I think I usually set that to 2 or 4 mb

For a reference:

Gigabyte GA-5AX
AMD K6-2+ 570 MHz 570ACZ @ mod K6-III+ 600 MHz/1,7 V
256MB SDRAM (CL2, fastest timing in BIOS, write allocation on)
nVidia GeForce4 MX440 128MB (128bit) AGP DVI XpertVision
3Dfx Voodoo 1 4MB PCI Yakumo (Biostar)
NEC D720101GJ USB 2.0 PCI
Creative Sound Blaster AWE32 IDE/MKE CSP CT3900
SD to IDE + 128GB SD card
IDE CD-ROM
GoTek Floppy Emulator
Seasonic G-360

I don't have the system running right now, this is from my archives (no special tweaks for performance):

3DMark 99 (drivers nvidia_9x_45.23)
--------------
3730 3DMarks
8582 CPU 3DMarks

Shogun wrote on 2022-09-01, 17:57:

I then took a look at CTU and tried to enable write combining using the method shown at My oldest rig, Hyundai HD-P5233M and per […]
Show full quote

I then took a look at CTU and tried to enable write combining using the method shown at My oldest rig, Hyundai HD-P5233M and performance didn't change. I tried both 32mb and 8mb in mtrr0. Also I'm unable to enable mtrr1. When I type something in it just changes back to all zeroes. Not sure why nothing changed with write combining on.

Image of how the write combining settings were applied from the thread referenced above but using the memory address shown on my system:

AFAIK, the CTU tweak is meant for systems that don't natively support the write combining trick of the AMD CPUs.

Also, are you sure you are applying the tweak to the right memory ranges? The Voodoo3 does in fact have more than two ranges allocated to it which you can see in the Device Properties. You can determine which ones are the ones meant for VRAM by their size. Both of the ranges that deal with VRAM will amount to 32 MB individually (and thus 64 MB together). The easiest way to determine you have the right ranges, is to type in the starting address for either bank, then choose the 32 MB size from the pulldown menu. If the end address matches what you see in the listed resources, you've found the right range.

What I discovered in my testing in the thread that you linked, is that of those two memory ranges only the first 8 MB actually represent local VRAM, and the rest is system RAM allocated to the Voodoo via the PCI bus.

leonardo wrote on 2022-09-01, 18:48:

AFAIK, the CTU tweak is meant for systems that don't natively support the write combining trick of the AMD CPUs.

Also, are you sure you are applying the tweak to the right memory ranges? The Voodoo3 does in fact have more than two ranges allocated to it which you can see in the Device Properties. You can determine which ones are the ones meant for VRAM by their size. Both of the ranges that deal with VRAM will amount to 32 MB individually (and thus 64 MB together). The easiest way to determine you have the right ranges, is to type in the starting address for either bank, then choose the 32 MB size from the pulldown menu. If the end address matches what you see in the listed resources, you've found the right range.

What I discovered in my testing in the thread that you linked, is that of those two memory ranges only the first 8 MB actually represent local VRAM, and the rest is system RAM allocated to the Voodoo via the PCI bus.

I went back into CTU and for whatever reason it wasn't saving my settings. I checked the box to maintain settings at startup. It then kept the settings and I was able to edit mtrr1. After setting everything it still isn't having any impact.

System properties showing memory range:

CTU settings:

3DMark 99 bench showing no improvement:

I said that I would provide some numbers for my two K6+ VIA machines. These are my two desktops.
I do have a laptop that I could run, but it would really only be useful for DOS benches, as it has a terrible Trident video chip.
Anyway here they are. These are Windows tests only, I can run some DOS numbers if you need them.

System 1
Soyo 5EMA+ 1MB L3 (latest BIOS) ATX
VIA MVP3 Chipset
Tightest RAM timings and fastest caches enabled, L3 Cache enabled
No other optimizations
Micron PC133 256MBx2 (512MB) Cl2 Ram @100MHZ
AMD K6-3+ 6x100mhz @2.2v 600mhz
3dfx Voodoo 3 3000 AGP @ 2x - Stock Clocks, latest official driver
Aureal Vortex 2 PCI
Creative SB16 ISA
Quantam Fireball HDD UDMA 2 Pirmary Master
Lite-On 52X CD-ROM Secondary Master
VIA PCI USB 2.0 (2.0 mode disabled)
VIA 4n1 4.35 version (AGP 2x "turbo mode")
Windows 98SE/ No unofficial patches
Direct X 7.0

3dMark 99:
Default Benchmark 800x600x16
3222 3dMarks
7510 CPU Marks

3dMark 2000:
1617 3dMarks

Final Reality:
3.63 (I was way off on my previous guess, must have been a p3 rig)

