feipoa wrote on 2021-06-22, 07:51:

Well, there is a 1 MByte difference in size reported in the BIOS when the HDD is setup as LBA vs, LARGE. If you are using LARGE and for sake of simplicity, you've filled up the 8070 MB. Then switch to LBA, you'll be missing 1 MByte worth of data from some file or files, would you not?

Yeah, such edge case can cause data loss.
But somehow I am unmoved about it.
😁

@megatron-uk

This is great info.
But also, we know that the rules of engagement change with old hardware and OSes.
Any numbers you have with 2/3/4x86 components and something like DOS ?
I have some of the cards listed in your chart and the perf is quite different when testing with old stuff.

Could you let me know how determined that your SanDisk Ultra 50 mb/s CF supports multi-sector transfers? If possible, could you provide a screenshot? Not to be overly particular, but time and human memory can have surprising results.

To match my tests, just grab a Promise Ultra100 TX2 and plug in a relatively fast CF card to serve as the host OS. Then run the same tests you just ran copying over these large files with a stop-watch.

I think I used a Maxtor DiamondMax 10 UDMA133 platter drive. I used a Sandisk Extreme Pro UDMA7 160 MB/s CF card in the Secondary IDE port. I am wanting to check the read/write speed of this CF card on the Secondary IDE port. Using something like the Promise Ultra100, which is an order of magnitude faster, will put the bottleneck onto the MB's IDE port.

Note that if you try to put the CF in the Primary IDE port, the MB will insist on booting from the Primary port. The MB lets PCI expansion cards boot to their HDDs only if the CF card is in the secondary IDE port on the MB. Unfortunately, the BIOS doesn't have a feature to select the boot source, other than A: or C:, so this is the workaround I use. NT4 is able to see the CF card's full 32 GB. Alternately, you can use the CF card in the Primary IDE port if the target OS is located there. As noted above, I would need to get another 8 GB CF card to do that.

I haven't started using the XTIDE BIOS, but if you understand what I'm trying to accomplish, then there isn't really an option to continue using a PCI hard drive controller expansion card. I am currently using a Promise Ultra100 card, but am wanting to give it up for a second Voodoo2 card in the system. It is a big sacrifice and I want to ensure a CF card or the Maxtor DiamondMax 10 aren't too sluggish on the MB's built-in IDE controller. If it bogs the whole system down, then the trade-off isn't worth it. However, the BIOS only sees 8 GB and I will need to use the XTIDE BIOS, e.g. on a 3Com network card, to workaround this limitation.

pshipkov wrote on 2021-06-22, 08:23:

Should we try to choreograph some cross benchmark in let's say DOS and some popular 486-correct version of Windows, like 98 or 95osr2.
Can be as simple as hand stopwatching a copy of ### mb file.
We can pick boards that are close/compatible. Even hw setup varies a bit, we will still have more info than we have now. :)

Yeah, perhaps. My immediate tests are just wanting to see if there's something goofy with my MB-8433UUD and to figure out why your ATTO, Speedsys, etc benchmarks for IDE transfer rates are twice as fast as mine. If you are able to do the Promise Ultra100 test to the IDE-CF, I should have a better idea if the synthetic benchmarks are just hot air. You going IDE-CF to IDE-CF seems like it would be a lot slower than Ultra100 to IDE-CF, so I cannot make any conclusion from the results you provided. And because of my CF card stash limitation, I cannot replicate your test until I order more parts.

pshipkov wrote on 2021-06-22, 08:29:

@megatron-uk […]
Show full quote

@megatron-uk

This is great info.
But also, we know that the rules of engagement change with old hardware and OSes.
Any numbers you have with 2/3/4x86 components and something like DOS ?
I have some of the cards listed in your chart and the perf is quite different when testing with old stuff.

Yes, on old hardware the performance is nowhere near what any of those cards can do! 😁

On both of my ISA bus systems (286@20MHz and 486DLC@40MHz) I get no more than 2.6-3.0MBytes/sec with any combination of card. That's clearly a bus limitation and not a processor (nor CF card) bottleneck since the DLC is many, many times faster than the 286.

@megatron-uk
Right on.
BTW, what hw runs your Linux ?

