GL1zdA wrote:

It seem that there are problems with SCSI controllers if you want to use Windows 3.x:

Adaptec provides a "FastDisk" VxD/386 driver for SCSI controllers that works with 32-bit disk access as explained above. The driver included with Windows 3.1x won't even work on translated drive over 504MB. 32-bit Disk Access bypasses using Int 13h to access hard drives, thus speeding up disk I/O since it uses a protected mode driver with no thunking to real mode.

Has someone been able to test the AHA1542CF card to see it can do 10MBps? I am only able to get 6MBps. Would the spped be die to the ISA bus speed?

Iconclas wrote on 2020-01-09, 01:18:

Has someone been able to test the AHA1542CF card to see it can do 10MBps? I am only able to get 6MBps. Would the spped be die to the ISA bus speed?

ISA can't achieve 10 MB/s, at least not with the standard 8 MHz clock.
Theoretical maximum is 8 MB/s, and 6 MB/s seems very good - I think it's only possible with bus mastering.
With ISA, Fast SCSI is hardly faster than the original SCSI...

I could swear my 5x86 had DMA & EIDE support (pio4)

When did udma/ ATA33 come out?

I would think a PCI equipped 486 could run a UDMA disk controller

I would HOPE a VLB could as well but it was too early

Grzyb wrote on 2020-01-09, 02:14:

ISA can't achieve 10 MB/s, at least not with the standard 8 MHz clock.
Theoretical maximum is 8 MB/s

That's just plain wrong. ISA is 16-bit at 8.33 MHz and thus has a bandwidth of 16.7 MB/s. You are apparently not aware that we have 16-bit ISA slots since 1984.

It is important to distinguish between bandwidth or transfer rate (how many times you can switch bit values per sec) and bus speed (how fast can you transfer data).

16.7 MB/s (or 128 Mbit/s) is the bandwidth of 16 bits ISA. It doesn't mean anything. It is just number of bits x clock.

You need several (2-8) bus cycles to transfer data on ISA. So if you take the best case of 2 cycles per transfer you'll get the theoretical maximum of 8.33 MB/s which is also what most books mention as max theoretical speed of 16bit ISA clocked at 8.33 MHz. Then there are practical limits...

The maximum AHA1542CF can do is about about 2.5-3 MB/s in my testing. I heard some people achieved 3.4 MB/s with some tricks, but can't really replicate that.

rmay635703 wrote on 2020-01-09, 04:46:

I could swear my 5x86 had DMA & EIDE support (pio4)

Yes, PIO4 was standard in late 486 onboard IDE.

When did udma/ ATA33 come out?

Looks like 1996..98 - http://www.t13.org/documents/UploadedDocument … ATA-ATAPI-4.pdf

I would think a PCI equipped 486 could run a UDMA disk controller

Sure, in such a machine you can install even a UDMA/133 controller.
You can also install a SATA controller (150 MB/s), but with this, the original PCI (132 MB/s) will be the bottleneck.

Grzyb wrote on 2020-01-09, 14:09:

rmay635703 wrote on 2020-01-09, 04:46:

I could swear my 5x86 had DMA & EIDE support (pio4)

Yes, PIO4 was standard in late 486 onboard IDE.

But likely via ATA-2 specs only, which maybe wasn't the most reliable.
I'm still looking for sources, but it seems that the early PIO 4 had some reliablility issues.

"[..]ATA-3, which builds on ATA-2, adds improved reliability, especially of the faster PIO mode 4 transfers;
however, ATA-3 does not define any faster modes. [..]"
Source: Upgrading and Repairing PCs

Edit: Typos fixed.

Jo22 wrote on 2020-01-09, 15:32:

But likely via ATA-2 specs only, which maybe wasn't the most reliable.
I'm still looking for sources, but it seems that the early PIO 4 had some reliablility issues.

I recall some problems with some late 486 with SIS chipset, resolved by disabling "IDE prefetch read buffer".
Also, there was that well-known bug in CMD640.

Grzyb wrote on 2020-01-09, 14:09:

Sure, in such a machine you can install even a UDMA/133 controller.
You can also install a SATA controller (150 MB/s), but with this, the original PCI (132 MB/s) will be the bottleneck.

Check this doesn't always work. PCI specification comes with versions. Most modern SATA PCI cards expects things like MSI/MSIX style interrupts (specially the ones based in VIA chips), which ofc aren't present even in some 440BX motherboards, much less 486 and Pentium boards. There also would be voltage incompatibilities. Symptoms of the lack of these features include non-booting systems, bios/dos crashes when accessing the device, and BSoDs from Windows.

Yes. PCI IDE on 486 and early Pentium is hit and miss

PCI IDE Controller spec was standardised relatively late (mid-94) after many of these early PCI chipsets were designed prior to that.

Early PCI boards might need IDE to work in compatibility rather than native PCI mode. That means they either need a ISA auxiliary board to route IRQ14/15 from the ISA slot to the PCI card or some other crazy workarounds. Like my Socket 4 board which has BIOS settings where you have to specify which slot exactly the PCI IDE card is inserted in.

It was less an issue in early Pentium era as many of these had I/O onboard which was wired properly and users were less likely to need PCI IDE controller cards.

mpe wrote on 2020-01-09, 20:35:

Yes. PCI IDE on 486 and early Pentium is hit and miss
PCI IDE Controller spec was standardised relatively late (mid-94) after many of these early PCI chipsets were designed prior to that.

Indeed.
We were talking about 5x86, so very late for 486, hopefully with mature PCI...

