Hello All,

I know this thread is getting old, but I have made some changes over the past month and wanted to do a final update in case someone else is looking to build something similar.

-I re-did the OS installations. I found that I was often times re-doing the same stuff (game installations, configs) in DOS/Windows 3.1 as I was in Windows Me. I got rid of both of these OSes and installed Windows 98. MS-DOS mode works well enough for most games that don't like to be run within Windows.

-I got rid of all the multi-boot complexity. I have Linux installed to another CF card entirely. It is easy enough to swap out the cards in the back, so why bother with LILO and complicated partitioning schemes.

-I found that the system still is not stable at 133 Mhz FSB. I have been through two different motherboards and all kinds of different RAM. It runs very solid at 112 Mhz and isn't worth fighting anymore. One theory I have that I have not been able to confirm is if this board has a 1/4 PCI divider. If it doesn't, I would be running the PCI bus at 44 Mhz when the FSB is at 133 Mhz, which is probably too high. Sandra and Rayers SMB report 33 Mhz, but I'm not sure if I can trust them.

-Probably most significantly: I ditched the Creative 3D Blaster Savage4 Pro PCI video card. This thing just is not usable in real life. I only found two drivers that don't immediately blue screen the computer on boot up. The final drivers from the Creative site dated Nov 1999 and the version 8.20.06 reference drivers dated Feb 2000. Even then, whether or not the game you want to play will be glitchy (or work at all) is a crap shoot. I cannot recommend this card to anyone unless you are looking for a PCI or AGP video card for a DOS and Win 3.1 only machine.

It can be hard to find different drivers for the S3 Savage4 Pro. This site has a bunch that I couldn't find anywhere else: http://ftp.isu.edu.tw/pub/Hardware/multimedia … drv/index-e.htm. I have ran virus scans on all the Windows 9x drivers and they came out clean despite the site looking a bit sketchy.

I replaced the main video card with a NOS 16 MB TNT2 M64 PCI card. Yes, its the M64, but it still outperforms the Savage4 by a fair amount in most cases, and you can actually expect software to work with it. I am using the Detonator 8.05 drivers and DirectX 7. These things are a dime a dozen so I went ahead and overclocked the RAM. I had no issues increasing it to 180 Mhz which gives a nice performance boost.

C3 Ezra 866@1.0 Ghz (112x9)
Unreal Tournament (800x600x16, high detail): S3 Savage4 Pro (Metal) - 53 FPS, TNT2 M64 @ 143 Mhz - 32 FPS, TNT2 M64 @ 180 Mhz - 38 FPS
Quake 3 Demo 001 (800x600x16, high quality, OpenGL): S3 Savage4 Pro - 31.5 FPS, TNT2 M64 @ 143 Mhz - 35.4 FPS, TNT2 M64 @ 180 Mhz - 40 FPS
3D Mark 2000 (default settings): S3 Savage4 Pro - 993, TNT2 M64 @ 143 Mhz - 1698, TNT2 M64 @ 180 Mhz - 1880
3D Mark 99 (default settings): S3 Savage4 Pro - crash, TNT2 M64 @ 143 Mhz - 4038, TNT2 M64 @ 180 Mhz - 4162

Only two real downsides to the M64: 1) You have a noticeable loss in performance and quality in the Unreal games and 2) you lose 8-bit palletized texture support (not a big deal for me because I have a Voodoo 2 in here as well). Other than that, you still have decent DOS compatibility, the drivers are much better, things actually work, and in most cases are noticeably faster.

Its remarkable how close our setups are.

My current setup:
-Antec NSK 4400 ATX case
-Seasonic ATX power supply
-Micron 512 MB PC133 CL2 RAM
-Gigabyte GA-6BXC Rev 2.0 motherboard (with PowerLeap BIOS)
-MSI MS-6905 Slotket Rev 2.0
-Via C3 Ezra-T 1000 (100MHz x10)
-nVidia Quadro FX500 128MB AGP
-Powercolor Voodoo2 12MB PCI
-Corsair F120 120MB SATA SSD
-Promise SATA150TX2 PCI controller
-intel Gigabit PCI NIC
-TSSTcorp TS-H492C IDE DVD-ROM (with digital out connected to Yamaha soundcard)
-Yamaha YMF744 XG/OPL3 PCI soundcard using SB-Link
-Gravis Ultrasound PnP 16MB

Like you, the above setup seems rock stable at or below 112MHz FSB. The CPU itself is only stable <= 1232MHz (112 x11)
I'm starting to think the FSB problems we are having are maybe due to the motherboard, rather than chipset itself.

