PARKE wrote on 2025-02-05, 16:30:

Mockingbird, which version is your Abit slotket ?

It's the Abit Slotket III... Must have bought it 25 years ago. I am not using it. When I refer to my 'new' slotket, I am specifically referring to a completely different model from a different company. I have not disclosed which in particular for an aforementioned reason.

So far so good with the new slotket I picked up. Passed 640x480x16 for around 24 hours, and now has been going two hours without issue at 1024x768x32 with cpu at 1.2Ghz.

Just to clarify - earlier I stated that the new slotket was not stable - but that was on the QDI board. I am now testing on the board I intend to use for this build, the Gigabyte GA-6BXC (and so far, it seems stable, but I won't know for certain until tomorrow)... So that's another thing to consider -- board/slotket variability where it relates to overall system stability. So far, I've found the P3B-F to be the most picky in this regard. I should also state that this is the newer GA-6BXC with the later voltage regulator that allows for lower voltages. Don't put an Ezra or Nehemiah on an early GA-6BXC unless you first swap out the voltage regulator or you'll be overvolting the CPU.

Reason for this system is CMOV capability, which is absent from Ezra and Ezra-T. Also, Ezra-T's write combining is broken (I was going to use that despite the missing CMOV, until I read about the WC errata which in my view vastly detracts from the value of this model). Nehemiah seems ideal for retro systems. At 1.2Ghz, you get around something of the speed of a Katmai 500 (some people claim faster, some people claim slower), and the "organic" throttling capabilities (that means to say - FSB/Multi/CPU Features and not other chicanery) are too numerous to list. Hopefully this survives 24 hours, and then I'll test again for another 24 hours with the graphics card I intend to use (testing on 64-bit GeForce 4 MX420, intend to use 128-bit MX440) and then I'll call it good to go. This system has been years in the making.

mockingbird wrote on 2025-02-05, 15:41:

I don't necessarily agree with this, but let's see.

You... don't agree with how the graphics rendering pipeline works? 😀

mockingbird wrote on 2025-02-05, 15:41:

Don't put an Ezra or Nehemiah on an early GA-6BXC unless you first swap out the voltage regulator or you'll be overvolting the CPU

It will not be overvolted, unless you use a slotket that allows you to override the voltage request signal and you specifically configure the slotket to request a higher voltage.
Otherwise, it will just fail to post and the CPU will receive no voltage (if the buck pwm controller is not able to provide the voltage that the CPU requested).

Getting back, it's now been 13 hours since I launched the 3DMark2000 loop and it's still going strong. Will leave it for another 2 - 3 hours, but I don't expect it to fail at this point. 😀

bloodem wrote on 2025-02-05, 18:11:

You... don't agree with how the graphics rendering pipeline works? 😀

I am only following Sherlock Holmes' adage of accusing the improbable once the impossible has been excluded. I think it will turn out that the slotket was the issue after all, so that argument is pointless for now.

Getting back, it's now been 13 hours since I launched the 3DMark2000 loop and it's still going strong. Will leave it for another 2 - 3 hours, but I don't expect it to fail at this point. 😀

Great, thank you! Things are humming along here fine too... I've been running 1024x768x32bpp for three hours and it's stable so far... I won't call it until tomorrow.

I don't want to sound disputatious, but here's a report of the P3B-F not working with the MSI Slotket.
You say yours does -- I wonder what was up with his setup.

I really wish someone would design a new slotket PCB... I think I have a source for inexpensive PPGA370 sockets... The PCBs couldn't be more than $5 a piece (maybe $10/pcb for ENIG) and the parts maybe another $5... There really is a shortage of good, compatible slotkets for a reasonable price.

bloodem wrote on 2025-02-05, 18:11:

You... don't agree with how the graphics rendering pipeline works? 😀

UPDATE: It crashed to desktop at 1.2Ghz (640x480x16 was done at 900mhz) 1024x768x32bpp after around 3 hours. I went to take a shower, came back, and I see the green desktop screen.

I'm going to reboot and re-do the test at 900Mhz.

So either there is something magical about the 370SPC and the MS6905 2.3, or perhaps it is something else.

Am I correct in characterizing your opinion as that of leaning towards my GeForce 4 MX cards being the issue?

If the 900Mhz test fails, I'm going to put in the GeForce 3 to match your specs.

If I'm forced to give up, I'll test with the new Nehemiah that is due to arrive to me.

