Figured it out!!

It was the god damn Barton CPU. Swapped Thoroughbred-B in and mobo takes FSB like no before, 146FSB and going for more.

Strange still... and yes, I have tried undervolting too, everything I did, didn't make any change in anything. Last FSB where it was stable was 136 and that's it.

dj_pirtu wrote on 2022-01-19, 10:53:

Figured it out!!

It was the god damn Barton CPU. Swapped Thoroughbred-B in and mobo takes FSB like no before, 146FSB and going for more.

Strange still... and yes, I have tried undervolting too, everything I did, didn't make any change in anything. Last FSB where it was stable was 136 and that's it.

Awesome 😀 what kind of difference in FPS do you see with overclocked FSB versus 133 ?

@cde

Just to update you sir, I'm probably going to abandon the Tualatin route. No matter what, I cannot get the PIIX4 (BX) southbridge stable with my Ultra133 TX2 card, and that's a mandatory requirement (TRIM in DOS).

I'm down to the all PCI P4 2.6 northwood build and a Yamaha card alongside a Vortex2, or a KT133A build with an ESS ISA card and the Vortex2...

I prefer the ESS card to the Yamaha, but I also prefer the Intel platform to the VIA.

I wanted to ask you, which method of throttling were you not able to get to work? Was it Throttle for DOS? I ought to test that for you. As was stated before, Abit disabled a certain ACPI function in their last BIOS revision... There is an earlier revision which keeps this feature and might allow the board to work properly, and needless to say, work without requiring BIOS multiplier changes.

EDIT:
Well turns out my instability was caused by a newly installed GeForce 2 GTS. Either the GeForce 2 GTS is a speed sensitive card (crashed less at 1Ghz on my P3 system), or I have two defective cards one with Samsung ram, one with Infineon... I don't know what's wrong with the former one, it won't post at all.

SETMUL does not work. I had the same problem as this gentleman, so it's not board specific.

Next I'll try THROTTLE.EXE... I would have preferred SETMUL because I had the BIOS modified to expose the disable ACPI option (and use APM instead), and I think THROTTLE needs ACPI.

EDIT 2:
THROTTLE.EXE works... Both in ACPI and APM mode. Next I'll try SETMUL two other brands (and models) of KT133A boards... There are initial reports of it working with KT133 in the thread, so I don't know what's going on. I tried several different CWSDPMI.EXE versions. If THROTTLE.EXE is doing what it claims to do, and not just disabling caches like its debug output says it is, then to heck with SETMUL.

cde wrote on 2022-01-19, 16:12:

dj_pirtu wrote on 2022-01-19, 10:53:

Figured it out!!

It was the god damn Barton CPU. Swapped Thoroughbred-B in and mobo takes FSB like no before, 146FSB and going for more.

Strange still... and yes, I have tried undervolting too, everything I did, didn't make any change in anything. Last FSB where it was stable was 136 and that's it.

Awesome 😀 what kind of difference in FPS do you see with overclocked FSB versus 133 ?

Well, actually it's about the same because that Thoroughbred is running lower clocks (1,6GHz) vs. previous Barton (1,7GHz). But I ordered Athlon XP 2400+ 266FSB version (Thoroughbred-B) so I can get CPU clock over 2GHz.

mockingbird wrote on 2022-01-19, 23:55:

@cde […]
Show full quote

@cde

Just to update you sir, I'm probably going to abandon the Tualatin route. No matter what, I cannot get the PIIX4 (BX) southbridge stable with my Ultra133 TX2 card, and that's a mandatory requirement (TRIM in DOS).

I'm down to the all PCI P4 2.6 northwood build and a Yamaha card alongside a Vortex2, or a KT133A build with an ESS ISA card and the Vortex2...

I prefer the ESS card to the Yamaha, but I also prefer the Intel platform to the VIA.

I wanted to ask you, which method of throttling were you not able to get to work? Was it Throttle for DOS? I ought to test that for you. As was stated before, Abit disabled a certain ACPI function in their last BIOS revision... There is an earlier revision which keeps this feature and might allow the board to work properly, and needless to say, work without requiring BIOS multiplier changes.

EDIT:
Well turns out my instability was caused by a newly installed GeForce 2 GTS. Either the GeForce 2 GTS is a speed sensitive card (crashed less at 1Ghz on my P3 system), or I have two defective cards one with Samsung ram, one with Infineon... I don't know what's wrong with the former one, it won't post at all.

SETMUL does not work. I had the same problem as this gentleman, so it's not board specific.

Next I'll try THROTTLE.EXE... I would have preferred SETMUL because I had the BIOS modified to expose the disable ACPI option (and use APM instead), and I think THROTTLE needs ACPI.