@feipoa
I see where you are going with is.
Will check tomorrow.
Bet perf will be similar - limited by the on-board controller.
Let's see.

pshipkov wrote on 2021-06-22, 09:01:

@megatron-uk
Right on.
BTW, what hw runs your Linux ?

My main Linux desktop is the following:

i7 4790S
Asus B85M-E
32GB RAM
Nvidia GTX1050
2x Intel 480GB SSD's in RAID 1
5x Seagate Exos 8TB HDD's in RAID 5

The GPU is mainly for video encoding (dvd's/blu-rays, as well as OBS captures).
I'm running Linux Mint. It's my combination workstation/NAS, and runs off an APC SmartUPS 750VA.

feipoa wrote on 2021-06-22, 08:47:

Could you let me know how determined that your SanDisk Ultra 50 mb/s CF supports multi-sector transfers? If possible, could you provide a screenshot? Not to be overly particular, but time and human memory can have surprising results.

Some drivers support explicit control over this feature.
When testing with these CF cards and changing the state of multi-sector transfers - results vary.
It may not be the CF card itself, but the controller/driver going through different logic/code path.
So, take it for what it is.
I can probably run more tests around this with different CF cards and see if all of them show different results - if that's the case - then either all they support this mode, or it is some discrepancy on controller/driver level.
If only some of the cards show difference - that's clear indication that multi-sector is supported by some of them only.

Below are some numbers from Windows 95 osr2 and Promise Ultra133 TX2 and Promise FastTrak TX 2000 controllers.
They are probably skewed by the fact that I don't have and cannot find working Windows 95 drivers for them. Do you happen to have that stuff ?
Tried several different versions, but nothing worked.

Otherwise here are some numbers:
DOS UATA to UATA: 7126 kb/s
WIN95 UATA to UATA: 2311 kb/s <- notice the large delta compared to DOS
WIN95 UATA to UMC: 2093 kb/s
WIN95 UMC to UATA: 1632 kb/s
WIN95 UATA to IDE: 1818 kb/s
WIN95 IDE to UATA: 1422 kb/s

Where UMC denotes the custom driver and IDE means default Windows one.

At the same time they are in the ballpark of the on-board only controller tests in Windows, but in the other hand DOS uata to uata shows significant boost. So there is probably Windows driver issue in addition to the hard limits of the onboard ide.

Those results are peculiar. Slower than IDE to IDE.

WIN95 UATA to IDE: 1818 kb/s
WIN95 IDE to UATA: 1422 kb/s

I guess you don't have an Ultra100 TX2?

Yes, I have the Ultra133 TX2 driver for Win9x. I think I use it in Win98SE, so not sure how it works in Win95osr2. Attached. I know the Ultra100 TX2 has working Win95 drivers.

Do you have an NT4 setup? NT4 is pretty good about not being too fussy when switching hardware from another install.

No 100 tx2 here.
I have the driver you provided, but will try the one from your archive.
If cannot get it to work with win95, will switch to nt4.

Yes, the win95 numbers are suspicious. Will give it another spin and see how it goes.

OK with two promise fasttrak raid controllers (100 tx2 and tx2000) the drivers hang hard in win95.
Will try tomorrow with Ultra 133 tx2 tomorrow one more time and if no go will switch to nt4.

Or better - I can just run the tests in dos and be done with it.

The NT4 tests will be the most comparable. But DOS will also allow you to compare the UMC driver to non-UMC driver. All results will be of value.

Was not able to get any of the Promise controllers working with a functional driver in Windows 95.
I can try the Eware and SIS ones, but feels like more wasted time.

DOS does not see both physical drives (one attached to the ultraata controller and the second attached to the on board controller).
As I reported in previous posts in DOS copy from/to the CF card attached to the ultra ata controller goes at 7+ mb/s.
I am unable to test in DOS uata to umc ide, because the umc dos driver hangs the system when sees the uata drive.

I tried to use one of the nt4 installations I have, but they are all smp kernel based and hang on the poor 486 system.

Will have to install fresh nt4 for the last round of tests.

Nt4 feels kind of wrong for this class hardware.
Win98 may be a better fit, but I have no idea if will hit similar driver issues.
Do you know ? I never bothered to check how it handles third party raid/ultraata controllers ...

