You already have a standard, IDE. I'd stick with that.

I'm hard pressed to recommend SCSI unless its simply to play with.
IDE SD/CF cards adapters are cheap. Not the case for SCSI equivalents.
IDE optical drives are pretty much give away prices, Not so for SCSI.
SCSI HDD's are noisy and will have come from servers so your guaranteed they have had many many hours clocked up on them.
SCSI performance is pretty much outdone by more modern drives or near instant seek times of flash media.

Only benefit from your list is larger HDD support as that would be limited to the controller card. but CF cards are cheaper, faster and easily swapped out.

Of course if you just want to play, nothing wrong with SCSI at all, but its a long dead technology with better alternatives now.

I exclusively use SCSI for my retro systems.

The main advantage is that the SCSI controller comes with own BIOS so that I can bypass any size/compatibility issues of many motherboards. I can boot even on systems with dead CMOS battery that doesn't remember hard disk settings. No need to remember storage configuration on old systems with no IDE detection, etc. I just put the card it and it boots....

I have a set of Adaptec controllers for ISA, EISA, VL-Bus and PCI and just pick the one I need to use.

As storage device, I use SD-SCSI v6 which can do over 8000 kB/s on PCI/VL-Bus/EISA and 3300 kB/S on ISA which is more than what I was to push over pre-UDMA IDE. I can also connect the device to my Macs which is another benefit.

IDE is always a mixed bag. Quality, compatibility and bugs vary mb from mb. Several of my motherboards have no IDE or buggy one. Even PCI IDE is complicated. Some mb support native IDE some only legacy and others only if you route IRQ14/15 via ISA. It is way too complicated. Then there is a problem with speed as it is motherboard dependent and no standard way how to configure speed in DOS.

By using SCSI I can avoid all of that. Just make sure my controller is using non-conflicting address/IRQ, load ASPI driver in DOS and I am settled.

but my integrated ide controllers work well and are quite modern. On a performance level, my only question was whether the SCSI controller having its CPU was more powerful for this reason.

Also could you consider my particular hardware solution instead of a generic computer? I should take a single ide controller which I then use on 4 computer classes, 386sx, 486-pod, pentiumMMX-K6III, Amiga.

tnks

AlessandroB wrote on 2020-06-08, 17:37:

but my integrated ide controllers work well and are quite modern. On a performance level, my only question was whether the SCSI controller having its CPU was more powerful for this reason.

With XT systems this might be relevant, but on 386-Pentium it's not an issue. The main reason SCSI drives were faster was that the tended to spin faster which gave them faster seek times. In fact 'intelligent' controllers tended to add latency.

Also could you consider my particular hardware solution instead of a generic computer? I should take a single ide controller which I then use on 4 computer classes, 386sx, 486-pod, pentiumMMX-K6III, Amiga.

That 'generic' idea sounds very bad tbh. DOS just uses BIOS calls so doesn't care, but any more advanced OS depends on HDD drivers, and most don't take kindly to being moved around, regardless of whether it's SCSI, IDE or something else. Expect corrupted Win9x installs for a start, and with WinNT/2k/XP you first need to uninstall the storage drivers from the system before moving and re-installing. That's bad enough for a one-off migration. Doing that every time you want to run a different system is a nightmare.

Better idea: give each system their own drive, but use an rsync-like tool to keep the contents of the drive synced between systems.

dionb wrote on 2020-06-08, 21:14:

With XT systems this might be relevant, but on 386-Pentium it's not an issue. The main reason SCSI drives were faster was that the tended to spin faster which gave them faster seek times. In fact 'intelligent' controllers tended to add latency

Before Pentium/Triton bus-master DMA wasn't a thing on IDE. Disk I/O used CPU bound PIO transfers. On the other hand, most SCSI controllers did use DMA bus-master long before that (although there were PIO SCSI controllers too). Chances are that on a non-multitasking OS like DOS this doesn't matter that much as you are unlikely to be doing disk I/O and CPU intensive task at the same time, but there is a substantial difference in the right use case.

So while I agree, I would shift the line where SCSI stopped having a significant edge (considering mainstream single HDD case)from XT era to about mid-1995.

With XT-IDE there is no issui using IDE drives.

ok, i stay with the ide drive.tnks for the hint

mpe wrote on 2020-06-08, 13:30:

IDE is always a mixed bag. Quality, compatibility and bugs vary mb from mb. Several of my motherboards have no IDE or buggy one. Even PCI IDE is complicated. Some mb support native IDE some only legacy and others only if you route IRQ14/15 via ISA. It is way too complicated. Then there is a problem with speed as it is motherboard dependent and no standard way how to configure speed in DOS.

Where did you get this from? I've been using ide drives since '91 with absolutely no problems at all. A long as the drive was a suitable size for the bios limitation and serviceable it was fine. XT-IDE takes that away in a single swoop. As for mobo speed that is totally irreverent. ISA and pci cards without their own bios are only pass troughs. The dives circuit board is were the controller resides on those mobos were the ide controller isn't on the mobo.

SCSI controllers also had limits though the the years on hdd sizes they could take.

I was referring to local-bus IDE (PCI or VLB) on 1993/1994 motherboards.

There were several problems:

1. On early PCI boards, it was widespread use of CMD640 / RZ1000 controllers which had several data corruption bugs.
2. On VL-Bus the clock of the timing (the duration of transfer pulses) is derived from the local-bus clock. So you need to configure disk data transfer rate to match the speed of your CPU and capability of your disk (or risk corruption). These become PIO modes later once bus speed stabilised.
3. Then once you pass 500MB you are in the world of BIOS/OS geometry translation problems (NORMAL, LBA, LARGE) and several IDE, BIOS or OS limits. Inability of Win3.1 to use 32bit access on disks with more than 1024 cyl, boot issues, overlay software, ...
4. Early PCI boards (those without I/O) did not support PCI native mode and required ISA paddle-boards or BIOS settings like (PCI IDE card located in slot x)

IDE is fine if you have fixed setup on a particular computer. But moving it between systems and dealing with IDE adaptors from this era could be problematic. For me, the SCSI just works...

I love the XT-IDE. it is the best thing ever for a XT. But for 486/Pentium it is 10 times slower than a local bus adaptor. So I built my own 😀 :

Just looks like a standard Promise vlb multi i/o EIDE card with CF adapter attached. Those cards a capable of taking 8gig IDE drives.

mpe wrote on 2020-06-09, 22:41:

I was referring to local-bus IDE (PCI or VLB) on 1993/1994 motherboards.

You never mentioned vlb, only PCI and ISA. VLB was a a different approach to either of those which use fixed frequencies not the cpu buss frequency like vlb did.

Wrt to vlb quite a few OEMs integrated it on the mobo along with IDE headers so no need to add an vlb expansion card at all. vlb svga video was also integrated on the mobo.

Overlay software was a solution to solve a problem at the time and worked/works quite well. I still have DDO on my 1989 386DX33s 4gig IDE hdd. The 100meg scsi hdd originally in it was dead as a dodo and scsi hdds are that common down our way outside IT. You can also put that drive in any other old system with IDE support ,including my Zenith 286/16 with internal IDE from 1990, and will boot up just fine. Just set the drive type to 2 in the bios hdd settings. Not terribly complicated at all.