I looked through the thread but found no mention of why you specifically chose PCI primary videocards.

tpowell.ca wrote:

Its remarkable how close our setups are.

Well, lots of things are different, but it is funny how we both arrived on the same motherboard, BIOS, Ezra CPU, slotket, and an NVidia + Voodoo 2 combo. Things get pretty different when it comes to storage and sound.

What is the stepping code on your CPU? I would be interested in a faster Ezra T, but I have trouble telling them apart from the Nehemiahs. I have a A350EJ0. I read somewhere that A350 = Ezra and A360 = Nehemiah, but apparently this isn't always true.

tpowell.ca wrote:

I looked through the thread but found no mention of why you specifically chose PCI primary videocards.

The original idea was that I didn't want to have to worry about finding AGP cards that would run reliably on an overclocked AGP bus (133 * 2/3). Many 440BX boards have a 1/4 PCI divider available, so PCI cards at least will run in spec. I haven't been able to confirm though that this 1/4 divider exists on the GA-6BXC. It's one of my theories about the stability issues.

It looks like one of the keys for an Ezra or Ezra T is to make sure the voltage is 1.35v. Nehemiah will be 1.4 or 1.45.

Tenorman wrote:

The original idea was that I didn't want to have to worry about finding AGP cards that would run reliably on an overclocked AGP bus (133 * 2/3). Many 440BX boards have a 1/4 PCI divider available, so PCI cards at least will run in spec. I haven't been able to confirm though that this 1/4 divider exists on the GA-6BXC. It's one of my theories about the stability issues.

The PLL Datasheet will tell you about the PCI divider at 133MHz FSB. https://www.idt.com/document/dst/9148-26-datasheet page 3. It is proper 33MHz.

I had this VIA Apollo Pro (133MHz) Motherboard with bad VRM capacitors. It would only run with a 66MHz FSB at that point. After replacing the bad VRM capacitors it did 133MHz again.
If I can find some time I will see if my recapped GA-6BXC does 133MHz FSB for multiple hours without errors. In the past it always did 133MHz without issues, but it was not like it was set at 133MHz by default.
Past week my GA-6BXC is having some issues with Rayers FSB tool for DOS in combination with VIA (Nehemiah) processors. Intel CPU works fine with this tool though. Well, this motherboard was used for lots of experiments already.

gerwin wrote:

If I can find some time I will see if my recapped GA-6BXC does 133MHz FSB for multiple hours without errors. In the past it always did 133MHz without issues, but it was not like it was set at 133MHz by default.

I know that this doesn't necessary mean anything, but I have been through two different GA-6BXCs. The second one was actually new old stock and had never been used. Seeing as tpowell.ca has these issues as well, it seems 133 mhz is very hit or miss on these things.

Yes you wrote that. But all these motherboards are about 20 years old. The capacitors are also 20 years old and come from a time close to the capacitor plague.

gerwin wrote:

Past week my GA-6BXC is having some issues with Rayers FSB tool for DOS in combination with VIA (Nehemiah) processors. Intel CPU works fine with this tool though. Well, this motherboard was used for lots of experiments already.

I've noticed that if I mess with this utility too much on my system, it will stop working until I power the computer off and back on again. Changing FSB settings once or twice seems to be about it, after that you will just get errors.

Keeping everything the same, my board worked fine with an Ezra-T AG60AB0 (from 2002 week 42) at 1224MHz (112MHz x11) and a Nehemiah AG60AE1Q (from 2004 week 48) 1224MHz (112MHz x11).
I can run Rayer's FSB tool all day as long as I keep the FSB below 133MHz and above whatever combination with the multiplier keeps the CPU at or above 200MHz.