Well that certainly explains my issues with the UMC driver. I'm glad the issue isn't related to the IBM 5x86. In all my tests I have had a Promise Ultra100 PCI card installed with a hard disk or CF card connected. The UMC driver doesn't like that 'eh? The UMC IDE controller is using IRQ's 14/15, while the Promise is on IRQ 11. Any idea why the UMC driver gets hung up? It is expecting/insisting that the HDD to be the first physical drive?

I've always run NT4 on a hyper 486, even back when NT4 was still mainstream. I can see how if you didn't run it on the hardware back in the day then it may give the impression of being "wrong". But even as recent as 2004, NT4 + macho 486 was usable for me in daily use. It was far more stable than Win95.

For RAID controllers, Win98 has pretty lousy support for the higher end stuff of the time, if at all. However, I'm pretty sure I have a Fasttrak 100 setup in my Dual Tualatin system with Win98SE as one of the 4 bootable operating systems.

I just pulled the Promise Ultra133 TX2 card out of my Cyrix MII-433 system to test in the Biostar. The card works fine in Windows 95c. I am using driver version 2.00.0.42 from the Promise website. I'm not sure what driver version I posted earlier, but it isn't 2.00.0.42. Did you try *.42 already? Also, the Ultra133 TX2 card seems to function alright in NT4 even if the driver is still set as Ultra100 TX2. Bios version on the card is 2.20.0.15.

I ran NT 3.51 on a 486DX2 66 back in the day, and it's slower than NT4

feipoa wrote on 2021-06-24, 08:46:

Well that certainly explains my issues with the UMC driver. I'm glad the issue isn't related to the IBM 5x86. In all my tests I have had a Promise Ultra100 PCI card installed with a hard disk or CF card connected. The UMC driver doesn't like that 'eh? The UMC IDE controller is using IRQ's 14/15, while the Promise is on IRQ 11. Any idea why the UMC driver gets hung up? It is expecting/insisting that the HDD to be the first physical drive?

Not sure what the issue is really. Even if i specify which disk the driver to attach to /D#:SPEED flag it still hangs.

feipoa wrote on 2021-06-24, 08:46:

I've always run NT4 on a hyper 486, even back when NT4 was still mainstream. I can see how if you didn't run it on the hardware back in the day then it may give the impression of being "wrong". But even as recent as 2004, NT4 + macho 486 was usable for me in daily use. It was far more stable than Win95.

Still on 486 in 2004 - wow. 😀
Around 1998ish i was on the Pentium II band wagon already and not looking back for any 486 experience.
The software i was interested at the time was very resource demanding, so motivation was different for me.
In 1996 when NT4 was released i was still on 486 DX5 hardware. Remember trying the OS on it - it was just the wrong thing to do for my needs.
Skipped the whole Pentium 1 thing (army service and stuff) and jumped into PII/NT4 right after.
So i associate 486 with so-so experience in Win9x. These 9x versions of the OS were so clunky compared to the NT cores - but good enough for gaming and casual computing.

feipoa wrote on 2021-06-24, 08:46:

For RAID controllers, Win98 has pretty lousy support for the higher end stuff of the time, if at all. However, I'm pretty sure I have a Fasttrak 100 setup in my Dual Tualatin system with Win98SE as one of the 4 bootable operating systems.

I hear you.

feipoa wrote on 2021-06-24, 09:38:

I just pulled the Promise Ultra133 TX2 card out of my Cyrix MII-433 system to test in the Biostar. The card works fine in Windows 95c. I am using driver version 2.00.0.42 from the Promise website. I'm not sure what driver version I posted earlier, but it isn't 2.00.0.42. Did you try *.42 already? Also, the Ultra133 TX2 card seems to function alright in NT4 even if the driver is still set as Ultra100 TX2. Bios version on the card is 2.20.0.15.

Same controller here, just the driver i use is .43 minor version.
Can you confirm your installation process ?
Maybe i am doing it wrong with this 9x business ...

maxtherabbit wrote on 2021-06-24, 15:17:

I ran NT 3.51 on a 486DX2 66 back in the day, and it's slower than NT4

Yep