i recently got this generic sii0649 controller to test how much the rz1000 on my batman's revenge board is limiting throughput. BIOS version on it is 1914 which is the same as on a 2002 archive.org mirror i found, apart from that site i found it rather hard to find any kind of downloads or documentation for this chip.

speedsys results for a '96 quantum fireball 1280mb on this card:

now, for comparison, this is how the onboard rz1000 performs. first results are with slowest settings, second with fastest, which is Type F DMA and 64 sectors per burst enabled. from what i understand "type f" DMA is a type of 3rd party DMA that uses the chipset for DMA transfers instead of onboard logic on the card (which would be busmastering DMA), so it's like a somewhat more efficient PIO while not reducing load on the CPU. i believe earlier motherboard BIOS versions also had a prefetch option, but on mine this has been removed to prevent the rz1000 data corruption issues.

so enabling these options in BIOS gives an extra 1600 kb/s of buffered read speed, which surprised me a bit as i also ran adaptec's threadmark to test the differences in windows, where i got completely identical results (2.31 mb/s transfer rate and 58.68% cpu utilization). the rz1000 does not have any DMA checkbox in win95 device manager. but perhaps even more interesting is how hard performance drops with the new controller, buffered read speed is just a bit over half the rz1000 dma result and linear read speed also takes a hit. loading up duke3d also seemed to take noticeably longer, though i didn't time that. any theories on this result when one would expect the opposite, maybe the board/cpu are too old or the sii0649 is just not geared for PIO at all?

not exactly impressed with this controller, frankly i was hoping that it might offer busmastering DMA support in real DOS via BIOS support but that's clearly not the case at all, instead it's outperformed by the infamous rz1000 which is not a busmastering controller either. now i did not try tracking down the drivers (download on archive.org is dead) to see how it performs in windows because the issue with real DOS performance would still remain, and there's apparently no DOS drivers for this thing either, so the upgrade would be trading worse performance in DOS for better performance in windows.

i've been looking what kind of PCI IDE controllers were around in the mid-'90s that would hopefully do busmastering and be more compatible, but only came across the ~'94 pc87410, which is just another PIO controller. after that there were the promise udma66 cards but those were roughly '98 from what i understand, so there seems to have been a lack of good add-on IDE controllers in the time when busmastering became popular, but maybe people upgraded to newer chipsets instead or just went straight to SCSI to get something more reliable.