Tenorman. For sound, I gave up on my SB32/AWE32 since it lacked crucial SBPro compatibility and lets be honest, if you have a proper MT-32 and SC-55, the EMU8000 portion is just a waste. I was initially very fond of the YMF701 and YMF719 but in the end the YMF744 PCI won me over.

tpowell.ca wrote:

I can run Rayer's FSB tool all day as long as I keep the FSB below 133MHz and above whatever combination with the multiplier keeps the CPU at or above 200MHz.

Tenorman. For sound, I gave up on my SB32/AWE32 since it lacked crucial SBPro compatibility and lets be honest, if you have a proper MT-32 and SC-55, the EMU8000 portion is just a waste.

Interesting. I wonder if that is a difference between the Ezra and Ezra-T, or if this is what I am doing to make it mad... I run at 150 Mhz (50 x 3) and all the caches and branch prediction turned off all the time. Works quite well with a lot of older games.

I like messing with sound fonts for the games that will run from inside Windows 9x. I can understand just wanting to use "the best" though and not worry about it. I have an MT-32, but not a SC-55. My best general MIDI device is still an original Soundscape S2000, but that went back in my 486 because it doesn't get along with this machine.

So, everyone having issues with 133MHz FSB - is your VIA C3 actually rated at 100MHz or 133MHz?

This is something else to consider because you can have variants rated at both FSBs for a given rated CPU speed, and 133MHz FSB can be unstable for a 100MHz rated CPU. You can check the rating by looking at what is printed on the CPU heat spreader.

j^aws wrote:

So, everyone having issues with 133MHz FSB - is your VIA C3 actually rated at 100MHz or 133MHz?

This is something else to consider because you can have variants rated at both FSBs for a given rated CPU speed, and 133MHz FSB can be unstable for a 100MHz rated CPU. You can check the rating by looking at what is printed on the CPU heat spreader.

Mine is an 866. Says 133 x 6.5 right on it.

gerwin wrote:

If I can find some time I will see if my recapped GA-6BXC does 133MHz FSB for multiple hours without errors. In the past it always did 133MHz without issues, but it was not like it was set at 133MHz by default.

The 133MHz FSB results are in:
Software: Windows 98 / 3DMark2001 demo and game demo.
CPU: VIA C3 Nehemiah at 6.0x / 800MHz. Model "6.5x133/1,25V", CPUID 6.9.8, on MS6905 Master v2.3 Slotket jumpered to 1,4V.
Motherboard: GA-6BXC v2.0, recapped, running at 133MHz FSB. Booted at 50MHz, then used Rayer's tool to gradually go to 133MHz. AGP divider at 2/3 -> 89 MHz AGP clock. Edit: using Powerleap BIOS Award v6.00PG
Video: Voodoo 3 AGP.
Running time: 4 hours
Lockups: None (Minor Problem: Rebooted once because 3DMark gave an error message and did not want to start the demo anymore)

At first I went ahead with the Geforce4 Ti 4200 which was already installed, but that one locked up after two hours. Probably because of the AGP overclock. So I restarted the entire test with the Voodoo 3, which is known to be more tolerant in that regard.

Also I had the Slotket set to 'auto' voltage initially, but that did not work very well. That figures because 1,25V is out of spec.

In addition to the re-cap, another potential difference is the MSI MS-6905 Slotket. Tpowell.ca and I are both using v2.0 and you are using v2.3. You mentioned jumpering the slotket to 1.4v. A v2.0 only goes down to 1.5v. I am using a 1.35v CPU. Both the 1.5v and auto settings have the same stability issues on my system. It would be interesting to test with a different slotket, but unfortunately anything better than the MS-6905 v2.0 is either prohibitively expensive or impossible to find.

Tenorman wrote:

In addition to the re-cap, another potential difference is the MSI MS-6905 Slotket. Tpowell.ca and I are both using v2.0 and you are using v2.3. You mentioned jumpering the slotket to 1.4v. A v2.0 only goes down to 1.5v. I am using a 1.35v CPU. Both the 1.5v and auto settings have the same stability issues on my system. It would be interesting to test with a different slotket, but unfortunately anything better than the MS-6905 v2.0 is either prohibitively expensive or impossible to find.

