One Click on the Banner shows your site support to my Sponsors

Accelerate Your Mac! Logo
PCI SCSI Card vs Onboard SCSI:
Tests w/Stock 9600 IBM 4GB Disk

Published: 9/25/98 12:00 AM EST


With the recent lower pricing on PCI Ultra-SCSI controllers, I've been getting many mails from readers who are asking about performance increases using a PCI SCSI card with their stock (narrow) hard disk. As I've commented many times I don't suggest using PCI SCSI cards with the narrow, slower drives that normally come standard with most Mac models. You won't 'see' a real improvement in speeds on these drives in most cases, as the onboard SCSI of most Macs is 10MB/sec, faster than most narrow SCSI drives can sustain. The real reason to buy a PCI UltraSCSI controller is to add wide ultra SCSI drives and to run Raid striped volumes (multiple drives configured as one volume). In these cases the onboard SCSI is far too limited to allow the drives to reach their full potential. The exception to this rule would be those Macs that have only 5MB/sec onboard SCSI (Apple G3, and some older/lower cost Macs have 5MB/sec SCSI onboard). For those Macs the onboard SCSI is a bottleneck for most quality SCSI hard disks (but not CDroms, ZIP drives, etc.).

To prove this point I decided to test the stock Apple/IBM 4GB hard disk in my 9600/350 (overclocked to 400MHz) with the Jackhammer and Initio Miles PCI SCSI cards vs the 9600's 10MB/sec onboard SCSI to see what benefit if any would be obtained. As I suspected, sustained rates were no better with the PCI SCSI cards as the drive itself was capable of only 5 to 6MB/sec sustained rates according to the ATTO benchmark tests As shown in the graphs I will present here, there were cases where peak rates were much higher with the PCI SCSI cards, but my (limited) real world tests in the finder were very disappointing. The PCI SCSI cards were actually slightly slower at standard file system operations in OS 8.1 than the builtin SCSI in 4 tests I performed:

  1. Duplicating a 10.5MB Folder (2 files)
  2. Launching Photoshop 5
  3. Loading a 4.4MB file in Photoshop
  4. Saving a 4.4MB file in Photoshop

I didn't expect any real improvement with the stock drive, but I was surprised to find it the PCI SCSI card was actually slower. Possibly due to the 50MHz bus speed vs 33MHz PCI bus speed affecting file system overhead. Results may have been different with another system, as I wonder if the slow memory performance of the 9600 Kansas (Mach5) motherboard could be affecting the results. As reported in the past here, the PCI video card scores in the 9600 are usually up to to 25% slower than they should be based on tests in comparable speed Mac models not based on the Kansas motherboard. Results may vary with other disk drivers or disk models as well, as I used only Apple's drive setup v1.5 and the stock Apple/IBM 4GB OEM drive in these tests.

Looking at the graphs of the data I could guess the size of the stock hard disk's cache - 128K. This is especially evident in the 256k-8MB test range, where you can see the rates drop after 128K file sizes are exceeded, and the rates drop dramatically to reflect the actual drive platter (disk surface) performance instead of the interface to cache speed. The graphs also show that the stock SCSI interface is a bottleneck for peak rates writing to the hard drive's cache, but it is not a limiting factor once file sizes exceed the size of the hard disk cache. For the graphed tests, the OS system disk cache was not used, to better show the actual interface and drive performance, without being affected by the memory control panel's disk cache size.

The ATTO Benchmark tests were run at all possible settings, representing file transfer sizes of up to 32K, 128K, 512K, 2MB and 8MB. The smaller file sizes would be indicitive of database applications for instance, the larger ones applications like video capture.

Test System:

  • 9600/350 (overclocked to 400MHz)
  • OS 8.1
  • 256MB RAM (VM Off)
  • 4MB Disk Cache Setting (not used in Benchmark tests)
  • Stock 4GB IBM Hard disk - model DCAS 34330 (500MB Free)
  • Radius Thunder 3D graphics card in PCI slot A1 (top slot)
  • PCI SCSI cards were tested in slot D2 (4th slot)

Summary of Results:

Benchmarks:

Notice the much higher peak rates with the PCI SCSI cards over the stock 10MB/sec SCSI interface. At times there is even a small sustained rate increase, but the hard disk itself is the bottleneck as far as sustained rates.


ATTO Benchmark Tests
(PR=Peak Read, SR=Sustained Read, PW= Peak Write, SW=Sustained Write)
All numbers are MegaBytes/Sec
File Size Test
MB SCSI
Jackhammer PCI SCSI
Initio Miles PCI SCSI
32K
PR: 7.88
SR: 7.73
PW: 4.51
SW: 3.69
PR: 15.00
SR: 13.14
PW: 5.49
SW: 4.88
PR: 14.68
SR: 13.62
PW: 5.54
SW: 4.81
128K
PR: 8.60
SR: 6.58
PW: 6.24
SW: 5.05
PR: 18.37
SR: 13.57
PW: 8.14
SW: 5.94
PR: 18.15
SR: 13.02
PW: 9.57
SW: 4.37
512K
PR: 8.55
SR: 4.56
PW: 7.29
SW: 5.01
PR: 18.26
SR: 5.01
PW: 8.71
SW: 5.44
PR: 18.09
SR: 4.98
PW: 8.32
SW: 5.21
2MB
PR: 7.01
SR: 4.93
PW: 5.83
SW: 5.09
PR: 12.15
SR: 5.44
PW: 5.77
SW: 5.53
PR: 10.32
SR: 5.43
PW: 6.89
SW: 5.45
8MB
PR: 5.08
SR: 5.06
PW: 5.44
SW: 5.08
PR: 5.59
SR: 5.57
PW: 6.03
SW: 5.52
PR: 5.59
SR: 5.56
PW: 6.25
SW: 5.50



Real World Tests:

These were tests performed in the finder. 4MB disk cache size. Stock HD had 500MB free and was not optimized/defragmented. Tests were repeated twice to verify results. These results may be different in another Mac (non-Kansas motherboard) and should show better performance with a faster drive (wide SCSI or 7200/10000 RPM drive).


Test
Stock HD/SCSI
Stock HD/PCI SCSI
Duplicate 10.5MB file
4.47
6.47
Launch Photoshop 5
10.44
10.63
Load 4.4MB file in PS 5
7.31
7.47
Save 4.4MB file in PS 5
3.50
3.65



Graphs of Results:

(Note the spikes in the transfer rates - not good...)

32K File Tests Results (Stock SCSI, Jackhammer, Miles)






128K File Tests Results (Stock SCSI, Jackhammer, Miles)





512K File Tests Results (Stock SCSI, Jackhammer, Miles)







2MB File Tests Results (Stock SCSI, Jackhammer, Miles)




8MB File Tests Results (Stock SCSI, Jackhammer, Miles)



 

Other site SCSI reviews:
For more SCSI tests with all the popular PCI SCSI controllers and wide drivces, see my SCSI Features page. For general information on SCSI basics - check out the SCSI area of my Links page.


Back to XLR8YOURMAC.COM
Your Source for the best in CPU/SCSI/VIDEO card reviews, daily news, and more!


Copyright © XLR8YOURMAC.com, 1998.

No part of this site's content is to be reproduced in any form without permission.
All brand or product names mentioned here are properties of their respective companies.
Disclaimer: Users must read and are bound by the Site Terms & Conditions of Use.