BitWrangler wrote on 2023-07-05, 15:30:

There is a survivorship bias in what we see of 440BX boards now, the earlier less flexible ones more likely got scrapped sooner. However, some of the early and crappy ones are still around. They have inflexible and weak voltage regs, don't go below 1.8 or so, and rarely do better than 115 FSB, I can't tell you if this is due to improvement in the BX silicon over the first year, or improvement in board design, because you don't get older silicon on newer boards and vice versa. Do not forget that 440BX was launched for PII @ 100Mhz FSB. With 450Mhz PIIs these boards are fine for their day and solid as you want, but try to get them running PIIIs and faster than 100 and you are struggling. Even with power regulated powerleap adapters and a bunch of workarounds these boards are a pain in the arse. The "famous" P2B had these problems in the earliest revisions, however community support did manage to alleviate some of these with known mods, voltage reg transplants etc. There are plenty of good boards released with baked in PIII support from after this, but the "all BX are golden" myth will bite you in the arse if you subscribe to it.

I like BX in the sense that it was the successor to LX and TX and that enabled you to go above 75Mhz, and they were all revolutionary in the sense that you went from Multi-Word DMA 2 (with the FX, HX, and VX) to UltraDMA 2 with the PIIX4 southbridge.

Remember that this was the advent of the CD Writer, and you could not get those things to work reliably unless you enabled UDMA mode (or were fortunate enough to be able to afford a SCSI drive).

Aside from that, what else was there at that time to compare with? ALI and VIA were not at all close to BX in terms of stability and performance (AGP fast writes, sideband support, southbridge performance/maturity), so yes, for a system with ISA, BX was the gold standard.

The P2B 1.02 and 1.03 revisions (I am not including 1.04, because Asus claimed that they never produced 1.04 -- and it is a counterfeit model) are in fact great boards... The voltage regulator is the HIP6004ACB, so you swap it to the HIP6004BCB and you get 1.3V and above ability, which isn't really a big deal, because with the 1.6V of the original regulator, you're not overvolting your Coppermine by much anyways. I did the mod, but in retrospect it was unecessary because I'm using a slotket with it's own VRM chip (Abit Slotket III).

I think it would have been more accurate for you to say that Pentium II (or newer) CPUs have a survivorship bias, because they have far less throttling ability than anything before them. Even the Pentium MMX is better in this regard.

But honestly, what else with ISA was better than BX? i810 and later needed a third-party PCI-ISA bridge if you wanted ISA, ditto with P4. Yes, technically, a P4 "slow" Win98 system is superior to a BX "slow" Win98 system if the following two conditions are met: a) You have ISA or SB-Link on your motherboard (SB-Link is even more rare with P4 boards than ISA is. I happen to have two models that have it), b) you plan to play non-speed sensitive DOS games only. I have such a P4 system (SB-Link), it's not as "easy" as you make it out to be. Yes, you get native UDMA5 support and a smoother experience in general, but there are the little things such as:

- P4 CPUs need to be de-lidded and have fresh heatsink compound applied between the core and the heatspreader (otherwise your CPU will be doing 60+ degrees with the crusty old Intel TIM acting as an insulator rather than a conductor)
- Of the two *good* sound cards that with SB-Link (YMF7xx, ESS Solo), only one of them works properly with it (the YMF7xx. though there is a single report here of success with a Solo, which I could not replicate)
- Wavetable support with this setup is difficult (but possible), because no YMF7xx cards of that era have a header (Some Solo models do, but again, good luck getting SB-Link working with it)
- P4 throttling is useless (discussed in length elsewhere on this forum... ODCM is made to sound a whole lot better than it actually is)

With regard to VRMs on BX boards, I have not seen a single BX board, no matter how generic it was, that didn't have a decent VRM design. There are plenty of Socket 7 and SS7 boards with less than ideal VRM designs, but not BX boards. Yes, the overwhelming majority of them have off-brand capacitors, but capacitors are easily replaced.