UPDATE: I stopped the test yesterday after almost 17 hours. As expected, everything went smoothly, no crashes to report.

mockingbird wrote on 2025-02-05, 18:24:

I don't want to sound disputatious, but here's a report of the P3B-F not working with the MSI Slotket.
You say yours does -- I wonder what was up with his setup.

Hard to say what his issue was. It might've been a BIOS related problem (my particular P3B-F refused to post - stuck at code C1 - even with a Coppermine when using an older BIOS from late 99, which is a bit unusual compared to other 440BX boards I've tested).
Having said that, I have yet to see a 440BX board that does not work with the Nehemiah. Some, with older BIOSes, don't even recognize it properly and yet they work just fine in day-to-day usage.
My particular board is running with a modified BIOS version (p3bf1084_20240219) from this thread and it's been a flawless experience so far (see the screenshot below). One of the listed changes is specifically mentioning the VIA C3 CPUs: "experimental support for Via C3/Cyrix III processors"

mockingbird wrote on 2025-02-05, 19:30:

UPDATE: It crashed to desktop at 1.2Ghz (640x480x16 was done at 900mhz) 1024x768x32bpp after around 3 hours. I went to take a shower, came back, and I see the green desktop screen.

I'm going to reboot and re-do the test at 900Mhz.

So either there is something magical about the 370SPC and the MS6905 2.3, or perhaps it is something else.

You are focusing too much on the slotkets, CPU, when all of the evidence points in a completely different direction. Again, increasing the resolution and color depth will decrease the CPU load and increase the GPU load. This is not debatable, you can treat it as a mathematical certainty. 😀
Now, the fact that it seems to work fine with other CPUs might only be an effect of a different bug, not the actual cause.

mockingbird wrote on 2025-02-05, 18:24:

Am I correct in characterizing your opinion as that of leaning towards my GeForce 4 MX cards being the issue?

If the 900Mhz test fails, I'm going to put in the GeForce 3 to match your specs.

If I'm forced to give up, I'll test with the new Nehemiah that is due to arrive to me.

Either the GeForce 4 MX, the drivers or even a quirk with the motherboard itself. I would test on a clean Windows install with the GeForce 3 and Detonator 7.76. If it's a GeForce 3 Ti 200/500, you will have to force the driver installation, but it will work fine. If you have a different, known working & stable motherboard, I would also replace it as well.

bloodem wrote on 2025-02-02, 13:12:

I'm thinking of removing it because, from my experience, this mod is not needed. Here are a few pics:

I usually don't rely on my experience and look at primary sources.
Obviously in some cases this capacitor is needed
https://elektrotanya.com/jetway_j-603tcf_rev_ … f/download.html

shevalier wrote on 2025-02-06, 07:16:

I usually don't rely on my experience and look at primary sources.
Obviously in some cases this capacitor is needed

Thank you for the schematic, it makes sense now. 😀
However, you misunderstood what I said. I was talking about my experience with the MSI slotkets: the vast majority of these slotkets don't have this factory mod and yet they do work just fine.
Also, my initial assumption was that this was a user mod, not a factory mod, so I thought that maybe this was some unofficial improvement that I was not aware of.

bloodem wrote on 2025-02-06, 07:38:

Also, my initial assumption was that this was a user mod, not a factory mod, so I thought that maybe this was some unofficial improvement that I was not aware of.

I don't think it's a custom mod.
Option 1, possible.
Someone wanted to improve Vtt filtering and missed a few pins.
Option 2, unlikely.
Someone has incredible skills in debugging motherboard issues and they found a problem with slotket instability.
Well, the probability of this option is close to 0.

A quick look at the schematics shows that the Intel Slot BX and i815 motherboards don't have it.
The Via chipsets for Tualatin/Coper C370 do.

This may or may not be a factory modification.
Looks like a factory modification for a specific MSI motherboard.

mockingbird wrote on 2025-02-03, 16:30:

bloodem wrote on 2025-02-03, 13:32:

Strange. I absolutely ran 3DMark 2000 stability tests for many hours when previously testing various Nehemiah overclock frequencies/voltages and, outside of general instability when exceeding the maximum frequency that a chip could handle, they ran fine.
Anyway, I have a P3B-F board with a Nehemiah right next to me, so I'll start a test now and report back. 😀

Thanks, much appreciated. I have another Nehemiah on its way in the mail and I'm picking up a different model slotket today so I'll see if that helps... But I am leaning towards the CPU being defective since my Ezra-T is completely stable.

