Wow Never new such converters exist. Thanks.

Anyway I’m in line with Darry. So next stop is to compare results with a 15k

darry wrote on 2024-03-28, 18:14:

Some results from an ISA only (probably 8MHz ISA) Pentium machine a 1542CF and a 10K drive.

Will this work? Adaptec 1542CF and Cheetah 10k with SCA-2 80-pin hot-swap connector

Thanks. Seems to make no difference in performance. Damn. I already ordered now 😒

Marco wrote on 2024-03-28, 18:53:

darry wrote on 2024-03-28, 18:14:

Some results from an ISA only (probably 8MHz ISA) Pentium machine a 1542CF and a 10K drive.

Will this work? Adaptec 1542CF and Cheetah 10k with SCA-2 80-pin hot-swap connector

Thanks. Seems to make no difference in performance. Damn. I already ordered now 😒

You still might get lucky as your ISA is running faster than the one on that Pentium.

Re: Will this work? Adaptec 1542CF and Cheetah 10k with SCA-2 80-pin hot-swap connector

Swiego wrote on 2020-05-02, 19:52:

I realized I can up the adapter speed to 10MB/s (actually makes a 10% difference) and this gets me to a reliable 2.4MB/s with the Atlas 10k.4...This is in DOS with aspi4dos.sys; it drops to 1.6MB/s without the driver, and is roughly the same in every W98 benchmark I’ve tried.

so Adaptec 1542CF tops out at 2.4MB/s, there goes my "Adaptec 1542CF is also much faster than 1-2MB/s seen here", its exactly 400KB/s much faster 😉

I think the 1542CF is holding you back. Here's a video that benchmarks bluescsi on several macs against the original hdd that shipped with the machines and the bluescsi is competitive, but not always the winner, on every machine.

https://www.youtube.com/watch?v=AR2boVHnbLk

But that’s the point. If both my tests will fail (show slower performance than via IDE) then my summary will be the same as from user kingcake. SCSI is - in some configs at least - slower than isa even on old hardware. Quite often heard feelings like: faster, more responsive etc is than partly proven to be not the case .

Some specific isa scsi controllers can be faster than ide ones in certain scenarios. The opposite is true as well.
Especially if CF cards are used as attached storage devices.
This is well established fact.

Your observations are correct within the context you present them.
Nothing to see here. : )

Thanks. One addon: as I wanted to max out performance on my pc I would have been somehow satisfied with a reduction of cpu usage. But as observed and shown via threadmark it’s identical also

this class hardware does not have the performance specifications for multithreaded workflows.
you can stage some scenario to disproove the statement, but that wont be a practical example.
scsi makes the promise of lowered system requirements for concurrent local storage i/o.
this is largely not the case with single user gaming PCs and workstations.

all ide ontrollers will have more or less the same performance for your setup.
scsi's - they are more volatile. you will have to test several different ones to find the best match.

but in short - your system is the limiting factor here, not the controllers, so spending more time on them wont move the needle.

yes I get the point. Thats why I hoped that at least the Adaptec Threadmark bench would simulate such scenarios and might show differences...

Anyway, what still gives me some "sleepless nights" is that DMA speed settings in the AHA Bios. Up tp 10Mb/s. As I could read everywhere no one was ever successful with anything higher 5,7MB/s settings. I have my DMA clock speed at 6,85MHz but even here - no difference. Yes maybe these two facts dont really have to do something with each other 😀

What you mean by nobody is able to run aha-1542 at 10mb/s ? Here is somebody. : )
Others around here do it as well.
It is a common thing, probably thats why it is rarely discussed, unless i miss your point.
I had somewhere screenshots posted in the past showing what aha-1542 produces when in 10mb/s configuration.
Will try to find them later. On the phone atm.

I've got a 386sx-40. Getting ~3.3 MB/s on a AHA-1542CF and ZuluSCSI v1.1.

Some things to check are
- Setting the max possible ISA speed in the AHA bios (keep increasing and running the test in the AHA bios until you get errors). I have mine set to 6.7 MB/s.
- Setting "yes" to the sync setting in AHA bios (ZuluSCSI would not auto-negotiate, synchronous transfers have to be initiated by the controller)
- Shadowing the AHA-1542CF rom in your computer's bios settings (pretty large gain here for me)
- Make sure your BlueSCSI is set up to allow SCSI-2 and synchronous transfers at 10 MB/s

Also ~3-4 MB/s is close to the real-world maximum that an ISA SCSI card can do (16-bit ISA bus IS the limitation here). Up to 8 MB/s might be possible with 0ws everywhere and a fast CPU.

Some good discussion and commentary here.

ahyeadude wrote on 2024-03-30, 01:40:

Also ~3-4 MB/s is close to the real-world maximum that an ISA SCSI card can do (16-bit ISA bus IS the limitation here). Up to 8 MB/s might be possible with 0ws everywhere and a fast CPU.

Some good discussion and commentary here.

Fascinating and arcane knowledge. The advent of PCI bus mastering really spoiled us .

Oh wow thanks. I really could find that statement (5,7mb/s) is the limit quite often in the net indeed.
I did use the same settings as ahyeadude recommended. Thanks.

Why the h… can’t I pass any bios tests while selecting >5,7mb. Thanks for other sleepless nights now 😀
Maybe it’s simply based on the max bandwidth your pc can provide on Isa (which is mixture of cpu and Isa clock speed mostly). Hm.

OS2 Museum carries a lot of credibility, but things evolve, new "discoveries" are made, so some of the information previously available becomes obsolete.

Two screenshots from SpeedSys.

Adaptec AHA-1542CF at 10Mb/s + Acard adapter (386DX CPU 1x50MHz, 25MHz ISA bus). Look on the right side.

and the same SCSI controller, same speed setting, with ZuluSCSI adapter (IBM BL3 CPU 2x50MHz, 25MHz ISA bus).

Puh ah i am amazed. Impressive on the faster pc.
I also reach 3500kb linear read with my maxtor at 13,75MHz ISA on my 386sx/27. wait. Let me double check that message again 😀 ok yes correct

But again I wonder why you with nearly double isa clock plus double cpu clock do get only the same bandwidth as I do. Why is that ? (Picture upper left)

What driver are you using for the SCSI controller in Windows 95?

@Marco

I think we switched the topic to IDE controllers now, right ?

The same PC from above but with Promise EIDE Pro.

DTK PT-227 on another, most pimped, 386 machine.

SpeedSys reports 486 class CPU, but it is confused.

Both tested with CF cards. Kind of touching the possible upper bound of ISA/IDE/PIO.
Latest models IDE mechanical HDDs with cache and everything can saturate these puny old ISA controllers very well, but there is a lot of volatility with mech HDDs, so i stopped testing with them.
In reality, good quality present day CF cards are much faster than old HDDs just because of seek time.

To your question.
Partially answered with the mech HDDs note, but also - things don't scale linearly.
Towards the upper end more and more MHz'es are needed for smaller and smaller improvement.

---

@douglar
If your question was towards me - no proprietary software drivers needed for all ISA SCSI controllers i touched so far, including the AHA-1542 noted above.

---

Maybe i should start a thread for ISA IDE similar to the VLB IDE one.
Have quite a bit of info that can be dumped for public consumption.