Unreal (OG version, not Gold) Glide mode, 800x600x16
3 pass castle flyby:
AVG: 40.6fps
High: 81.8
Low: 17.5

System 2:
Tekram P5MVP-B4-1M (Baby AT)
Patched BIOS, 1MB Cache - Fastest RAM timings, AGP 2X forced. L3 Cache enabled
VIA MVP3 Chipset
AMD K6-2+ 550mhz (100x5.5 2.0v)
Micron PC133 CL2 256MBx2 @100mhz (512MB)
BFG Nvidia FX5500 AGP 4x @ 2x - Stock Clocks, 45.23 Driver (FX5200)
256MB DDR 128bit
WD Caviar UDMA2 IDE Primary Master
Lite-On 52x CD-ROM - Secondary Master
Aureal Vortex 1 PCI
Onboard USB 1.1
VIA 4.1 4.35 AGP installed in "turbo" (2x) mode
Windows 98SE No unofficial patches
Direct X 8.0a

3dMark 99:
Default Benchmark 800x600x16
2292 3dMarks
6359 CPU Marks

3dMark 2000:
1422 3dMarks

Final Reality:
3.16 (AGP test 29.73 MAX VRAM)

Unreal (OG version, not Gold) D3D, 800x600x16
3 pass castle flyby:
AVG: 23 fps
High: 49.5 fps
Low: 7.6fps

This took a few minutes to run. Hope it helps.
No memory or CPU tweaks were ran.
If anyone wants to see any other benchmarks let me know.

elcrys wrote on 2022-09-01, 18:19:

For a reference: […]
Show full quote

For a reference:

Gigabyte GA-5AX
AMD K6-2+ 570 MHz 570ACZ @ mod K6-III+ 600 MHz/1,7 V
256MB SDRAM (CL2, fastest timing in BIOS, write allocation on)
nVidia GeForce4 MX440 128MB (128bit) AGP DVI XpertVision
3Dfx Voodoo 1 4MB PCI Yakumo (Biostar)
NEC D720101GJ USB 2.0 PCI
Creative Sound Blaster AWE32 IDE/MKE CSP CT3900
SD to IDE + 128GB SD card
IDE CD-ROM
GoTek Floppy Emulator
Seasonic G-360

I don't have the system running right now, this is from my archives (no special tweaks for performance):

3DMark 99 (drivers nvidia_9x_45.23)
--------------
3730 3DMarks
8582 CPU 3DMarks

ram.png

SSTIMG01.PNG

SSTREP01.TXT

You’re running 600 at 1.7v?

Nice. How far does it go at 2.2 I wonder 🤔

Of some relevance are my overclocked scores in Sandra. Epox MVP3G-M.

For comparison, here's an Asus TXP4 with a K6-3+ @500 MHz (6x83).

Here's my Soyo 5EMA+.

Sphere478 wrote on 2022-09-02, 04:41:

elcrys wrote on 2022-09-01, 18:19:

For a reference: […]
Show full quote

For a reference:

Gigabyte GA-5AX
AMD K6-2+ 570 MHz 570ACZ @ mod K6-III+ 600 MHz/1,7 V
256MB SDRAM (CL2, fastest timing in BIOS, write allocation on)
nVidia GeForce4 MX440 128MB (128bit) AGP DVI XpertVision
3Dfx Voodoo 1 4MB PCI Yakumo (Biostar)
NEC D720101GJ USB 2.0 PCI
Creative Sound Blaster AWE32 IDE/MKE CSP CT3900
SD to IDE + 128GB SD card
IDE CD-ROM
GoTek Floppy Emulator
Seasonic G-360

I don't have the system running right now, this is from my archives (no special tweaks for performance):

3DMark 99 (drivers nvidia_9x_45.23)
--------------
3730 3DMarks
8582 CPU 3DMarks

ram.png

SSTIMG01.PNG

SSTREP01.TXT

You’re running 600 at 1.7v?

Nice. How far does it go at 2.2 I wonder 🤔

Probably one of the 1.6v chips. I doubt that it will go much further than 600 even with higher voltages.
Still like to have one of the 1.6v chips to play with though.

Repo Man11 wrote on 2022-09-02, 05:15:

Here's my Soyo 5EMA+.

Same board as mine. I guess I should do some memory benchmarks to try and compare.
Will your chip do stable overclocked? I assume you just matched the OP speed for your tests.

Jasin Natael wrote on 2022-09-02, 13:45:

Repo Man11 wrote on 2022-09-02, 05:15:

Here's my Soyo 5EMA+.

Same board as mine. I guess I should do some memory benchmarks to try and compare.
Will your chip do stable overclocked? I assume you just matched the OP speed for your tests.

