10MHz works stably on just about every machine and card set I've tried it with. It's a very conservative OC. I would not be afraid to run 13MHz for an extended duration if it is stable for you.

maxtherabbit wrote on 2020-11-29, 16:09:

10MHz works stably on just about every machine and card set I've tried it with. It's a very conservative OC. I would not be afraid to run 13MHz for an extended duration if it is stable for you.

I appreciate the feedback. I will give the 13.3Mhz a try for awhile and see how it goes. Thanks!

jasa1063 wrote on 2020-11-29, 16:19:

maxtherabbit wrote on 2020-11-29, 16:09:

10MHz works stably on just about every machine and card set I've tried it with. It's a very conservative OC. I would not be afraid to run 13MHz for an extended duration if it is stable for you.

I appreciate the feedback. I will give the 13.3Mhz a try for awhile and see how it goes. Thanks!

After running at 13.3Mhz for a few hours I had stability problems with my Compact Flash drive. I am going to just keep it at 10Mhz

jasa1063 wrote on 2020-11-29, 23:22:

After running at 13.3Mhz for a few hours I had stability problems with my Compact Flash drive. I am going to just keep it at 10Mhz

Yes matches my experience too. I narrowed it down the the controller though, some were able to cope with a divider of 3 with the same CF card but most couldn't.

As you can see there is a point of diminishing returns

Try Doom and there won't be any diminishing returns.

At 20Mhz my Sound Blaster 16 does not work at all

Some late sound cards like ALS100 will work fine.

I made a few changes to get things running at 20Mhz with a AT bus divider of 2. I swapped out the GoldStar Prime 2C Multi IO card I was using and for one that has a UMC chipset. I have had no more issues with my Compact Flash card. I added the /CS parameter to UNISOUND to treat the CT4170 SB16 card as a non Plug-n-Play card and it initializes and works just fine.

That's interesting about the Goldstar Prime 2 card. I wonder if they're all more or less constrained to 10MHz. If the UMC can handle 20MHz that's good to know. Which UMC chip is on there?

I wonder how the Winbonds fare.

Anonymous Coward wrote on 2020-12-05, 16:51:

That's interesting about the Goldstar Prime 2 card. I wonder if they're all more or less constrained to 10MHz. If the UMC can handle 20MHz that's good to know. Which UMC chip is on there?

I wonder how the Winbonds fare.

Here is a picture of the Multi IO card I am using.

jasa1063 wrote on 2020-12-06, 01:58:

Here is a picture of the Multi IO card I am using.

Does the IDE interface go faster when you over clock the bus?

douglar wrote on 2020-12-06, 02:12:

jasa1063 wrote on 2020-12-06, 01:58:

Here is a picture of the Multi IO card I am using.

Does the IDE interface go faster when you over clock the bus?

Yes, see my first post in this thread. The results from Coretest 3.03 are for IDE disk transfer rates.

Thanks. I'll keep my eyes open for cards with those chips. Maybe they're the same ones used on the VLB version?

Anonymous Coward wrote on 2020-12-06, 02:44:

Thanks. I'll keep my eyes open for cards with those chips. Maybe they're the same ones used on the VLB version?

The answer is yes to a VLB version (although these links are currently out of stock):

https://www.ebay.com/itm/ISA-I-O-card-UMC-UM8 … G-/184416599325
https://www.amazon.com/UMC-UM82C865F-Multi-I- … uct_top?ie=UTF8

That would explain the ability to run at the 20Mhz bus speed, since VLB cards run at 33Mhz-40Mhz.

jasa1063 wrote on 2020-12-06, 02:28:

douglar wrote on 2020-12-06, 02:12:

Does the IDE interface go faster when you over clock the bus?

Yes, see my first post in this thread. The results from Coretest 3.03 are for IDE disk transfer rates.

I see that the IDE transfers are faster. Love it.

I wasn't clear enough when I asked my question. I was wondering if you increased the ATA signaling speed over the IDE cable when you raising the AT bus speed or if the crystal on the card keeps the ATA speed constant and all of the performance gains came from the faster AT bus.

douglar wrote on 2020-12-22, 17:04:

jasa1063 wrote on 2020-12-06, 02:28:

douglar wrote on 2020-12-06, 02:12:

Does the IDE interface go faster when you over clock the bus?

