Haha maybe AI was imagining things like it often does. Damn.

I got another possible lead though,

https://www.philscomputerlab.com/rudolph-r-loew-patches.html

I can’t look into this with the devices I have currently, lmk what you find. Hopefully not another dead end.

“The Rudolph Loew (RLoew) Patches (Confirmed 9x/ME Solution)
The late Rudolph Loew was the premier developer for fixing Windows 9x/ME storage bottlenecks. He created a commercial SATA Driver Patches package specifically designed to force-enable 32-bit protected-mode disk access for non-native IDE and SATA controllers under Windows 95/98/ME. Following his passing, his family released his entire library to the public.
• Where to find it: Go to the Internet Archive Rudolph Loew Collection.
• What to look for: Download the package named SATA.INF or the SATA driver patch tools. This installer patches the Windows 9x/ME I/O subsystem so it can properly communicate with newer SATA hardware like your Promise card without needing a manufacturer-specific 32-bit driver.“

Unfortunately it appears these patches are more for getting native SATA controllers that resemble ordinary IDE controllers to appear differently and be usable by ESDI.

The TX4 is a mass-storage controller. Slightly different to an IDE controller. You can check this in the SATA.INF where it shows what device classes it attaches to.

trim.exe might be usable for, well, TRIM, though - not sure if the card exposes trim support via 13h! Similarly diskdvr.exe might tell us what precise PortDriver is being used at any given time (ESDI, Compatibility, Promise, etc) which could be handy in future.

I suspect the only workable solution for wanting to break out of 13h is to use a fully fledged promise driver. The TX4 is just about different enough to the TX2 to require signifficant modification to the MPD.

Riikcakirds wrote on 2022-06-25, 14:40:

The card is a SATA300 TX4, picture below. […]
Show full quote

Solo761 wrote on 2022-06-24, 22:25:

Just to be clear, are we talking about SATA150 or SATA300 TX4? There is TX4 card in both SATA I and SATA II variant.

The card is a SATA300 TX4, picture below.

The attachment tx4by0.jpg is no longer available

This card is very fast in dos, realmode or protected mode (with emm386 loaded). Benchmarks at 83-87MB/s on a Pentium 75 with SSD attached in DOS7.1. No drivers like are loaded in config.sys and no tsrs in autoexec.bat

Win9x should do the same speeds in compatibilty mode (unless something is intercepting the calls to the bios on the card).

How are you benching Riik? I get ~28MB/s with a SanDisk 128GB SSD (says UDMA 6 in promise bios) in Speedsys for buffered and linear reads. Only verify goes above 90. P100 and K6-2/3@400, no difference.

Win9x in compatibility mode is similar... ATTO reports about 25MB/s tops with 8MB lengths with the P100, and K6-2. 32MB RAM.

I did manage to get reads near 70MB/s in ATTO with a K6-3 and 44mhz PCI. Though writes were under 5MB/s. So perhaps something interesting is going on with L2/Memory?

Perhaps did you try the TX4 pata driver (after adjusting the inf, I mean).

I'm sure i tried that back in the day, but I really can't remember if it worked.

CrustyCeleron wrote on 2026-10-06, 22:42:

How are you benching Riik? I get ~28MB/s with a SanDisk 128GB SSD (says UDMA 6 in promise bios) in Speedsys for buffered and lin […]
Show full quote

Riikcakirds wrote on 2022-06-25, 14:40:

The card is a SATA300 TX4, picture below. […]
Show full quote

Solo761 wrote on 2022-06-24, 22:25:

Just to be clear, are we talking about SATA150 or SATA300 TX4? There is TX4 card in both SATA I and SATA II variant.

The card is a SATA300 TX4, picture below.

The attachment tx4by0.jpg is no longer available

This card is very fast in dos, realmode or protected mode (with emm386 loaded). Benchmarks at 83-87MB/s on a Pentium 75 with SSD attached in DOS7.1. No drivers like are loaded in config.sys and no tsrs in autoexec.bat

Win9x should do the same speeds in compatibilty mode (unless something is intercepting the calls to the bios on the card).

How are you benching Riik? I get ~28MB/s with a SanDisk 128GB SSD (says UDMA 6 in promise bios) in Speedsys for buffered and linear reads. Only verify goes above 90. P100 and K6-2/3@400, no difference.

Win9x in compatibility mode is similar... ATTO reports about 25MB/s tops with 8MB lengths with the P100, and K6-2. 32MB RAM.

I did manage to get reads near 70MB/s in ATTO with a K6-3 and 44mhz PCI. Though writes were under 5MB/s. So perhaps something interesting is going on with L2/Memory?

Please try at command prompt on windows 98

Copy bigfilename.XXX NUL (use a file larger than 1gb) (NUL means read only doesn't write)
time it with a watch.

Atto was giving false information in my tests

CrustyCeleron wrote on 2026-10-06, 22:42:

How are you benching Riik? I get ~28MB/s with a SanDisk 128GB SSD (says UDMA 6 in promise bios) in Speedsys for buffered and linear reads. Only verify goes above 90. P100 and K6-2/3@400, no difference.

Win9x in compatibility mode is similar... ATTO reports about 25MB/s tops with 8MB lengths with the P100, and K6-2. 32MB RAM.

I did manage to get reads near 70MB/s in ATTO with a K6-3 and 44mhz PCI. Though writes were under 5MB/s. So perhaps something interesting is going on with L2/Memory?

Get a file about 200MB in size. Boot to realmode dos and type
copy filename > NUL

Then just divide size of the file, (200 in this case) by the time in seconds it takes to complete the copy.

I've attached a dos program to show the time. It just displays the time and date without asking for a prompt like the default ms-dos time command does.

Ceate a simple batchfile using edit.com in DOS. Call it bench.bat and write:

c:\Datetime.com
copy c:\filename > NUL
c:\datetime.com

save it as bench.bat and run it.
That will give you a more accurate time of how long the copy takes than timing it with a watch.

I just tried it now, on a 200MB file in DOS7.1 with only Himem.sys loaded and umbpci (no emm386) on a Pentium 100mhz.
It took 2.8 seconds = 71.4MB/s

I don't trust the Speedsys benchmark in DOS beause i think it's using the cache of the drive in the benchmark result.
The same problem the user MarmotaArmy posted about above with Atto in Win9x, it needs to be set up to not include the cache in speed results.
Any result below 16MB/s means the drive is in a PIO mode with high CPU usage.

You shoud be able to get Win9x to use the Promise controller bios like DOS does with UDMA support using Int13Ext.
MSCONFIG in Win9x and the advanced button has options to disable options on disks.