Also, you say that older BX silicon did not do 133Mhz reliably... I will test this at some point. I've got a P2B and a P2B-S, among others... Do you have any references for this claim?

bloodem wrote on 2023-07-05, 16:15:

Why do you think it's fake? The ICS9148BF-26 is the reason why SoftFSB works. This clock generator can be found on a handful of motherboards (personally, I only saw it on the Gigabyte 6BXC and the Amptron PII-3100B). I think some older revisions of the Asus P2B might also have it, but never seen one myself.

It's a "counterfeit" Asus P2B 1.04. Asus claimed it never released the 1.04. Nothing has ever been posted about its deficiencies vis-à-vis the 1.03 or 1.02, but I think one such deficiency is in fact the counterfeit or deficient clock generator. I say it's not a "real" ICS9148BF-26 because it will not do above 100Mhz, not with the jumpers, nor with SMB (but SMB works otherwise with it). I will be replacing it with a real ICS9148BF-26 I managed to order and will be reporting back.

Either way, if you downclock the Ezra-T to 150MHz (50MHz x 3) without disabling any caches, it will still be way too fast (Pentium MMX equivalent speed, IIRC). So you will need to experiment with the "L1D" and "ICD" setmul arguments (in combination with the CPU's frequency) and see which combo will give you 25 - 30 FPS in 3DBench 1.0c

Indeed. Thanks for that point of reference. Sounds to me like we need a new table. I know that my Cyrix 5x86 performs identically to a 486 DX/33 when set to 33Mhz (setmul 1).

You really don't need a 486/386 if you're only interested in games. Trust me, I have quite a few 386/486 PCs, and while they are very nice to tinker with, you are better off playing games on the Ezra-T which can be tuned for any speed point, starting with a slow 386. If you encounter any games that, for some reason, don't work with the Voodoo 3, you can always get a secondary S3 PCI video card and it will surely work with virtually anything you throw at it.

Two videocards in the same box sounds complicated... I concur, with a good VBE 2.0 compliant card (the Voodoo3 is no slouch, but it is not perfect in this regard), you don't need a 486.

mockingbird wrote on 2023-07-05, 16:53:

It's a "counterfeit" Asus P2B 1.04. Asus claimed it never released the 1.04. Nothing has ever been posted about its deficiencies vis-à-vis the 1.03 or 1.02, but I think one such deficiency is in fact the counterfeit or deficient clock generator. I say it's not a "real" ICS9148BF-26 because it will not do above 100Mhz, not with the jumpers, nor with SMB (but SMB works otherwise with it). I will be replacing it with a real ICS9148BF-26 I managed to order and will be reporting back.

Oh, I see! That's very interesting, I didn't know that. Indeed, the ICS9148BF-26 should work perfectly at FSB133. Let me know if there's any change after replacing it.

mockingbird wrote on 2023-07-05, 16:53:

Two videocards in the same box sounds complicated... I concur, with a good VBE 2.0 compliant card (the Voodoo3 is no slouch, but it is not perfect in this regard), you don't need a 486.

Complicated? Not at all... Ugly? Yes... 😅

bloodem wrote on 2023-07-05, 18:57:

Oh, I see! That's very interesting, I didn't know that. Indeed, the ICS9148BF-26 should work perfectly at FSB133. Let me know if there's any change after replacing it.

Yes! So another quite large difference with this counterfeit P2B is indeed also the counterfeit clock generator. I replaced it, and now 133Mhz FSB is working:

And here are the two different ICs. The authentic one is on the left:

Looks like an inside job to me. The people responsible for this counterfeiting project had to have had knowledge of the PCB design, and also some influence with someone over at ICS who must have sold them some rejected clock chips that worked at 100Mhz but not at 133Mhz (also of some interest is that the counterfeit is labeled "Korea" on the back and the authentic one is labeled "CHINA")... Either that or they took some other ICS part that was somewhat pin-compatible and re-labeled it. This is all conjecture, and we'll never know. The people from back then are long gone and that era is completely lost. And soon the country itself will be swallowed up by the large dictatorship nearby.

Anways, I don't know if the chip is stable at 1.3Ghz, but it does at least post. Quite happy with that.