There is nothing wrong with your CPU. Instability is to be expected if your motherboard does not officially support the Nemiah core - this is even more relevant for slot 1 boards. Some slotkets improve stability, others make it worse.

For best results with a C3 nemiah, use a socket 370 VIA 694T chipset motherboard. They usually have an ISA slot as well so you can keep using that ISA sound card for DOS compatibility.

Socket3 wrote on 2025-02-06, 15:33:

There is nothing wrong with your CPU. Instability is to be expected if your motherboard does not officially support the Nemiah core - this is even more relevant for slot 1 boards. Some slotkets improve stability, others make it worse.

For best results with a C3 nemiah, use a socket 370 VIA 694T chipset motherboard. They usually have an ISA slot as well so you can keep using that ISA sound card for DOS compatibility.

Excellent point. I concur. I disagree with @bloodem's conclusions. I am happy for him that his setups are stable, but I am not convinced that his type of setup would be universally stable for everyone else.

That said, at 900Mhz, I am a few hours short of a 24 hour run of 3DMark2000 at 1024x768x32bpp. This failed on all other slotkets except for the one I am using now (which fails at 1.2Ghz).

I'll be out for most of the day, but if it is still running when I return, I'll call it good and settle for Nehemiah at 900Mhz (9x100). That is good enough for me for this type of build (and I'll stick with the GF4 MX440 and Detonator 45.23).

I think Samuel or Ezra (not Ezra-T, because of the aforementioned CPU errata) is a much easier path than Nehemiah. Ezra-T worked without issue on almost everything I tested it on. Nehemiah worked on almost nothing I tested it on.

Off the top of my head, the Nehemiah is 100% stable on the following 440BX motherboards that I personally tested, using 3 different slotkets (two of them in the "no-name" category):

- Gigabyte GA-6BXC rev 1.9 (with VRM mod)
- Gigabyte GA-6BXC rev 2.0
- MSI MS-6163 Ver 1.0
- MSI MS-6163 Ver 2.0
- Asus P3B-F V 1.03
- Amptron PII-3100B (with VRM mod / doesn't even recognize the CPU, but works fine).

Mind you, I didn't just run a few tests on these boards and deemed the platforms stable. I actually played with them for weeks, even months in some cases. I even finished entire games on some of them (such as Max Payne, on the Amptron - which was surprisingly playable with the overclocked Nehemiah).

My conclusion is based on what you've told me, nothing points to the CPU as being the actual root cause (though, it might play a role in triggering a different bug/fault), since you mentioned that 3DMark2000 (sometimes) fails at higher resolutions, but appears to run just fine at lower resolutions (when the CPU load is higher).

mockingbird wrote on 2025-02-06, 16:34:

Excellent point. I concur.

The excellent point is due to confirmation bias. 😀 If you think about it, the point barely stands. For example, let me ask you this: you mentioned that the Ezra-T works fine; do you think that CPU, which launched in the Summer of 2002, is "officially supported" by the 6BXC?
Last but not least, if it's "normal" for the Nehemiah to be unstable on "motherboards that don't officially support it", why does it work just fine for me and many others, on so many different boards?

Edit: to be clear, I'm not some Nehemiah fanboy 😀 I couldn't give a rat's ass if it's unstable or not, but it has always been stable for me so... it would be weird to claim otherwise.

bloodem wrote on 2025-02-06, 17:05:

Off the top of my head, the Nehemiah is 100% stable on the following 440BX motherboards that I personally tested, using 3 different slotkets (two of them in the "no-name" category):
<snip>
Mind you, I didn't just run a few tests on these boards and deemed the platforms stable. I actually played with them for weeks, even months in some cases. I even finished entire games on some of them (such as Max Payne, on the Amptron - which was surprisingly playable with the overclocked Nehemiah).

I appreciate that fact, but this is a unique kind of crash which underlines some inherent stability. Keep in mind, in most cases, Nehemiah also "worked" on all my boards (except the P3B-F -- but I don't trust my P3B-F, I am not concluding anything about it at the moment until I can confirm it is stable with other CPUs), but crashed only after several hours of 3DMark2000. That might be considered "stable" for 99% of folks, but not me. You would never see this instability except maybe if you were playing several hours straight of some intensive 3D game.

My conclusion is based on what you've told me, nothing points to the CPU as being the actual root cause (though, it might play a role in triggering a different bug/fault), since you mentioned that 3DMark2000 (sometimes) fails at higher resolutions, but appears to run just fine at lower resolutions (when the CPU load is higher).

