|Accelerate Your Mac!|
Bring in the Noise
Mac Audio Column
by Thad Brown
|A $640 24 GB ATA 20MB/second Audio Dweeb Approved Disk Array
As I mentioned last week, I saw what Mike got for performance from a bargain Maxtor IDE drive, and got to thinking. I said to myself, "Thad, if you could save some money on drives, you could get your girlfriend some roses, or help an important political cause, or take some guitar lessons, or best of all, buy some more microphones." With the TurboMAX PCI ATA card costing only $99, and with my mini-tower case having room for four hard drives and two CD drives internal, I decided to give it a try. Luckily for me I didn't have to take on faith the level of performance--I had a Maxtor 6.5GB drive laying around, so I could just buy the card and test its performance. The Maxtor benchmarked more than adequately and I ordered a pair of IBM 12.9GB 7200RPM drives for $253 apiece. I chose IBM for a few reasons, the main one being that in my experience they are some of the quietest drives around, and it's not like they are likely to go under any time soon. Also, the ability to use a standard Apple driver and format utility is nice.
Installation went easily as far as hardware was concerned. The UMAX S900 case is beautifully designed and easy to work in. The main drives bays are in a cage, with another bay perpendicular near the base and a final bay where a plate can be screwed into the bottom of the drive and then it is suspended upside down just below the floppy drive. The case seems identical to a Dell case that I was working in recently, so if you are familiar with a good PC case, you know what this is like. Apple Drive setup recognized the IBM drives no problem, but of course the Maxtor is "unsupported" so I tried pretty much every driver I had, and settled on an ATTO driver. I had some flakiness with the Maxtor, and I may just get another IBM to replace it when I have a few extra dollars. The main function of the third ATA drive in my setup will be for low level performance things like sample and REX file archiving, Linux, and whatever else I can think of. I am really looking forward to being able to keep the contents of a half a dozen (and maybe later many more) of my sample CDs on an internal drive, so I can grab a sample or drum loop for a song without hunting for audio CDs and ripping and importing and what not. Here is a table with the ATTO benchmarks for each drive. I ran all of the tests at least three times to be sure I didn't have any really strange results. The drives are my IBM 5400 RPM SCSI boot disk that came with the UMAX, a 4.5 GB narrow Barracuda that I got real cheap, it's a first generation drive that probably does not reflect current Barracuda capabilities, the IBM ATA drives, and a Maxtor DiamondMax 5400 RPM ATA drive.
These results look fairly predictable to me. The higher bandwidth drives performed better, and the IBM 7200 RPM drive topped everything else by a long shot. The lowest sustained write on the IBM was enough to theoretically record 80 tracks of 24 bit audio, and the 19 MB/s is enough bandwidth for well over 100 tracks of playback. I wish I could have had a newer high density Maxtor to test, and also it would have been nice to do RAID testing, but I don't own (or plan to use) any RAID tools, so I'll have to leave that to somebody else. I also tried the card in various PCI slots in my machine. Since I don't think this is of general concern I didn't put it in the table, but swapping the card into different PCI slots in the S900 DID make a difference in performance. Sustained reads and writes were nearly identical, but in each benchmark, peak reads and writes were much higher, though they were all good. Looks like the DEC PCI bridge chip does make a difference.
As unsurprising as this chart is, do remember one thing--I paid $253 for an IBM drive that had an average sustained write over 10 MB/s and peak reads over 20. File copies were comically fast, and opening and closing was also as snappy as one would expect. Not bad for that kind of money
In audio specific applications the combination of the ATA drives was excellent for playback, and after some tweaking, just as good for recording. A critical difference between SCSI and ATA is that the ATA drives use many more CPU cycles than SCSI drives do. It seems that the drive controllers take a static amount of bandwidth--I tried testing at different CPU speeds, and surprise, the faster the clock speed the less there seemed to be a noticeable hit with CPU performance. In my unscientific testing of "turn on reverbs until the thing chokes" the ATA drives would max out the CPU a little earlier, but it was a fairly small difference, a handful of EQs or one compressor. In playback, the ATA drives truly shined, however. Playing back a 32 track 24 bit tune was not even enough to make the disk meter got to halfway up.
