Intel datasheet https://theretroweb.com/chip/documentation/27 … 6b098362128.pdf :

"The 82371SB PCI ISA IDE Xcelerator (PIIX3) may contain design defects or errors known as errata which may cause the
product to deviate from published specifications. Current characterized errata are available on request"

Troubleshooting
Common System Configuration Issues
Windows* 95/98 INF Update Utility
Bus Master IDE Driver for Windows 95 http://ps-2.kev009.com/ncr3xxx/pcfiles/Machin … 3271/INF_BM.PDF

rasz_pl wrote on 2023-02-03, 03:31:

Intel datasheet https://theretroweb.com/chip/documentation/27 … 6b098362128.pdf : […]
Show full quote

Intel datasheet https://theretroweb.com/chip/documentation/27 … 6b098362128.pdf :

"The 82371SB PCI ISA IDE Xcelerator (PIIX3) may contain design defects or errors known as errata which may cause the
product to deviate from published specifications. Current characterized errata are available on request"

Troubleshooting
Common System Configuration Issues
Windows* 95/98 INF Update Utility
Bus Master IDE Driver for Windows 95 http://ps-2.kev009.com/ncr3xxx/pcfiles/Machin … 3271/INF_BM.PDF

Much appreciated. By all accounts, it seems it should just work, given its functionality in W95 and DMA confirmation in Sandra. Just to triple check given Intel’s language in the first doc, I got another CF card out and installed W2K; after a few hours, was able to confirm functioning and stable DMA with that install as well using native drivers. Only W98SE gives any fuss.

For what it's worth, the fastest/best and most stable DMA I've been able to get out of my Sandisk Extreme CF card on a Packard Bell 810 motherboard (430VX/PIIX3 also) with a PMMX 233MHz was MWDMA mode 1, which is 13.3MB/s.

While I instead use Linux on that board, I found I had to explicitly set the DMA mode otherwise at every boot Linux would spend a lot of time automatically downgrading the DMA modes until it found something usable. I'm not entirely sure how Windows handles DMA configuration, but perhaps the fix might be to similarly force an explicit DMA mode in Windows (perhaps via a registry entry or similar) rather than just the regular on/off checkbox in the GUI?

mscdex wrote on 2023-02-03, 07:47:

For what it's worth, the fastest/best and most stable DMA I've been able to get out of my Sandisk Extreme CF card on a Packard Bell 810 motherboard (430VX/PIIX3 also) with a PMMX 233MHz was MWDMA mode 1, which is 13.3MB/s.

While I instead use Linux on that board, I found I had to explicitly set the DMA mode otherwise at every boot Linux would spend a lot of time automatically downgrading the DMA modes until it found something usable. I'm not entirely sure how Windows handles DMA configuration, but perhaps the fix might be to similarly force an explicit DMA mode in Windows (perhaps via a registry entry or similar) rather than just the regular on/off checkbox in the GUI?

In W2K at least, it shows DMA vs UDMA status, though it doesn’t show which mode in my experience. If forcing a mode could be an option though for 98, that would be neat.

In any case, my situation ended up being a funky CF card; got a new Verbatim 233x in direct from Amazon warehouse this morning, verified it with h2testw, then got 98SE installed and… now DMA works just fine. The install that doesn’t play nice with DMA is on a Transcend 133x.

rasz_pl wrote on 2023-02-03, 03:31:

"The 82371SB PCI ISA IDE Xcelerator (PIIX3) may contain design defects or errors known as errata which may cause the
product to deviate from published specifications. Current characterized errata are available on request"

Sounds like the answer is: "Very possible, but it depends on your board, your bios, and your storage device"