Good you mentioned it. These voltage jumpers of the MS-6905 slotket are for passing a request signal to the Motherboard VRM. It is an intel defined five-digit bit pattern. This page has a copy of just that table. So it is not a very demanding slotket feature. Indeed this MS-6905 box scan I found does not mention the lower voltages, maybe because these settings were not deemed practical at the time. But all five jumpers are on the slotket. Attached is the full list of jumper combinations. When in doubt: with a multimeter you can verify which VCC is supplied to the CPU.

Looking at it, I don't see any difference between a Ver.2 and Ver.2.3 MS-6905 Slotket version. It is likely the PCB is exactly the same and therefor they had to sticker the Ver.2.2 and 2.3 ones. Below the sticker is probably the original Ver.2 PCB marking. No other visible differences AFAIK, but maybe some capacitor or resistor value was changed? I opened up the backside of the Ver.2.3 and nothing to see there either; no components, no correction wires.
All good news for MS-6905 Ver.2 owners: You are not missing out.

Another thing about the voltages: Obviously VIA C3 Nehemiah uses intel Tualatin voltage request patterns, otherwise there would not be a 1,25V variant. Both the 440BX motherboard VRM and the MS-6905 are not prepared for that. I have to compare what the intel Tualatin voltage patterns actually changed, but until then I would say: For Ezra-T and Nehemiah it is safer to specify a voltage manually with the slotket jumpers, instead of using the auto-voltage jumper setting.

The MS-6905 Master 2.0 and 2.3 are slightly different.
Ver 2.0 has 3 pin J3 and Ver 2.3 has 4 pin J3.

I've seen pics of a Ver 2.2, and seem to only have a 3 pin J3 as well, so I'm not sure how 2.0 and 2.2 are different.

That is the story, except when I checked my Ver.2.3 it has 3 pin J3...

gerwin wrote:

That is the story, except when I checked my Ver.2.3 it has 3 pin J3...

Interesting, and good to know.

Btw, any idea what the 3 pin J2 (reserved) does ?

I tried the manual setting for 1.35 volts and got the same result as always when using 133 FSB: 3DMark 2000 will crash sometime during the first test (helicopter) and 99 will crash during the second test (FPS). I have not verified the voltage with a multimeter.

I also upgraded the video card to a "real" TNT2 AGP card and got another nice performance increase. Unreal Tournament in particular seems to really like the new card. I've more than made up for the UT performance that was lost by ditching the S3 card.

C3 Ezra 866@1.0 Ghz (112x9)
Unreal Tournament (800x600x16, high detail):
S3 Savage4 Pro PCI 32 MB (Metal) - 53 FPS
TNT2 M64 PCI 16 MB @ 143 Mhz - 32 FPS
TNT2 M64 PCI 16 MB @ 180 Mhz - 38 FPS
TNT2 AGP 32 MB @150 Mhz - 56 FPS

Quake 3 Demo 001 (800x600x16, high quality, OpenGL):
S3 Savage4 Pro PCI 32 MB - 31.5 FPS
TNT2 M64 PCI 16 MB @ 143 Mhz - 35.4 FPS
TNT2 M64 PCI 16 MB @ 180 Mhz - 40 FPS
TNT2 AGP 32 MB @150 Mhz - 48 FPS

3D Mark 2000 (default settings):
S3 Savage4 Pro PCI 32 MB- 993
TNT2 M64 PCI 16 MB @ 143 Mhz - 1698
TNT2 M64 PCI 16 MB @ 180 Mhz - 1880
TNT2 AGP 32 MB @150 Mhz - 2161

3D Mark 99 (default settings):
S3 Savage4 Pro PCI 32 MB - crash
TNT2 M64 PCI 16 MB @ 143 Mhz - 4038
TNT2 M64 PCI 16 MB @ 180 Mhz - 4162
TNT2 AGP 32 MB @ 150 Mhz - 4444

I did not know about these 1.35v settings. I always used AUTO for my Ezra/Nehemiah/Ezra-T and had no issues -- same motherboard.

HOWEVER.... my modded Tualatin-S 1.4ghz (from Korea) would always crash if I had the FSB set to 133. It could go as high as 112, but not higher. I'm very interested to see if using one of these 'hidden' jumper settings on my MS6905 2.0 allows it to work. The seller had told me I could send him my motherboard (6BXC 2.0) and that he would mod it to work, but I didn't want to go through the trouble and pay the shipping.