Reviews Leaderboard Database Reference Search StorageReview Discussion Reliability Survey Search About StorageReview.com Contents

ATA-66 vs. ATA-33 - Test Results and Conclusion

  August 23, 1999 Author: Eugene Ra  

HDTach v2.52 under Windows NT 4.0
BenchmarkQuantum Fireball Plus KX - ATA-33Quantum Fireball Plus KX - ATA-66
Burst Read Speed (MB/sec) 28.2 46.7

The first test consists of the HDTach v2.52 Burst Read Speed test. To confirm that the Fireball Plus KX was indeed running in two separate modes for our tests, we needed to confirm that Burst Speeds between the two modes did indeed straddle the 33 MB/sec barrier. As we can see, ATA-66 operation nets a 46.7 MB/sec burst transfer rate, ATA-33 a 28.2 MB/sec results. A warning for those of you who like to jump to conclusions: This is a low-level test, a test of buffer-to-host transfer rate. A substantial increase should be expected when jumping from ATA-33 to ATA-66. Lack of a difference would cast doubt on whether the drive was indeed running in two separate modes. Let's take a look at some other low-level figures:

Ziff Davis WinBench 99 under Windows 95 OSR 2.1 using FAT 32
BenchmarkQuantum Fireball Plus KX - ATA-33Quantum Fireball Plus KX - ATA-66
Disk/Read Transfer Rate 
Beginning (KB/sec)23100 23100
End (KB/sec)14700 14700
Disk Access Time (ms)12.2 12.2
Disk CPU Utilization (%)6.38 6.60

Ziff Davis WinBench 99 under Windows NT 4.0 using NTFS
BenchmarkQuantum Fireball Plus KX - ATA-33Quantum Fireball Plus KX - ATA-66
Disk/Read Transfer Rate 
Beginning (KB/sec)23100 23100
End (KB/sec)14700 14700
Disk Access Time (ms)11.8 11.9
Disk CPU Utilization (%)2.75 2.85

As we can see here, going from ATA-33 to ATA-66 operation results in no changes to mechanically-limited low-level test results. There's zero change in transfer rate both at the outside and inside tracks. The Fireball Plus KX is capable of sequential transfer rates up to 23 MB/sec. Should increasing the speed limit from 33 MB/sec to 66 MB/sec result in higher sequential rates? Of course not. Similarly, access time, which has nothing to do with maximum transfer rates, remains unchanged when going from 33 to 66. Lastly CPU utilization scores remain fairly uniform between the two modes. Reports that heading to ATA-66 from ATA-33 decreases CPU utilization are simply false. Such conclusions, once again, are likely drawn from the variables left uncontrolled by an unaware tester.


Ziff Davis WinBench 99 under Windows 95 OSR 2.1 using FAT 32
BenchmarkQuantum Fireball Plus KX - ATA-33Quantum Fireball Plus KX - ATA-66
Business Disk WinMark 99 (KB/sec) 3647 3677
High-End Disk WinMark 99 (KB/sec) 12267 12467
AVS/Express 3.4 (KB/sec)8443 9313
FrontPage 98 (KB/sec)38067 38000
MicroStation SE (KB/sec)12633 12833
Photoshop 4.0 (KB/sec)8823 8843
Premiere 4.2 (KB/sec)9773 9313
Sound Forge 4.0 (KB/sec)17233 17433
Visual C++ (KB/sec)13867 13867

Ziff Davis WinBench 99 under Windows NT 4.0 using NTFS
BenchmarkQuantum Fireball Plus KX - ATA-33Quantum Fireball Plus KX - ATA-66
Business Disk WinMark 99 (KB/sec)4310 4340
High-End Disk WinMark 99 (KB/sec)13367 13733
AVS/Express 3.4 (KB/sec)16200 18100
FrontPage 98 (KB/sec)42133 42767
MicroStation SE (KB/sec)18900 19300
Photoshop 4.0 (KB/sec)8060 8140
Premiere 4.2 (KB/sec)10707 11400
Sound Forge 4.0 (KB/sec)12500 12333
Visual C++ (KB/sec)11500 11400