But... there was this attempt at building The Ultimate 486 Machine - https://www.youtube.com/watch?v=UiTSuv29bkY
Based on Biostar MB8433UUD-A, with Am5x86, and the results were:
- PCI PATA+SATA controller with some Promise chipset - works OK with a 160 GB PATA HDD, but can't see an SSD
- PCI PATA+SATA controller with VIA VT6421A chipset - can see the 120 GB SATA SSD, but isn't stable

Grzyb wrote on 2020-01-09, 21:34:

- PCI PATA+SATA controller with some Promise chipset - works OK with a 160 GB PATA HDD, but can't see an SSD
- PCI PATA+SATA controller with VIA VT6421A chipset - can see the 120 GB SATA SSD, but isn't stable

First one: Doesn't implement certain commands to work with SSDs (even if them look as HDDs, not all the ATA commands are implemented the same).
Second one: As i said before, without MSI/MSIX style interrupts, these chips will never work stable. These interrupts are implemented in mandatory way in the last PCI versions (2.3 and 3), which are found only in P4 and some Athlon XP chipsets. Same happens with many USB2 controllers.

In any case that doesn't change what i said. SATA and UATA133 cards not always work in systems with older PCI version slots, even if their slots look mechanically compatible. You need to check these details before putting modern PCI cards in such older systems.

Stojke wrote on 2013-06-15, 12:36:

Oh i see. Well i will know once they arrive.

Can you tell me about high end controllers? And those that were special/rare etc.

Have a look at the DPT PM2021. it can be upgraded to a caching raid controller and is probably one of the highest end SCSI controllers for ISA.
The DPT PM2041W even features a wide SCSI interface and similar upgrade options.
Don´t ask me about the sense installing a wide SCSI RAID controller in an ISA based system.
I own a PM2041W but never detected any other one during the last 20 years, so these might be pretty rare.
eisapc

mpe wrote on 2020-01-09, 13:24:

It is important to distinguish between bandwidth or transfer rate (how many times you can switch bit values per sec) and bus spe […]
Show full quote

It is important to distinguish between bandwidth or transfer rate (how many times you can switch bit values per sec) and bus speed (how fast can you transfer data).

16.7 MB/s (or 128 Mbit/s) is the bandwidth of 16 bits ISA. It doesn't mean anything. It is just number of bits x clock.

You need several (2-8) bus cycles to transfer data on ISA. So if you take the best case of 2 cycles per transfer you'll get the theoretical maximum of 8.33 MB/s which is also what most books mention as max theoretical speed of 16bit ISA clocked at 8.33 MHz. Then there are practical limits...

The maximum AHA1542CF can do is about about 2.5-3 MB/s in my testing. I heard some people achieved 3.4 MB/s with some tricks, but can't really replicate that.

I have read an old Byte Magazine where the author was hot rodding an IBM AT for 12MHZ. I have read that certain cards for ISA can go quite fast such as controllers and ethernet adapters. I was talking about the DMA transfer rate of the card itself. The setup screen has a setting for 10 MB/s.

Iconclas wrote on 2020-01-20, 13:36:

The setup screen has a setting for 10 MB/s.

But where exactly?
AHA-1540CF is Fast SCSI, and Fast SCSI means 10 MB/s on the SCSI bus, not on the ISA bus.

I can see "DMA Transfer Rate" of 6.7 MB/s there - do you want to say you can select 10 MB/s as well?
Hard to believe it...

AHA-1540B has jumpers to select DMA transfer speed, and the max setting is 8 MB/s...

I'm going to chime in on the Isa scsi card.

That option is indeed for the Isa bus transfer rate, and 9/10 cases >5MB isn't going to work reliably. At least as far as emperical evidence goes.

Also, the vast majority of scsi drives (of this era) are not good enough to even surpass 5mb/s anyway. Unless you are putting new dries on adapters and shoehorning them on the narrow bus, and that's a whole 'nother kettle of fish.

Grzyb wrote on 2020-01-20, 14:46:

But where exactly? AHA-1540CF is Fast SCSI, and Fast SCSI means 10 MB/s on the SCSI bus, not on the ISA bus. […]
Show full quote

Iconclas wrote on 2020-01-20, 13:36:

The setup screen has a setting for 10 MB/s.

But where exactly?
AHA-1540CF is Fast SCSI, and Fast SCSI means 10 MB/s on the SCSI bus, not on the ISA bus.

I can see "DMA Transfer Rate" of 6.7 MB/s there - do you want to say you can select 10 MB/s as well?
Hard to believe it...

AHA-1540B has jumpers to select DMA transfer speed, and the max setting is 8 MB/s...
adaptec.png

Max selectable by jumpers is 8MB/s but you can push it to 10MB/s via a command line switch on ASPI4DOS

Also, the later revisions have it selectable in the bios to 10mb/s. (usually the PnP)

10mb/s is very possible in isa/pci setups when the isa bus has very little other traffic. Setups like pentium and newer. That said, if your system has PCI, you should use SATA and be done with it.

luckybob wrote on 2020-01-20, 20:02:

That said, if your system has PC, you should use SATA and be done with it.

Again, stop with this "recomendation". SATA is a no go in older pre-Pentium3 platforms, except maybe if you are lucky to get one of these Promise SATA controllers with the first revisions of the chip (And even these will not work in PCI slots found in 486/Pentium 90 boards and less). Newer ones need features only present in PCI versions 2.2/3 and up (Pentium 4+ PCI chipsets).

Just because a board has PCI slots doesn't mean will be compatible with all PCI cards, specially for newer SATA and USB2 controller hardware.