Yes, see my first post in this thread. The results from Coretest 3.03 are for IDE disk transfer rates.

I see that the IDE transfers are faster. Love it.

I wasn't clear enough when I asked my question. I was wondering if you increased the ATA signaling speed over the IDE cable when you raising the AT bus speed or if the crystal on the card keeps the ATA speed constant and all of the performance gains came from the faster AT bus.

I did not increase ATA signaling speed. All the performance gains are from the faster AT Bus.

I had good experience with GoldStar Prime 2 IDE controllers, CF cards and overclocking 386 hardware.
On some better 386 boards they can handle 3.6+ Mb/s @ 25MHz AT bus, no problem.
Didn't notice much difference between them and UMC ones.
Maybe the one you got is tired from the many years of service. 😀

jasa1063 wrote on 2020-12-31, 01:02:

douglar wrote on 2020-12-22, 17:04:

I wasn't clear enough when I asked my question. I was wondering if you increased the ATA signaling speed over the IDE cable when you raising the AT bus speed or if the crystal on the card keeps the ATA speed constant and all of the performance gains came from the faster AT bus.

I did not increase ATA signaling speed. All the performance gains are from the faster AT Bus.

You did increase the ATA signalling speed. You did not increase the ATA signalling speed explicitly, though. An ISA IDE interface card (it's more correct to not call it a controller, as it does not control anything itself) just passes the ISA signals to the ATA cable. If the ISA bus runs at a higher clock rate, the signalling speed on the ISA bus is increased, and thus the signalling speed on the IDE cable is increased, too.

jasa1063 wrote on 2020-12-06, 01:58:

Anonymous Coward wrote on 2020-12-05, 16:51:

That's interesting about the Goldstar Prime 2 card. I wonder if they're all more or less constrained to 10MHz. If the UMC can handle 20MHz that's good to know. Which UMC chip is on there?

I wonder how the Winbonds fare.

Here is a picture of the Multi IO card I am using.

I'm using a similar IO card in my 386DX-40, same UMC chips, similar board dimensions, just slightly different layout. So this thread got me curios to try and play with the ISA bus clock. 10 MHz worked fine, 13.3 MHz caused lock-ups in some benchmarks, yet so far 16 MHz (my board has a setting for it) seems to be working without issues; haven't tried 20 MHz yet.

In benchmarks, the difference going from 8 to 16 MHz was:
3D Bench 1.0: 15.3 -> 16.9
PC Player Bench: 3.9 -> 4.1 fps (was 4.0 at 10 MHz and 4.1 at 13.3 MHz, so this test doesn't seem to scale well)
TOPBENCH: 83 -> 98
Doom (lowest quality): 28.5 -> 30.3 fps
Quake 320x200: no measurable difference, 1.6 fps in both cases
NU System Info Disk benchmark: 2.2 -> 3.1 MB/s

I even saw about a 1% increase in speedsys memory throughput (20.6 -> 20.8 MB/s), not much but it was consistent and repeatable.

Seems like a decent boost for very little work 😀

mkarcher wrote on 2020-12-31, 10:46:

jasa1063 wrote on 2020-12-31, 01:02:

douglar wrote on 2020-12-22, 17:04:

I wasn't clear enough when I asked my question. I was wondering if you increased the ATA signaling speed over the IDE cable when you raising the AT bus speed or if the crystal on the card keeps the ATA speed constant and all of the performance gains came from the faster AT bus.

I did not increase ATA signaling speed. All the performance gains are from the faster AT Bus.

You did increase the ATA signalling speed. You did not increase the ATA signalling speed explicitly, though. An ISA IDE interface card (it's more correct to not call it a controller, as it does not control anything itself) just passes the ISA signals to the ATA cable. If the ISA bus runs at a higher clock rate, the signalling speed on the ISA bus is increased, and thus the signalling speed on the IDE cable is increased, too.

Thanks for the clarification.

You should try your tests again with a DLC-40. If the bottleneck is the CPU, then you might get better gains from overclocking ISA with the DLC.