Ultra ATA 66 Hard Drive Roundup - March 2000

The AnandTech storage test bed is currently set up using Windows 98, Windows NT 4.0 Service Pack 6 and Windows 2000. Each operating system is set up using its own 4 GB partition with Windows 98 using the Fat 32 file system, Win NT 4.0 using the NTFS file system and 2 instances of Windows 2000, one using the FAT 32 and the other using the NTFS file system. Since Windows 2000 supports both file systems, we decided to set up two platforms to help compare the performance under both systems. This will help the reader make an informed decision as to which drive performs best under each file system, for users planning to upgrade to Windows 2000. Also, all of the tests were run using the Ultra ATA 66 protocol, using an 80-pin IDE cable. In all instances, the only installed software is the operating system and the benchmark programs.

Some of you may be wondering about our choice of a test bed, most notably the use of a Celeron 500 processor and the Intel 810E motherboard. While this would not make a good choice for the average user looking for maximum performance, as the integrated video graphics leave much to be desired in terms of gaming, it is perfect for use as a test bed. What we are looking for is a solid system, with an integrated Ultra ATA/66 controller, so as to ensure consistent results between tests. The idea here was to eliminate as many variables as possible. Also the 810E supports SDRAM, which was another concern.

It was decided early on to use a motherboard based on the i810E platform because of the fact that the i810E chipset makes use of Intel’s 801AA I/O Controller Hub (ICH) which functions properly as an Ultra ATA/66 compliant controller. The i810E also boasts Intel’s Accelerated Hub Architecture which features a dedicated 266 MB/sec path to the ICH, which helps eliminate any potential bottlenecks that are present with non-AHA chipsets that use the PCI bus as an all-purpose bus used to connect the integrated disk controller with the CPU/memory.  While there are other options such as AMD’s 750 chipset, the performance of a hard drive from one platform to the next should not vary much, so the selection of an i810E motherboard as the test bed should not make a difference.

Using a Celeron 500 as the CPU for our storage test bed was a decision made in order to represent the average of what most users find in their systems as well as because the CPU is sufficient for the needs of our test bed.  Using a faster CPU would not change the outcome of any of our tests.

The complete test bed is as follows:

·         Intel Celeron 500 MHz CPU

·         Intel CA810E motherboard with native Ultra ATA/66 support

·         128 MB Corsair PC100 SDRAM

·        Ultra ATA/66 40-pin 80-conductor HDD cable

·         Microsoft Windows 98

·         Microsoft Windows NT 4.0 SP6

·         Microsoft Windows 2000 (FAT 32)

·         Microsoft Windows 2000 (NTFS)

·         Ziff Davis Winstone 2000 CC

·         Ziff Davis Winstone 99

·         Ziff Davis Winbench 99

·        Adaptec Threadmark 2.

Each hard disk was partitioned and formatted before each suite of tests on their respective operating systems, as to prevent any skewing of the test results. For purposes of consistency, each benchmark was run a total of 5 times, with the final score being the average of those five runs.

Adaptec’s Threadmark 2.0 is a popular benchmark and we used it to help further illustrate the differences between the drives. However, as it is almost 2 years old and is no longer supported by Adaptec; it would not run under Windows 2000, so only Threadmark results for Windows NT and Windows 98 are shown. Ziff Davis’ Winbench 99 was used to show the real-world transfer rates achieved by the individual drives. These two tests are good at demonstrating the sustained data transfer rates of the drives and are useful indicators for the type of applications that access the disk in this type of manner. A good example of this would be digital video and audio editing where data is often accessed sequentially and the raw transfer speed is important. The main consideration for this type of work is going to be a disks rotational speed, as a higher rotational speed will translate into better sequential read/write performance.

Ziff Davis Business Winstone 99 and Content Creation Winstone 2000 were used to show the real world performance differences between the drives. These tests open multiple applications simultaneously and switch between them doing various tasks. These types of tests utilize a lot of random disk accesses and a low seek time in addition to a higher rotational speed and data cache will improve performance here. These tests will more closely represent how the majority of home and business users will access their hard drives and are a useful all around indicator of a hard drives performance.

It should be noted that we could not get all of the commercial tests to run properly under all of the operating systems, and as such you should only compare results across the same platform when comparing the drives. The problems were mainly under Windows 2000 and we don’t know if it was driver problems or if Winstone isn’t 100% Windows 2000 compatible yet. Adaptec’s Threadmark 2.0 refused to even display the drives under Windows 2000, but considering it is a two year old benchmark that is no longer officially supported, it will only be useful for Windows 98 and NT tests.