This board overclocks quite well but it doesn't have some of the memory settings that the Epox board does, so it lags a bit behind. Unfortunately, both boards have memory cache that cannot work above 112 MHz FSB so if I run either of them at 620 MHz (5x124) the onboard memory cache has to be disabled. The CPU used for the above screenshot is a K6-3 450 - they were never great overclockers, so I've never bothered to try and run it at anything other than default speed.

Repo Man11 wrote on 2022-09-02, 14:44:

Jasin Natael wrote on 2022-09-02, 13:45:

Repo Man11 wrote on 2022-09-02, 05:15:

Here's my Soyo 5EMA+.

Same board as mine. I guess I should do some memory benchmarks to try and compare.
Will your chip do stable overclocked? I assume you just matched the OP speed for your tests.

This board overclocks quite well but it doesn't have some of the memory settings that the Epox board does, so it lags a bit behind. Unfortunately, both boards have memory cache that cannot work above 112 MHz FSB so if I run either of them at 620 MHz (5x124) the onboard memory cache has to be disabled. The CPU used for the above screenshot is a K6-3 450 - they were never great overclockers, so I've never bothered to try and run it at anything other than default speed.

Yeah I had the same experience with onboard cache. But honestly 112mhz FSB was never much of a priority to me.
I did not notice that the chip was a vanilla K6-3, so that makes perfect sense.

Jasin Natael wrote on 2022-09-02, 13:43:

Sphere478 wrote on 2022-09-02, 04:41:

elcrys wrote on 2022-09-01, 18:19:

For a reference: […]
Show full quote

For a reference:

Gigabyte GA-5AX
AMD K6-2+ 570 MHz 570ACZ @ mod K6-III+ 600 MHz/1,7 V
256MB SDRAM (CL2, fastest timing in BIOS, write allocation on)
nVidia GeForce4 MX440 128MB (128bit) AGP DVI XpertVision
3Dfx Voodoo 1 4MB PCI Yakumo (Biostar)
NEC D720101GJ USB 2.0 PCI
Creative Sound Blaster AWE32 IDE/MKE CSP CT3900
SD to IDE + 128GB SD card
IDE CD-ROM
GoTek Floppy Emulator
Seasonic G-360

I don't have the system running right now, this is from my archives (no special tweaks for performance):

3DMark 99 (drivers nvidia_9x_45.23)
--------------
3730 3DMarks
8582 CPU 3DMarks

ram.png

SSTIMG01.PNG

SSTREP01.TXT

You’re running 600 at 1.7v?

Nice. How far does it go at 2.2 I wonder 🤔

Probably one of the 1.6v chips. I doubt that it will go much further than 600 even with higher voltages.
Still like to have one of the 1.6v chips to play with though.

I wrote about the CPU here (a standard 2.0V CPU):
Re: SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod

In short, with air cooling:
660 MHz @ 1.9V (stable in Prime95)
690 MHz @ 2.2V (unstable in Prime95)
700 MHz @ 2.4V (testing motherboard itself was unstable at 140 MHz FSB)

In the end I went with efficiency and 600 MHz @ 1.7V (Prime95 stable). I was also surprised to have these voltages stable, I was suspecting that the board (GA-5AA) feeds higher voltage, but no - I later confirmed the results with GA-5AX.

elcrys wrote on 2022-09-02, 16:55:

I wrote about the CPU here (a standard 2.0V CPU): Re: SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod […]
Show full quote

Jasin Natael wrote on 2022-09-02, 13:43:

Sphere478 wrote on 2022-09-02, 04:41:

You’re running 600 at 1.7v?

Nice. How far does it go at 2.2 I wonder 🤔

Probably one of the 1.6v chips. I doubt that it will go much further than 600 even with higher voltages.
Still like to have one of the 1.6v chips to play with though.

I wrote about the CPU here (a standard 2.0V CPU):
Re: SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod

In short, with air cooling:
660 MHz @ 1.9V (stable in Prime95)
690 MHz @ 2.2V (unstable in Prime95)
700 MHz @ 2.4V (testing motherboard itself was unstable at 140 MHz FSB)

In the end I went with efficiency and 600 MHz @ 1.7V (Prime95 stable). I was also surprised to have these voltages stable, I was suspecting that the board (GA-5AA) feeds higher voltage, but no - I later confirmed the results with GA-5AX.

That is interesting. I would assume that the returns would be pretty diminishing at 660mhz. But that is still neat to see one running at that clock speed.
My K6-3+ chip will likely do 600 at 2.1v, I've not tried honestly, it has always been ran at 2.2 and stays cool and happy.
My K6-2+ 533 chip is rated at 2.0v and it will do 600 at 2.2 but isn't stable.
The heat increase is rather noticeable too. It will run at 550 at stock and stays as cool as can be, so I left it at that.

