In order to evaluate the various UDMA drivers available for Win95 OSR2,  I looked for benchmarks whose results could be duplicated and whose numbers seemed to correspond to one another. Benchmarks such as Wintune97, whose results of 52 Mb per second cached and 2.5 Mb per second uncached, did not correspond to other benchmarks' results and were not used.  I have enjoyed using various incarnations of Wintune for years and was a bit dismayed by the HD results of this version. I also decided against using a number of very good DOS based tests. This was done mainly for ease of testing and the fact that I found a Windows based benchmark that agreed with the DOS based tests' results.  I settled on three benchmarks.  WinBench98 and Threadmark 2.0 seemed to be widely quoted and the results are very easily duplicated.  Both of these also measure CPU utilization, which was a primary concern. The third, HD Tach 2.00, is a DOS-like benchmark that runs in Windows. It seems little affected by which drivers are used and served as a reference point.  Though I was interested in the actual numbers, I was more interested in whether they went up or down using each combination of patches and drivers.

One frequent question has been on how to remove the Intel drivers and get back to the default Microsoft units.  I have read about and tried many methods, but using Intel's uninstaller remains the best choice.  However, my best benchmarks have been achieved by a complete reinstall of OSR2. (article)

While some may disagree, I have found that the order the patches and drivers are installed does make a difference in performance.  I installed the files in the order that you find them listed on the charts below.

These tests are a "work in progress," with new drivers added as they arrive.  Check the bottom of the page for links to the newest tests and Win95 UDMA articles.

Links to the benchmarks and drivers are located on the last page.

The system tested is assembled as follows:

Abit AX5 revision 2.22 - 512KB L2 cache
AMD K6 233 o/c'd to 250 (83 X 3)
Quantum Fireball ST 3.2  - FAT 16
Quantum Fireball SE 3.2 - FAT 16
32 MB generic SDRAM
Matsushita CR-574 CD Each group of tests were started with a clean installation of OSR2.1 with only the following programs loaded:

TLIET600 Video driver
CPUIDLE 2.4
Direct X 5.0 (for WinBench98)
WinBench98 v 1.0
Threadmark 2.0
HD Tach 2.00

The system was rebooted after each file's installation and before each benchmark.  The HD was defragged before each benchmark.  Each group of tests was run three times and the results were averaged.

The top section of each chart shows the changes I observed and any changes I made in the Device Manager. The bottom section shows the results of the benchmarks.

My Choice
At the present I am not using any hardware requiring the installation of the USB patch.  This set up produces the best results for my needs.

Chart 1 - AMD K6 250 MHz. (83 X 3) No USBSUPP installed.  Blue = Best  Red = Worst

 USBSUPP
If you require the USB patch or don't want to look at the ? in the Device Manager, this is how the installation and benchmarks look.  Note that the column for "OSR2 Only" is missing. As you might imagine, there was very little difference in benchmarks between clean installations of OSR2.  "OSR2 Only" is shown in Chart 1.

Again, the system was rebooted after each file's installation and before each benchmark.  The HD was defragged before each benchmark.

Chart 2 - 250 MHz. (83 X 3)  USBSUPP installed.

Next - Intel 3.01 BMIDE Drivers