So weird! I would argue that your board is quite special, then! 😁 Thanks for the update, @mockingbird!

mockingbird wrote on 2023-07-25, 16:21:

Looks like an inside job to me. The people responsible for this counterfeiting project had to have had knowledge of the PCB design

You can today download Asus P2B boardview files ( ASUS_Motherboard_P2B_1.10.rar ASUS_Motherboard_P2B_1.12.rar ) and they turn out to be whole Fabmaster DFT package (http://www.smtworldwide.com/pf/test-expert/) ripped from one of automatic In-Circuit-Tester machines Asus fab was using at the time. Includes bom, netlist, XY. This alone is enough to program P&P machine. Its not inconceivable someone also borrowed gerbers and did a midnight run of whole board in nearby fab. Taiwan used to have ~100 board assembly vendors and probably as much small PC motherboard manufacturers at some point. Another option is doing overnight shift in legit Asus fab using legit spare PCBs with externally sourced chips/parts that failed some stricter validation.

mockingbird wrote on 2023-07-25, 16:21:

The people from back then are long gone and that era is completely lost.

back then was merely 20 years ago 😀 people are still around, but you have language barrier to cross first.

rasz_pl wrote on 2023-07-26, 01:55:

Its not inconceivable someone also borrowed gerbers and did a midnight run of whole board in nearby fab. Taiwan used to have ~100 board assembly vendors and probably as much small PC motherboard manufacturers at some point. Another option is doing overnight shift in legit Asus fab using legit spare PCBs with externally sourced chips/parts that failed some stricter validation.

Things were different back then... Companies were different, people were different, but I suppose you could be right. After all, the Asus branded AS97127F was only a re-labeled W83781D (or so I read). Still, my gut tells me this was done from the inside.

mockingbird wrote on 2023-07-25, 16:21:

back then was merely 20 years ago 😀 people are still around, but you have language barrier to cross first.

The relationship between societal changes versus the length of time isn't always linear. No -- this is a bygone era. I knew people from back then from China who I used to to business with here. I still have their cards. They have all left and disappeared.

mockingbird wrote on 2023-07-25, 16:21:

Looks like an inside job to me. The people responsible for this counterfeiting project had to have had knowledge of the PCB design, and also some influence with someone over at ICS who must have sold them some rejected clock chips that worked at 100Mhz but not at 133Mhz (also of some interest is that the counterfeit is labeled "Korea" on the back and the authentic one is labeled "CHINA")... Either that or they took some other ICS part that was somewhat pin-compatible and re-labeled it. This is all conjecture, and we'll never know. The people from back then are long gone and that era is completely lost. And soon the country itself will be swallowed up by the large dictatorship nearby.

Interesting subject. I also have an Asus P2B rev 1.04. Same PLL chip print as the one you labelled fake. But back in 2013 when I tested this board: I could go to 133MHz FSB with the original PLL, I never replaced the PLL, it is still there.
( It was not 133MHz that was the problem, 50MHz FSB was. In the sense that 50MHz FSB worked as well, but once in this slow 50MHz mode it could not be raised again through software. A very minor issue. This issue remained even when I forced the AGP divider to 2/3 with a custom jumper mod. )
When I got the board it was defective, a MOSFET had partially detached from the PCB. I figure it had overheated when used by the previous owner.

gerwin wrote on 2023-07-31, 00:28:

Interesting subject. I also have an Asus P2B rev 1.04. Same PLL chip print as the one you labelled fake. But back in 2013 when I tested this board: I could go to 133MHz FSB with the original PLL, I never replaced the PLL, it is still there.
( It was not 133MHz that was the problem, 50MHz FSB was. In the sense that 50MHz FSB worked as well, but once in this slow 50MHz mode it could not be raised again through software. A very minor issue. This issue remained even when I forced the AGP divider to 2/3 with a custom jumper mod. )

Interesting -- I'll test that with setmul and SMB and report back. Can you elaborate on that 2/3 AGP divider mod please?

mockingbird wrote on 2023-07-31, 01:13:

Can you elaborate on that 2/3 AGP divider mod please?

Sure. I made this mod 10 years ago, So I may not remember 100% and the soldering may not be up to my current standards. But see attached image. There is a resistor just barely visible behind the capacitor, 10kOhm IIRC. The jumper pin header is for toggling the MAB12# pin on the i440BX northbridge.