EDIT 2:
THROTTLE.EXE works... Both in ACPI and APM mode. Next I'll try SETMUL two other brands (and models) of KT133A boards... There are initial reports of it working with KT133 in the thread, so I don't know what's going on. I tried several different CWSDPMI.EXE versions. If THROTTLE.EXE is doing what it claims to do, and not just disabling caches like its debug output says it is, then to heck with SETMUL.

Little hint for the Throttle: use -c switch, that will keep L1 cache enabled and then choose from the different speed options.
Without that -c it disables L1 cache and whatever CPU you are using it's just too slow and those speed options doesn't make any sense.

mockingbird wrote on 2022-01-19, 23:55:

No matter what, I cannot get the PIIX4 (BX) southbridge stable with my Ultra133 TX2 card, and that's a mandatory requirement (TRIM in DOS).

How are you issuing TRIM to an SSD under DOS?

KT7AGuy wrote on 2022-01-20, 11:36:

mockingbird wrote on 2022-01-19, 23:55:

No matter what, I cannot get the PIIX4 (BX) southbridge stable with my Ultra133 TX2 card, and that's a mandatory requirement (TRIM in DOS).

How are you issuing TRIM to an SSD under DOS?

https://rloewelectronics.com/distribute/TRIM/1.2/

dj_pirtu wrote on 2022-01-20, 10:37:

Little hint for the Throttle: use -c switch, that will keep L1 cache enabled and then choose from the different speed options.
Without that -c it disables L1 cache and whatever CPU you are using it's just too slow and those speed options doesn't make any sense.

Thanks! But with this 2+ Ghz beast, I need the L1 disabled... According to Speedsys, even at the lowest setting with L1 enabled, it resembles a P3 1Ghz!

With combinations, I can get it very very slow.

Ok, this is a great incentive to go this route as opposed to the P3B-F system, and a better GeForce2 GTS equivalent is on its way (obtained at a bargain price, of course).

mockingbird wrote on 2022-01-19, 23:55:

Next I'll try THROTTLE.EXE... I would have preferred SETMUL because I had the BIOS modified to expose the disable ACPI option (and use APM instead), and I think THROTTLE needs ACPI.

Interesting, could you share the version of the BIOS you use?

cde wrote on 2022-01-21, 17:08:

mockingbird wrote on 2022-01-19, 23:55:

Next I'll try THROTTLE.EXE... I would have preferred SETMUL because I had the BIOS modified to expose the disable ACPI option (and use APM instead), and I think THROTTLE needs ACPI.

Interesting, could you share the version of the BIOS you use?

Sure, it's here:

https://kijibato.sakura.ne.jp/bios/index.html

Look for "kt7_3c.bin" on that page, he refers to it as "last low temperature official BIOS", that means to say, the last BIOS that reports the temperature correctly and does not have Abit's ACPI "fix". To expose enabling/disabling ACPI, you need to open the file in a bios editor and unhide the option. I may be able to upload this if you can't do it yourself.

Regarding on-the-fly multipilier changing via SETMUL, please see this post on the CPUSPD thread:
Re: CpuSpd - A Hardware Based CPU Speed Control Utility for DOS/Win9X Retro Gaming

"'Cool'n Quiet', or 'PowerNow', chipset registers are available from Via KT133A (second revision, not all KT133A) and newer VIA chipsets, but only enabled by default on KT266 or newer, because switching multiplier on the fly in KT133A was highly unstable, i did some bios modifications for my KT133A boards to support PowerNow, and i think that was the reason they did not came with the registers enabled by default hehe."

It seems as though we're out of luck in that regard.

So now I've swung back to my P4 + SBLink build, because I intend to use CpuSpd with ODCM. The P4 board also supports 3GB of SDRAM, and I found a stick type that it likes (finally!).

mockingbird wrote on 2022-01-23, 04:36:

Look for "kt7_3c.bin" on that page, he refers to it as "last low temperature official BIOS", that means to say, the last BIOS that reports the temperature correctly and does not have Abit's ACPI "fix". To expose enabling/disabling ACPI, you need to open the file in a bios editor and unhide the option. I may be able to upload this if you can't do it yourself.

Many thanks! Could you kindly share your BIOS patcher? So, this will only work with KT7A 1.1 or 1.3?

cde wrote on 2022-01-23, 11:20:

mockingbird wrote on 2022-01-23, 04:36:

Look for "kt7_3c.bin" on that page, he refers to it as "last low temperature official BIOS", that means to say, the last BIOS that reports the temperature correctly and does not have Abit's ACPI "fix". To expose enabling/disabling ACPI, you need to open the file in a bios editor and unhide the option. I may be able to upload this if you can't do it yourself.