I spent lots of time and effort on getting the ATA drives to record well, but in the end it paid off. The first time I wrote this, I was convinced that the TurboMAX and IBM drives simply would not record well. The problem LOOKED like an extension conflict, the CPU meter would be a fifth or a sixth of the way up, and then suddenly, "bam" overloads and playback cuts our for a short time and then resumes. I tried different slots for my audio card, I tried using the built in converters, I tried disabling one or both drives, I ran 256 colors instead of millions and disabled graphic acceleration. Pretty much everything that could be done, but still, recording a new track while playing back a very modest number of other tracks (like say 12) would invariably produce at least one or two stutters during a one minute recording pass. My much lower benchmarking Barracuda had no such problems with such a small number of tracks. After nearly giving up, I decided to try maxing out the disk cache in Cubase to see if I gave maximum memory to the app with maximum cache, that would help. With that setting, practically nothing would record at all, which seemed strange. Then it clicked, the problems were with CPU overloads, not with disk overloads. Audio apps have an easier time dealing with smaller caches (less data to access at a time), and drives get an easier go with larger caches (less time spent seeking and waiting). By dropping the cache to a lower level, suddenly all of the problems disappeared. I could play back 31 tracks of 24 bit audio and record another without CPU overloads or disk problems. Different caches provided different levels of performance, but a good compromise setting was 48kb. It appeared that the drives and computer could have done even more impressive record/playback fears, but I think that 32 tracks of 24 bit 44.1k audio is, well, sufficient would be an understatement. I don't know how much more I would ever need.
The only negative I have found is I still have not gotten what I feel are acceptable levels of reliability and performance with CD burning. Partly this is due to the maze of drivers that one must load to use a non-Apple CD-ROM and a CD burner in the same machine. It would be very nice to get a generic CD-ROM driver that will allow basic functions of SCSI CD drives and burners. In any case, I'm working on why I get more errors that I expect, and why I can't dupe audio CDs. So far I have had to create an image of a disk before I could burn it. I'll post updates and workarounds as I find them, but I have had promising results by moving the Toast or Jam swap file onto a disk other than the system disk. You can choose which disk is used for the cache in the preferences menu, accessible from the edit menu in Toast or Jam. Don't know why it should matter, but it seems to.
It's hard to argue with this kind of performance for this kind of money. Despite some hassle getting things to work, this much storage that specs like it does for me is worth a little time figuring out how to make things work. One of the best things that Apple has been doing recently is trying to conform to industry standard hardware and components. I think that this is a good thing, and support for ATA drives as a cost cutting measure in the new G3 boxes has the side benefit of Apple drivers that support many ATA drives that perform well at a fraction of the price of SCSI iron. Thumbs up to the cats in Cupertino.
Like many Mac users, and in particular Mac audio users, I have always been a SCSI snob of the worst kind. One of my chief complaints about new Macs was what I used to think of as sketchy IDE drives. SCSI has attained nearly unquestioned superiority in the world of audio, and the Barracuda in particular has become the automatic drive of choice for audio people. Of course, there is absolutely nothing wrong with an U/W SCSI card and high end drives, and I still believe that they would outperform my ATA system in the same machine. I understand the attitude that more performance is just better, and if you have the cake, get the best. I honestly respect the "performance is its own reward" way of life, but since I have to buy analog equipment in addition to my performance enhancing digital equipment. So if you don't need the best (and I don't), it may be time to look into non-SCSI drives. My SCSI snobbery was not nearly enough to keep me getting 24 gigs of audio approved drives for $525. As I keep on saying here at Bring in the Noise, all I care about is audio, and the money I saved on drive bandwidth over a similar SCSI system is enough to buy a very nice microphone or a bunch of plug-ins. That's not a very hard choice for me to make.
Your Source for the best in CPU/SCSI/VIDEO card reviews, daily news, and more!
All brand or product names mentioned here are properties of their respective companies.