Here's the ultimate in results, what most of you have probably been waiting for while wading through this sermon . As we can see, there's little difference in aggregate high-level scores between ATA-33 and ATA-66 modes. The Business Disk WinMark 99, under both operating systems, reveals a difference of less than 1%. The High-End Disk WinMark 99 posts similar results, the two modes coming within 3% of each other. When it comes to the individual application breakdowns, significant differences arise in only two cases, Premiere and AVS/Express. Differences in Premiere can be dismissed as variation between trials. Our three trials, sufficient for nearly all tests, show one sub-standard ATA-33 score among two other scores that match ATA-66's performance. AVS/Express' results, however, are a different story. Here a consistent and uniform advantage of about 10% is exhibited in ATA-66 operation. Perhaps data visualization is one of those few applications that feature a high hit-rate when searching for data in a disk's buffer. It should be clear from these test results, however, that such applications are by far the exception rather than the rule.

ThreadMark 2.0 under Windows 95 OSR 2.1 using FAT 32
BenchmarkQuantum Fireball Plus KX - ATA-33Quantum Fireball Plus KX - ATA-66
Data Transfer Rate (MB/sec) 10.06 10.74
Average CPU Utilization (%) 35.52 39.51
KB/sec per 1% CPU Utilization (KB/sec) 283 272

ThreadMark 2.0 under Windows NT 4.0 using NTFS
BenchmarkQuantum Fireball Plus KX - ATA-33Quantum Fireball Plus KX - ATA-66
Data Transfer Rate (MB/sec) 12.95 13.02
Average CPU Utilization (%) 21.56 20.75
KB/sec per 1% CPU Utilization (KB/sec) 601 627

ThreadMark 2.0, a purely synthetic test, displays a negligible difference in Windows 9x between the two modes, but a 6.7% difference under NT. If one insists, he may conclude from this result that ATA-66 is indeed faster than ATA-33. In our book, it's yet more proof that ThreadMark is indeed a flaky benchmark.

The conclusion? Combine these test results along with the "performance increases" we've witnessed from previous jumps (up the PIO ladder and DMA mode 2 to DMA33) and it should be obvious that going from ATA-33 to ATA-66 in itself does not provide statistically significant performance increases. At this point, some readers may be inclined to think "Well, it's obvious that Promise is ripping the community off with its bogus ATA-66 controller, the Ultra66. A real ATA-66 product, such as Abit HotRod, would show the big differences that just have to be there between ATA-33 and ATA-66." A similar cynical accusation could be leveled against the ATA-66 implementation in Quantum's Fireball Plus KX. For our part, we've never had problems with either Promise ATA controllers or Quantum ATA drives. Here we're going to invoke the simple law of parsimony: Is it more likely that Promise and Quantum, two veteran players in the ATA industry, have no idea what they're doing and implemented "crippled" ATA-66 in their products? Or is it more likely that ATA-66, as shown by some carefully controlled tests, provides no significant increase in performance with today's drives?

Is ATA-66 a sham? Absolutely not. ATA-33 first shipped over two years ago in the form of the Quantum Fireball ST. Seagate's Barracuda ATA ST328040A, a drive due to ship in the near future, can push 28 MB/sec of data. ATA drives are coming very close to ATA-33's limits. Introducing ATA-66 drives and controllers at this time simply allows the industry and the community to place next-generation infrastructure in place before drives that truly need it arrive. It's advanced planning, folks, something that will benefit us down the road.

* Note: Threadmark 2.0 and HDTach v2.52 test results are the average of five trials.
WinBench99 test results are the average of three trials.


HOME | ARTICLES | LEADERBOARD | PERFORMANCE DATABASE | REFERENCE GUIDE
COMMUNITY | RELIABILITY SURVEY | SUPPORT SR! | ABOUT SR |

Copyright © 1998-2005 StorageReview.com, Inc. All rights reserved.
Write: Webmaster