The point is that most motherboards don't allow the user to manually define the AGP divider, but instead determine the AGP divider according to the bootup FSB speed. (In more restrictive boards this is tied to the CPU 66 or 100MHz FSB identification). In such a setup when you bootup at an FSB below 100MHz it will set the AGP divider to 1. If you then use software to increase the FSB speed to 100 MHz more, the AGP frequency becomes unnecessarily high.

i have a 1.02 P2B with the supposedly fake ICS 9148BF-26 part, and have to wonder, where is the idea coming from that the part is supposed to be specified for 133 mhz operation? i can't find a data sheet that would suggest this. print changes on ICs aren't unheard of either, especially when there is a two year difference in date codes. what's more, the 1.12 board has a different ICS part, 9248DF-39, that can come with yet another, totally different white print style.

auron wrote on 2023-07-31, 23:27:

i have a 1.02 P2B with the supposedly fake ICS 9148BF-26 part, and have to wonder, where is the idea coming from that the part is supposed to be specified for 133 mhz operation?

The pdf datasheet I have for "ICS9148-26" does list 133MHz as a select-able option, The P2B v1.04 motherboard silkscreen shows it too.
By the way, It is the same PLL part number as on the Gigabyte GA-6BXC motherboard.

Ok, the 50Mhz issue is not unique to you, you can rest assured. I have the exact same issue - it just hangs when it tries to apply the new settings (I also tried to explicitly set the PCI speed, did not help):

According to this post, 50mhz is useful for fast/slow 386. So I guess a reboot would be required then to switch back. Not such a big deal.

gerwin wrote on 2023-07-31, 00:28:

Interesting subject. I also have an Asus P2B rev 1.04. Same PLL chip print as the one you labelled fake. But back in 2013 when I tested this board: I could go to 133MHz FSB with the original PLL, I never replaced the PLL, it is still there.

IIRC, when I attempted 133Mhz operation with the original chip, I got an 83Mhz FSB. I might try re-soldering in the old one just for a sanity check (maybe there were tin whiskers between the pins).

gerwin wrote on 2023-07-31, 16:57:

Sure. I made this mod 10 years ago, So I may not remember 100% and the soldering may not be up to my current standards. But see attached image. There is a resistor just barely visible behind the capacitor, 10kOhm IIRC. The jumper pin header is for toggling the MAB12# pin on the i440BX northbridge.

Thanks, your pic makes it look easy enough to implement.

The point is that most motherboards don't allow the user to manually define the AGP divider, but instead determine the AGP divider according to the bootup FSB speed. (In more restrictive boards this is tied to the CPU 66 or 100MHz FSB identification). In such a setup when you bootup at an FSB below 100MHz it will set the AGP divider to 1. If you then use software to increase the FSB speed to 100 MHz more, the AGP frequency becomes unnecessarily high.

Got it... 89Mhz is one thing, I can't imagine they would like 100Mhz (well maybe some very late AGP cards). I should be ok at 66 and 75mhz AGP (100 and 112mhz FSB) then without the need for the mod.

auron wrote on 2023-07-31, 23:27:

i have a 1.02 P2B with the supposedly fake ICS 9148BF-26 part, and have to wonder, where is the idea coming from that the part is supposed to be specified for 133 mhz operation? i can't find a data sheet that would suggest this. print changes on ICs aren't unheard of either, especially when there is a two year difference in date codes. what's more, the 1.12 board has a different ICS part, 9248DF-39, that can come with yet another, totally different white print style.

I'm pretty certain your board is fine and your clock generator is fine. Revision 1.02 should be authentic.

Good news! I can confirm RayeR SMB.EXE functionality with the ICS9250CF-08 (same switch as the 9148BF-26). No need to look for unobtanium motherboards. The only caveat is that 50Mhz does not work (because the clock chip doesn't support it)... Everything else I tried worked fine.

mockingbird wrote on 2023-08-01, 22:30:

Ok, the 50Mhz issue is not unique to you, you can rest assured. I have the exact same issue - it just hangs when it tries to apply the new settings (I also tried to explicitly set the PCI speed, did not help):

Yes, that is how I remember it too. Good to know it was not just my particular board being weird.

mockingbird wrote on 2023-08-01, 22:30:

IIRC, when I attempted 133Mhz operation with the original chip, I got an 83Mhz FSB. I might try re-soldering in the old one just for a sanity check (maybe there were tin whiskers between the pins).

