ATA 133:
The new ATA 133 standard is providing a theoretical maximum data throughput of 133 MB/sec. That means in theory a 33% higher data throughput than with ATA 100 at 100 MB/sec. However, ATA speeds are referring to the direct transfer speed between the disk and the interface. In practice data are running or handled by various layers of drivers, the OS and a transfer speed of 100 or 133 MB/sec is in reality not achievable. In their ATA 133 white-paper Maxtor gives a conservative efficiency estimation of 62%. This means a transfer rate of around 62 MB/sec for ATA 100 disk and around 82 MB/sec for ATA 133 disk. A closer look shows that ATA 133 is nothing else than an enhancement of the existing ATA 66 and ATA 100 protocols and is not a new technology by itself. The data width of 16 Bit and the 40-pin plug are identical and the same 80-core ribbon cable used for ATA 66 and ATA 100 is used for ATA 133. The higher transfer rate of 133 MB/sec is achieved by shortening the cycle time from 40 ns (ATA 100) to 30 ns (ATA 133). The cycle time refers to the time window within a certain task has to be completed. ATA 133 is currently not an open standard like ATA 66 or ATA 100, but a Maxtor-owned proprietary technology. Maxtor is asking other manufacturers who want to use it for license fees.The T13 committee responsible for setting the ATA standards is currently working on authorizing ATA 133 as an official ATA standard.

We were using for this review 4 samples of Maxtor’s ATA 133 D740X with 40 GB capacity and 7,200 rpm spindle speed, all of them made in Japan. These drives come in 2 flavours with a fluid ball bearing motor (“L” type) to minimize noise, and a standard ball bearing motor (“J” type). We used the J drives with the BB motor; they were silent nonetheless. For the ATA 100 counterpart we used 4 Western Digital Caviar WD400BB with 40 GB capacity and 7,200 rpm. These WD drives are widely used. They are acknowledged as solid and issue free ATA 100 drives. In the huge database of the soon defunct StorageReview they are ranking in most benchmarks among the top performing ATA 100 disks. 

Maxtor and Western Digital Drives Tested

The Tests and the Test-bed:
Our question here is: how fast are the ATA 133 based HPT 372 and 374 controllers in comparison with an ATA 100 based solution? Although we don’t regard the widely used STR as very valid to benchmark hard disks, we took transfer speeds anyway. Transfer rate, disk access time and CPU utilization were taken with the Winbench 99 “Disk Inspection Test” and the disk benchmarking utility “HDTach 2.61”. Of more importance for the evaluation of hard disks and RAID 0 performance are the application based Winbench 99 (Version 2.0) “High-end Disk Mark” suit and Ziff Davis’ “Content Creation 2001”. In particular Content Creation is apt to bench RAID 0, because it is mainly based on audio, video, and picture editing applications like Sound Forge, Adobe Premiere, and Adobe Photoshop. Both benchmarking programs are for free available from Ziff Davis here. Content Creation has to be ordered on CD. All test were run as demanded in the test manual:

• de-fragmentation  and rebooting before each test run
• idle tasks before each test run
• 10 repetitions of all tests

Recently there has been a lot of discussion about the optimal RAID 0 block stripe and file cluster size. Promise is recommending in the FastTrack 100 TX manual a stripe size of 8k for usual PC jobs and a minimum of 64K when working with large files such in audio/video editing and for server operations. The default for Promise Raid FastTrak 100 controllers is 64K. Highpoint doesn’t give a recommendation, but the default stripe size for RAID 0 in the BIOS and the Windows based RAID management utility is also 64K. A recent AnandTech review showed that both controllers, the Promise FastTrak series and the Highpoint are indeed working best with 64K. Another review on VIAHW using the same benchmarks as we do in our review, is pointing at a smaller stripe size of 16k or 32K. We tested stripe sizes of 16K, 32K, and 64K with the respective matching file cluster size and did not find any significant or consistent difference. In fact, the performance difference between different stripe sizes were smaller than the fluctuation within the results of the 10 consecutive test runs we execute for reach test setting. The results reported hereafter are based on a stripe size of 32K and a matching NTFS cluster size of 32K, which seems to be a good compromise.

Issues and Non-issues:
Soon after the first ATA 133 capable boards hit the market it was discovered that VIA chipsets have obviously a problem handling the increased burst speed of ATA 133. “Motherboards with VIA chipsets start to interrupt the burst shortly after the cycle has started. These interruptions taking place every 1.2 µs make the transfer rates of ATA 133 controllers drop down to 64 MB/sec in a worst case scenario”.  An in-depth explanation can be found here at techchannel.de which discovered the issue shortly after boards with the KT266A chipset started retailing.

2 patches are available to remedy the problem: George Breese’s PCI Latency Patch and VIA’s own VIA PDF patch. We have tested both patches and found:

• Both patches increase the transfer speed by around 20% or up to 10 MB/sec.
• Both patches have very little impact on application-based benchmarks ranging from 3% in Winbench 99 applications to 2% with Content Creation 2001.
• The latest patch from George Breese (1.9b) worked slightly better than the VIA PDF 1.04 patch. We therefore ran all tests with the Breese patch.

Unfortunately we experienced some serious problems with our Promise FastTrak 100 TX/4. When running the TX/4 on our KR7A we noticed that several ACPI related functions like stand-by, hibernation, and the full power shutdown did not work. Only after we removed the TX/4 and booted from a single disk the ACPI functions were working properly again. We then tried the TX/4 on 2 other KT266A boards, the Epox 8KHA+ and the Gigabyte 7VTX-H from our previous KT266A comparison. On none of these KT266A boards the TX/4 was working properly showing the same symptoms we experienced already with the KR7A. We even went back to Windows 2000 to exclude the possibility of a Windows XP related issue, but the problem remained. On 2 of our AMD 760 boards, the Gigabyte 7DXR and the Epox 8K7A, the same TX/4 however was working without a hitch. We therefore have to assume that the TX/4 has a problem with the KT266A chipset. In addition to the already described problem with the missing ACPI functions we experienced several times BSOD errors related to the TX/4 “evian.sys” driver. At that point we had a look at the Promise website for help. We could not find anything about the TX/4. Although the TX/4 is selling for almost 1 year now, the BIOS has not been updated and the TX/4 is not even listed in the FAQ section. We then contacted Promise by e-mail. After several attempts we finally received a beta driver for Windows XP without any further comment. After installing this driver neither Windows 2000 nor Windows XP would boot anymore. Several attempts to clean install XP with this beta driver failed, we were only greeted by a BSOD again hinting at the TX/4 “evian.sys” drivers and we had to reinstall Windows XP with the original drivers from the TX/4 installation disk. We again tried to contact Promise but did not receive any comment or help. We also found that the TX/4 is very sensitive in regard of over clocking attempts, even mild attempts to overclock the PCI bus, resulted in BSOD with error messages again hinting at the dreaded TX/4 “evian.sys” driver.

The only problem we had with the Rocket 404 been that our test board (KR7A) did not recognize the card in PCI slot 3. As recommended in the Rocket 404 manual we changed the slot and the card was initialized without further problem. We were running our 8 disks for 2 weeks almost 24 hours, in all kind of torture tests. RAID 0 x 4 is a delicate affair, if only one of the drives fails, all data are lost. Our 8 disks from Western Digital and Maxtor were running all tests flawless and never failed us.

NEXT PAGE - Results, Conclusions and Recommendations

---

Other Recent Reviews:

DIY Parallel-Cable Remote LCD!

TwinMOS PC2700 DDR

MSI K7N420 Pro nForce Motherboard!