more important to performance than the ATA66 vs UDMA33 issue. In fact, I once claimed that UDMA33 would perform at the same level as ATA66. Was I right? Hmm...

Please note that the following is not exactly "scientific"....don't take it without a grain of salt.

The test system:

Abit BE6 Mobo
Onboard HPT66 ATA66 controller
Onboard Intel 82371AB/EB IDE controller
Celeron 400Mhz @ 500Mhz (83Mhz FSB)
128MB PC100 SDRAM
7200 RPM Maxtor Diamond Max HD, 2MB buffer
CL Riva TNT Blaster 16MB
Netgear FA310TX NIC
Ensoniq AudioPCI
Windows 2000, 1 FAT32 Partition

Benchmark Software:

dskbench - http://www.sesa.es/dskbench/dskbench.htm

Testing procedure:

1) Connect the hard drive to the HPT66 controller via a ATA66 cable. Run dskbench three times.

2) Connect the hard drive to the IDE controller via a standard cable. Run dskbench 3 times.

(scientific, eh?) :)

RESULTS (all numbers are from an average of THREE consecutive test runs):

Very very interesting.

Dskbench measures disk performance when the disk is being nailed with many large files all at once...very much like in a multitrack audio environment. I am fairly confident in the numbers it generates.

The nice thing about dskbench is that it compiles results at various block sizes. A "block size" is the amount of data that is read from or written to the HD at once. I'm not certain, but I believe that most audio software will use 16K or 32K block sizes. I'm going to contace Flavio Antonioli, author of n-Track, about this.

Let's see now, initially things are right where we might expect...with ATA66 the leader:

ATA66 Read Speed: 28.07 MB/s
ATA33 Read Speed: 24.26

ATA66 Write Speed: 27.95 MB/s
ATA33 Write Speed: 22.90

Wanna see a rather large graph? http://www.imt.net/~blarson/ata/initial.gif

"That's great, but how about pushing the drive a bit?" Ok, let's take a look at sustained transfer rates at various block sizes:

ATA66, 128K: 11.35 MB/s
ATA33, 128K: 9.47

"Yeah, ATA66 is smokin!" How true, how true. Almost 2MB/s faster! (that's a lot BTW)

ATA66, 64K: 6.62 MB/s
ATA33, 64K: 5.49

"Hmm...that's not quite as noticable, but ATA66 is still over 1MB faster than ATA33!" True, true!

ATA66, 32K: 3.44 MB/s
ATA33, 32K: 4.08

"Hey smackasser, you made a typo!" Nope, at a block size of 32K, ATA33 is slightly faster.

ATA66, 16K: 1.90 MB/s
ATA33, 16K: 2.79

"Oh my god, my world is crumbling!" Using a block size of 16K, ATA33 is almost 1MB/second faster than ATA66!!!!

Here's a big giant graph of the data in...ummm...a big giant graph: http://www.imt.net/~blarson/ata/sustained.gif

"Ok, slack...enough of your fancy numbers. Gimme somethin I can use!"

Ok, you understand tracks? Let's look at it again, considering maximum 16bit, 44.1Khz audio tracks:

ATA66, 128K: 134.7
ATA33, 128K: 112.6

ATA66, 64K: 78.7
ATA33, 64K: 65.3

ATA66, 32K: 40.9
ATA33, 32K: 48.5

ATA66, 16K: 22.6
ATA33, 16K: 33.2

"Yipes!!! You got a huge-ass graph for this shit too?"

Sure: http://www.imt.net/~blarson/ata/tracks.gif

So what does this REALLY mean? I dunno. It's one system with one hard drive on one controller. Remember that. Plus we don't know what block size our software is using.

Wanna talk CPU usage? Well, the standard IDE controller was a little better than the HTP66 controller at about 1 - 2%. The HPT66 controller was more around 1.5-4%.

So, I'll be getting some more info to ya soon.

In summary:
http://www.imt.net/~blarson/ata/initial.gif http://www.imt.net/~blarson/ata/sustained.gif http://www.imt.net/~blarson/ata/tracks.gif

BTW, these numbers that I'm seeing are MUCH MUCH MUCH MUCH better than my old 5400RPM drive. More on that later.

Happy tracking.

Slackmaster 2000

[This message has been edited by Slackmaster2K (edited 04-24-2000).]
 