I understand you are curious now, but doing a desoldering/resoldering just for that? Your call of course...

mockingbird wrote on 2023-08-03, 05:56:

Good news! I can confirm RayeR SMB.EXE functionality with the ICS9250CF-08 (same switch as the 9148BF-26). No need to look for unobtanium motherboards. The only caveat is that 50Mhz does not work (because the clock chip doesn't support it)... Everything else I tried worked fine.

Noted, thanks!
There is another similar PLL that works with SMB like that: ICS9150-08. That is one of the 50..133MHz ranged chips. It is found on Shuttle HOT-661 v1.1 motherboards. These boards also have the i440BX chipset.
Where the Asus P2B v1.04 does not like 50MHz FSB that much, this Shuttle HOT-661 v1.1 does not like 133MHz.

gerwin wrote on 2023-08-03, 20:49:

There is another similar PLL that works with SMB like that: ICS9150-08. That is one of the 50..133MHz ranged chips. It is found on Shuttle HOT-661 v1.1 motherboards. These boards also have the i440BX chipset.
Where the Asus P2B v1.04 does not like 50MHz FSB that much, this Shuttle HOT-661 v1.1 does not like 133MHz.

Ah, thanks! Great to know. The 9250CF-08 is pin compatible and in fact original to the P2B-S (I replaced mine)... I am very tempted to solder the original back in (I care more about 50Mhz with the Ezra-T than I do about 133Mhz), and use this instead of the P2B. Yes, the datasheet specifies "operation not guaranteed" for 133mhz.

With regard to re-soldering in the original clockgen on the P2B, I purchased a Slot1 PIII with a 133Mhz FSB to rule out instability caused by a slotket or what have you. When that arrives I intend to re-do the testing to get a more definitive answer of whether the clock chip was indeed not genuine.

gerwin wrote on 2023-08-03, 20:49:

Noted, thanks!
There is another similar PLL that works with SMB like that: ICS9150-08. That is one of the 50..133MHz ranged chips. It is found on Shuttle HOT-661 v1.1 motherboards. These boards also have the i440BX chipset.
Where the Asus P2B v1.04 does not like 50MHz FSB that much, this Shuttle HOT-661 v1.1 does not like 133MHz.

The 9150-08 is back in on the P2B-S and 50Mhz is now functioning again (same issue with the ICS9148 with raising it back up after reducing it to 50 though). Don't upgrade the clockgen on your P2B-S people, 🤣! A couple of things:

1) The replacement they sent me is almost identical to my original except for the engraving on the chip's underside. Same week, same logo, same serial number (and no, I did not send who I purchased it from a pic of my original). What a coincidence!
2) When removing the chip, the silscreen erroneously says 9150-06. I don't think ICS ever produced such a model. Must be a misprint.

EDIT: On a separate note, I also want to confirm operation of 133Mhz FSB with my P2B 1.04 and its upgraded clockgen (ICS9148BF-26). I obtained a Slot1 P3 866 to rule out any slotket shenanigans, and it turns out it was the infineon RAM that was causing my issues. So I installed two double-sided 256mb sticks, and it sees 512mb fine and the system is Prime stable. I have not re-tested the 'counterfeit' clockgen yet. It is on my list of things to do.

I have been recently tinkering with a Nehemiah 1.2A GHz on an ASUS P2-99 rev 1.12 motherboard, and I'm finding it way more versatile than my Socket 7 setup equipped with a K6-2+/450 overall despite lacking flexibility of the Ezra-T when it comes to 486 speeds.

The motherboard has the ICS9248DF-39 clock generator that works with SoftFSB and even RayerR's SMB using the options meant for the ICS9148-26. Similar with mockingbird's experience with the ICS9250CF-08, "50 MHz FSB" doesn't work and in my case, it actually sets the FSB to 124 MHz with the PCI clock at 41.33 MHz.