Many thanks! Could you kindly share your BIOS patcher? So, this will only work with KT7A 1.1 or 1.3?

Yes, this is for the 1.0-1.2, he also has the 1.3 BIOS there... I don't know the intricacies so much of that board revision, though I will attach my modified version of his BIOS to unhide APM (disable ACPI). I did this with modbin63, and I can't share that here, you will have to find it yourself. As for which patcher he uses, he just ran it through BP 4.23, which adds microcode and some other fixes (but removes 8x8 pixel support, or something like that, according to the report of a user here).

Excellent! Thanks 😀

cde wrote on 2022-01-23, 21:33:

Excellent! Thanks 😀

You're welcome, you did inspire me after all in designing my build.

Just a small update:

I tried CpuSpd (issued command cpuspd m14), that locked up just like setmul did, so I think the KT7A has the Cool'n'Quiet register disabled in the BIOS.

I've PM'ed member @RoberMC to see if he can share his mod.

I also removed my northbridge heatsink (and replaced it with something more substantial - albeit fanless - good riddance!) and took a photo:

The date code makes me think it is the second revision of the KT133A... If RoberMC comes through, then we may finally have a way to change the multiplier for XP-M at the command prompt.

What did you use to replace the north bridge heatsink? I've been thinking about replacing mine but I have found something yet.

chrismeyer6 wrote on 2022-01-24, 00:35:

What did you use to replace the north bridge heatsink? I've been thinking about replacing mine but I have found something yet.

Out of my entire box of junk NB heatsinks, I only found one that was adequate and had the holes that lined up properly. It's not a big deal though, you can use that inexpensive double-sided thermal tape you buy by the square from Aliexpress with any heatsink you want, or you can put a 30 or so Ohm resistor in-line with the fan to quiet it down. This is just what I did because it was easy for me, and it was what I had on hand... That said, the aluminum is much more substantial than the stock heatsink so it should not require any fan (in my case).

KT7AGuy wrote on 2022-01-20, 11:36:

How are you issuing TRIM to an SSD under DOS?

Speaking of which, here's another small update:

The Ultra133TX2 + JMD330 (successor to the JM20330) was not stable with this setup. After some testing, I discovered that it does work ok with the Marvell PATA to SATA bridge, but write performance in ATTO is reduced (not more than 40MB/sec)... I need to do more testing, maybe with the VPD bridge, but this instability wasn't apparent with Prime95 because it had to do with the disk i/o subsystem and not the CPU. But the mode of failure that led to this discovery was two-fold: 1) inability to deltree the progra~1 (Program Files) directory, and crashing at the second stage of the Windows 98 install (after the reboot, when it's done with the device scan and starts the menu configuration and whatnot).

The write speed reduction is a small price to pay for the ability to TRIM in DOS. Reads were in the 90MB/sec range. This is with an Intel SSD530 (provisioned to 80GB or so).

I'm wondering, why not use the Marvell 88SA8052 mSATA combo? It is compatible with the VIA686B (so no external card needed) and supports TRIM.

cde wrote on 2022-01-24, 19:57:

I'm wondering, why not use the Marvell 88SA8052 mSATA combo? It is compatible with the VIA686B (so no external card needed) and supports TRIM.

The bridge chip is probably the same as on the Marvell SATA to PATA bridge, so that's just a question of preference of what you prefer, a 2.5" drive or an mSATA card with an adapter, but the crux of the matter is the southbridge IDE. I've tested all southbridges... PIIX4, ICH2, VIA... None of them are 100% reliable. the only one that works properly is the Promise.

mockingbird wrote on 2022-01-24, 22:49:

cde wrote on 2022-01-24, 19:57:

I'm wondering, why not use the Marvell 88SA8052 mSATA combo? It is compatible with the VIA686B (so no external card needed) and supports TRIM.

The bridge chip is probably the same as on the Marvell SATA to PATA bridge, so that's just a question of preference of what you prefer, a 2.5" drive or an mSATA card with an adapter, but the crux of the matter is the southbridge IDE. I've tested all southbridges... PIIX4, ICH2, VIA... None of them are 100% reliable. the only one that works properly is the Promise.

Thanks. So, I've noticed that the UDMA stability issues I was seeing with the 88SA8052-based SATA to PATA bridge went away with the mSATA version of the bridge (using either Transcend or Samsung mSATA drives); it's pretty weird.
The UDMA issues I was seeing also seemed linked to the quality of the IDE cable itself, eg. some 80-pin cables lead to errors why others work fine.