Slack...I went with a Diamond Max because my board is set for dma33 (Asus p2b-2 or something) and I didn't want to spend the money on a controller at the time. The reason for the Max is my computer guru showed me a chart he downloaded (maybe I can find it) showing performance compared to 4 other major dma66 contenders. I think Western Digital did the comparison so it wasn't biased in favor of the Maxtor. It just happened that the Diamond Max held it's own when tested along with every other HD on the market, and ended up as one of the top 5 finalists when Western did the test.
 
Ok, the word from Flavio is:

"The block size for mono 16 bit files should be equal to the playback buffer
size multiplied by the "disk loading buffers x times bigger..." setting.
For stereo files the block size is obviously twice as big."

With n-Track's default settings, a block size of around 70K is used. Therefore ATA66 will perform better. However, using lower buffer settings (for better response) will lower the block size, and UDMA33 would perform better.

Interesting.

Slackmaster 2000
 
Thank you very much Slack- or shall I say Ultra Slack. I wonder how a different controller (Promise) would change the results. Maybe ther would be a way to switch the controller on and off depending on your application.

This stuff really is confusing isn't it! I've read so many reviews on hardware the past few weeks that I think I'm going crazy!

Thanks for making me even a little CrAzIeR!!

-jhe
 
Yeah...like I said, this is just what I'm seeing on my system. Maybe I can get someone (ED?) with a Promise controller to repeat these experiments.

Maybe we should start some sort of benchmark database for DAWs...

Slackmaster 2000
 
Hey Slack, send me details about what I would need to do to conduct some benchmarks with the Promise controller and my two hard drives. I need to know where to get the software, and of course, how to use it. But I would really like to get involved with doing so.

I am a firm believer in the "real world" tests. Kind of like the Monster Cable test I did. There is just so many variables involved with upgrades and supposed "improvements" that is it hard to know what you will "really" get with any new purchase.

I am not sure though how much a test I would do will really say anything. My Maxtor drive is udma66 7200 rpm, while my Western Digital is udma33 5400 rpm. Will the rpm's effect test results? Oh wait, I just re read the post. I see. You want to see if the Promise card will increase performance.

I am game. Just tell me what to do.


Ed
 
I read your test results very carefully and all the other postings for that matter. I'll just be brief here........I have a new Maxtor 20.5 GB Diamond Max/7200 rpm with 2mb cache still in the box that I bought a couple days ago(Haven't had time to get it installed yet).....I don't think my motherboard chipset supports DMA66..Will I still notice a major increase in performance from my Fujitsu/5400 rpm with 11msecs. access time???????? I hope so....that's why I bought it.
Running p2 450, 192mb ram, windows 98.
Thanks......
 
Supersonic...I had an old Samsung 4 gig dma33 in my computer before I switched to the Maxtor Diamond Max 20 gig dma33. My performance actually went down. This was according to two tests I ran. One from my computer guy which is stored on my other computer so I don't remember the name of the test program, and one that anyone can access on the Winmag website called Wintune. The circumstances were a bit strange, however, as I had to use the Win95 drivers while running Win98. Otherwise both drives would have ran miserably.
 
And the access time is negligable. What you're after is uncached transfer rate. My old Samsung was putting out close to 20 megs whereas the Maxtor, using the same drivers, is only putting out around 15 megs.
 
Monty,

Access time has quite a bit to do with sustained transfer. Think about it. It is actually directly related to RPM's.

Supersonic,

Yes you *should* notice a considerable difference in overall disk performance. *However*, the word is that Maxtor drives don't like to do UDMA in Windows95/98 (Is this true ED, or was it NT4?). I knew that I would be using Windows 2000, which is why I decided to buy the drive anyway (got a good price locally). If the drive runs in DMA mode, just as your 5400RPM Fujitsu drive did, you'll notice the difference.

Ed,

Download dskbench from the location I specified in the original post. Then basically run it three time for each hard drive, for each controller, and record the results. If you don't have Excel, you can send me the results & I'll make cute little (gigantic) charts for ya.

Dskbench is a console app so the best way to run it is from the command line.

I just wrote a big giant text about how to do this....but I think it'd just confuse the issue. I'll just say, "re-read my post and recreate it using your drive & controller". Remember, I did three runs and took the average for a total of 6 runs.

Here's a tip:

dskbench > filename.txt

The ">" operator will pipe the output of a console app into a file. This way you don't have to write down the results by hand each time...just specify a different file for each run.

Also, dskbench takes about 5-10 minutes to run and you won't see any output while it's running if you're piping the output to a file. Be patient.

If you have any questions then we should continue this via email.

Thanks, this'll be interesting!

Slackmaster 2000
 
You must log in or register to reply here.
Back
Top