Pro tip: The remaining FSB speeds with SMB can be set by entering the -0.5% parameter for the spread spectrum modulation, which will set the FS3 bit to 1. To set the bit back to 0, enter the +0.5% parameter.

I'm providing a cheat sheet if anyone is interested.

166.8/33.40:  smb /sp 10 66 33.40 +0.5%
275.0/37.50:  smb /sp 10 75 37.50 +0.5%
383.3/41.65:  smb /sp 10 83 41.65 +0.5%
4100.2/33.30: smb /sp 10 100 33.30 +0.5%
5103.0/34.30: smb /sp 10 103 34.30 +0.5%
6105.0/35.00: smb /sp 10 66 33.40 -0.5%
7110.0/36.67: smb /sp 10 83 41.65 -0.5%
8112.0/37.30: smb /sp 10 112 37.30 +0.5%
9115.0/38.33: smb /sp 10 75 37.50 -0.5%
10120.0/40.00: smb /sp 10 50 25.00 -0.5%
11124.0/31.00: smb /sp 10 133 33.30 -0.5%
12124.0/44.33: smb /sp 10 50 25.00 +0.5%
13133.3/41.65: smb /sp 10 133 33.30 +0.5% (PCI clock won't be set to 33.30 MHz)
14133.3/33.30: smb /sp 10 100 33.30 -0.5%
15140.0/35.00: smb /sp 10 103 34.30 -0.5%
16150.0/37.50: smb /sp 10 112 37.30 -0.5%

Edit: Adjusted cheat sheet, so every FSB can be set properly thru SMB.

Gmlb256 wrote on 2023-08-31, 02:32:

I have been recently tinkering with a Nehemiah 1.2A GHz on an ASUS P2-99 rev 1.12 motherboard, and I'm finding it way more versatile than my Socket 7 setup equipped with a K6-2+/450 overall despite lacking flexibility of the Ezra-T when it comes to 486 speeds.<snip>

Great post! Very nice indeed.

My Ezra-T is in fact running Prime stable (so far) @ 1.33Ghz now that I got the RAM issue sorted out, so I think I might use this with the Asus P2B 1.04 (ICS9148BF-26, which works with 50Mhz FSB) instead of the P2B-S. Heh, maybe I'll put that new clockgen back on it afterall.

EDIT: Prime95 23.x seems incompatible with the Ezra-T... The symptom is that Test 1 never finishes (the system is still responsible, but the test just won't progress). A P3 866 in this same board performs the tests fine. I have seen old posts regarding this issue but on other system configurations. Also, while 133Mhz is stable, I don't like the heat that's coming out of the northbridge so I'm going to keep it at 1Ghz afterall.

EDIT 2: Prime95 24.14 works with the Via C3. There's a floppy bootdisk image available here for download, which I will be attaching to this post.

mockingbird wrote on 2023-08-31, 03:33:

Also, while 133Mhz is stable, I don't like the heat that's coming out of the northbridge so I'm going to keep it at 1Ghz afterall.

This is one of the reasons why I set the FSB/PCI to 124/31 MHz on mine instead of 133/33 MHz. With 133 MHz FSB (aside from the issue with the AGP @ 89 MHz), I had to set the VIO jumper setting to Test for stability and that wasn't great for the northbridge's lifespan.

Gmlb256 wrote on 2023-08-31, 13:18:

This is one of the reasons why I set the FSB/PCI to 124/31 MHz on mine instead of 133/33 MHz. With 133 MHz FSB (aside from the issue with the AGP @ 89 MHz), I had to set the VIO jumper setting to Test for stability and that wasn't great for the northbridge's lifespan.

Interesting... Maybe the 440ZX was less capable in that regard. It wasn't only the heat coming out of the northbridge, the CPU @ 1.6V was an order of magnitude hotter than it was when it was running at 1Ghz. 0.15V shouldn't have that effect. This silicon just wasn't meant for that. The clockgen is also hot -- I might put a heatsink on it. See the second edit of my post above about Prime95, btw.