I'm not done testing yet. You missed the part where I say I also clocked the CPU to 1.2Ghz. For your information, I have been running at 1024x768x32bpp @ 900Mhz for more than 24 hours, and it is still running fine.

Now what I'm going to do for you is I'm going to solder on jumper blocks to my slotket and bump the voltage a little and test again at 1.2Ghz. If that passes a 24 hour 3DMark, then I will blame all my other slotkets (note: I tried this on my Abit Slotket III - did not help one bit, even at 900mhz).

The excellent point is due to confirmation bias. 😀 If you think about it, the point barely stands. For example, let me ask you this: you mentioned that the Ezra-T works fine; do you think that CPU, which launched in the Summer of 2002, is "officially supported" by the 6BXC?
Last but not least, if it's "normal" for the Nehemiah to be unstable on "motherboards that don't officially support it", why does it work just fine for me and many others, on so many different boards?

Edit: to be clear, I'm not some Nehemiah fanboy 😀 I couldn't give a rat's ass if it's unstable or not, but it has always been stable for me so... it would be weird to claim otherwise.

We'll see. There are still more tests to be run. Higher voltage, different CPU... We'll see. I'll report back Saturday night.

mockingbird wrote on 2025-02-06, 22:58:

I appreciate that fact, but this is a unique kind of crash which underlines some inherent stability. Keep in mind, in most cases, Nehemiah also "worked" on all my boards (except the P3B-F -- but I don't trust my P3B-F, I am not concluding anything about it at the moment until I can confirm it is stable with other CPUs), but crashed only after several hours of 3DMark2000. That might be considered "stable" for 99% of folks, but not me. You would never see this instability except maybe if you were playing several hours straight of some intensive 3D game.

I'm starting to wonder: you mentioned that you were using Detonator 45.23 for your tests, and I just realized that this driver in particular is known to use SSE instructions (at least for some of the newer GPUs?). Well, it just so happens that the Nehemiah does have SSE instructions, while the Ezra-T does not (that one has AMD's 3DNow! instructions). Maybe the driver bugs out in certain conditions with the Nehemiah's SSE implementation. Anyway, I'm just throwing ideas, I have no evidence to back this up. Depending on your findings, I might do a bit of digging myself, because at this point I'm just curious as to what could cause the instability you're seeing. One thing I know for sure, though: I've never tested the Nehemiah with drivers newer than Detonator 7.76 or with GPUs newer than the GeForce 3 (never really saw a point for these types of builds, since these older drivers are always faster and perfectly stable).

mockingbird wrote on 2025-02-06, 22:58:

We'll see. There are still more tests to be run. Higher voltage, different CPU... We'll see. I'll report back Saturday night.

It would be very strange if higher voltage had any impact because it would not explain why your CPU seems to work fine at lower resolutions, when the CPU load is higher. Anyway, keep us posted. 😀

bloodem wrote on 2025-02-07, 07:23:

I'm starting to wonder: you mentioned that you were using Detonator 45.23 for your tests, and I just realized that this driver in particular is known to use SSE instructions (at least for some of the newer GPUs?). Well, it just so happens that the Nehemiah does have SSE instructions, while the Ezra-T does not (that one has AMD's 3DNow! instructions). Maybe the driver bugs out in certain conditions with the Nehemiah's SSE implementation. Anyway, I'm just throwing ideas, I have no evidence to back this up. Depending on your findings, I might do a bit of digging myself, because at this point I'm just curious as to what could cause the instability you're seeing. One thing I know for sure, though: I've never tested the Nehemiah with drivers newer than Detonator 7.76 or with GPUs newer than the GeForce 3 (never really saw a point for these types of builds, since these older drivers are always faster and perfectly stable).
<snip>
It would be very strange if higher voltage had any impact because it would not explain why your CPU seems to work fine at lower resolutions, when the CPU load is higher. Anyway, keep us posted. 😀

Good thought you had there. I'll give you a teaser -- something I jumpered on my slotket seems to have made it completely stable at 1.2ghz (in addition to bumping it up by 0.05V to 1.5V I also installed missing jumper blocks and installed the "Cyrix" jumper and changed the FSB from "auto" to 100Mhz). It hasn't been 24 hours yet, but I'll continue to let it run and report back tomorrow night.

Ok, here's the final word: I've had my Nehemiah running 3DMark2000 at 1024x768x32bpp with 32-bit textures for around 48 hours on my new slotket at 1.2Ghz (12 x 100) and it is still going fine. My conclusion is that the Abit Slotket III is stable for Ezra-T but not for Nehemiah.

I also have another slotket (different model than the aforementioned two) in the mail and I'll run some further tests with it once it arrives.

Thanks @bloodem for your help.

mockingbird wrote on 2025-02-09, 00:01:

Ok, here's the final word: I've had my Nehemiah running 3DMark2000 at 1024x768x32bpp with 32-bit textures for around 48 hours on my new slotket at 1.2Ghz (12 x 100) and it is still going fine. My conclusion is that the Abit Slotket III is stable for Ezra-T but not for Nehemiah.

I also have another slotket (different model than the aforementioned two) in the mail and I'll run some further tests with it once it arrives.

Thanks @bloodem for your help.

Well, that's so weird and doesn't make any sense to me, but glad you finally have stable build. 😀

bloodem wrote on 2025-02-10, 06:24:

Well, that's so weird and doesn't make any sense to me, but glad you finally have stable build. 😀

Yes, but my MX440-8X without a heatsink did not survive 24 hours 3DMark like my MX440-64 bit did, so I'll re-test with my ordinary MX440. It seems like I had a combination of factors which may have also been the GPU. Maybe it was overheating after some time. I'll report back.

mockingbird wrote on 2025-02-10, 14:07:

Yes, but my MX440-8X without a heatsink did not survive 24 hours 3DMark like my MX440-64 bit did, so I'll re-test with my ordinary MX440. It seems like I had a combination of factors which may have also been the GPU. Maybe it was overheating after some time. I'll report back.

Yeah, that would make much more sense! Sorry about your MX440. 😀
In the meantime, by pure chance, I snagged another slotket for $8 including shipping (so I obviously couldn't let the opportunity pass 😁 ). It's another no-name of sorts, the "AA370TS".

For now I've tested it with one of the Tualatin 1.4 GHz CPUs that I modded, and it has been running fine for the past 3 hours in 3DMark and Prime95. Will definitely also test the Nehemiah, though. 😀

bloodem wrote on 2025-02-10, 17:15:

Yeah, that would make much more sense! Sorry about your MX440. 😀

It is now testing with an ordinary Mx440 (128 bit)... I have the MX440-8X in the P3B-F now running a different test.

In the meantime, by pure chance, I snagged another slotket for $8 including shipping (so I obviously couldn't let the opportunity pass 😁 ). It's another no-name of sorts, the "AA370TS".

For now I've tested it with one of the Tualatin 1.4 GHz CPUs that I modded, and it has been running fine for the past 3 hours in 3DMark and Prime95. Will definitely also test the Nehemiah, though. 😀

Yes, please do let me know, I just purchased that exact slotket (AA370TS - paid 10 Euro+shipping). Can you go into detail on that Tualatin mod you have? I might just attempt it with one of my own. Which glue did you use?

mockingbird wrote on 2025-02-10, 17:42:

Yes, please do let me know, I just purchased that exact slotket (AA370TS - paid 10 Euro+shipping).

Will do!

mockingbird wrote on 2025-02-10, 17:42:

Can you go into detail on that Tualatin mod you have? I might just attempt it with one of my own. Which glue did you use?

It's the usual mod, which is needed in order to make the Tualatin CPU compatible with the older FC-PGA socket (i.e.: 440 BX boards & co): You can find the guide here: https://mrufer.ch/pc/tualatin2_e.html
Basically, you need to remove (or isolate) 3 pins and connect the PWRGD pin to the VttPWRGD pin through a 1k ohm resistor (since I have A LOT of P3-S Tualatin CPUs, I chose to remove the pins - not proud of it, I don't usually like doing this to retro parts but, in my defence, I was having a lazy day when I did it 😅 ).

WARNING: If you can't manually force the correct voltage (1.45 - 1.5V) through either the slotket or the motherboard, you absolutely need to also connect VID4/VID25m to VSS (ground), otherwise you will fry the CPU, since it will default to requesting ~2V from the motherboard.
WARNING 2: you need pretty decent soldering skills and equipment (including a very, very, very fine soldering tip, 0.2mm or similar).

I used transparent Bison Epoxy, which is my favorite for this type of stuff: https://www.amazon.de/-/en/Bison-Epoxy-Adhesi … e/dp/B089FNHGZB