I do have another K6-2+ that is rated at 550mhz. It is in my only working K6 laptop, so I've not bothered to raid it. But I might one day...

I ended up switching out the motherboard. I swapped out the Shuttle for a Gigabyte GA-5AX Rev. 4.1. Oddly the memory performance went up a decent amount but the video card performance dropped a bit - 38fps vs 32fps in Quake 3 for example. I could have done the swap a bit more gracefully as I didn't uninstall stuff before putting in the new mobo but all seemed fine. I think I'll stick with this motherboard though as it isn't displaying the errant behavior I saw in the shuttle like shutting off incorrectly or restarting during the boot up process.

I compiled all the results provided in this thread and similar results I found elsewhere. I'll post here shortly in case it can be of any use to someone else.

Jasin Natael wrote on 2022-09-02, 13:43:

Sphere478 wrote on 2022-09-02, 04:41:

elcrys wrote on 2022-09-01, 18:19:

For a reference: […]
Show full quote

For a reference:

Gigabyte GA-5AX
AMD K6-2+ 570 MHz 570ACZ @ mod K6-III+ 600 MHz/1,7 V
256MB SDRAM (CL2, fastest timing in BIOS, write allocation on)
nVidia GeForce4 MX440 128MB (128bit) AGP DVI XpertVision
3Dfx Voodoo 1 4MB PCI Yakumo (Biostar)
NEC D720101GJ USB 2.0 PCI
Creative Sound Blaster AWE32 IDE/MKE CSP CT3900
SD to IDE + 128GB SD card
IDE CD-ROM
GoTek Floppy Emulator
Seasonic G-360

I don't have the system running right now, this is from my archives (no special tweaks for performance):

3DMark 99 (drivers nvidia_9x_45.23)
--------------
3730 3DMarks
8582 CPU 3DMarks

ram.png

SSTIMG01.PNG

SSTREP01.TXT

You’re running 600 at 1.7v?

Nice. How far does it go at 2.2 I wonder 🤔

Probably one of the 1.6v chips. I doubt that it will go much further than 600 even with higher voltages.
Still like to have one of the 1.6v chips to play with though.

the 1.6v chips are the same cores as other + chips, it's just because of the low clocks they can take less volts. what's the highest clocked 1.6? like 450mhz? I'm sure there was binning involved also.

edit:

elcrys wrote on 2022-09-02, 16:55:

I wrote about the CPU here (a standard 2.0V CPU): Re: SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod […]
Show full quote

Jasin Natael wrote on 2022-09-02, 13:43:

Sphere478 wrote on 2022-09-02, 04:41:

You’re running 600 at 1.7v?

Nice. How far does it go at 2.2 I wonder 🤔

Probably one of the 1.6v chips. I doubt that it will go much further than 600 even with higher voltages.
Still like to have one of the 1.6v chips to play with though.

I wrote about the CPU here (a standard 2.0V CPU):
Re: SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod

In short, with air cooling:
660 MHz @ 1.9V (stable in Prime95)
690 MHz @ 2.2V (unstable in Prime95)
700 MHz @ 2.4V (testing motherboard itself was unstable at 140 MHz FSB)

In the end I went with efficiency and 600 MHz @ 1.7V (Prime95 stable). I was also surprised to have these voltages stable, I was suspecting that the board (GA-5AA) feeds higher voltage, but no - I later confirmed the results with GA-5AX.

^see haha they have a unicorn chip. ( I want one 🙁 )

Sphere478 wrote on 2022-09-03, 04:18:

the 1.6v chips are the same cores as other + chips, it's just because of the low clocks they can take less volts. what's the hig […]
Show full quote

Jasin Natael wrote on 2022-09-02, 13:43:

Sphere478 wrote on 2022-09-02, 04:41:

You’re running 600 at 1.7v?

Nice. How far does it go at 2.2 I wonder 🤔

Probably one of the 1.6v chips. I doubt that it will go much further than 600 even with higher voltages.
Still like to have one of the 1.6v chips to play with though.

the 1.6v chips are the same cores as other + chips, it's just because of the low clocks they can take less volts. what's the highest clocked 1.6? like 450mhz? I'm sure there was binning involved also.

edit:

elcrys wrote on 2022-09-02, 16:55:

I wrote about the CPU here (a standard 2.0V CPU): Re: SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod […]
Show full quote

Jasin Natael wrote on 2022-09-02, 13:43:

Probably one of the 1.6v chips. I doubt that it will go much further than 600 even with higher voltages.
Still like to have one of the 1.6v chips to play with though.

I wrote about the CPU here (a standard 2.0V CPU):
Re: SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod

In short, with air cooling:
660 MHz @ 1.9V (stable in Prime95)
690 MHz @ 2.2V (unstable in Prime95)
700 MHz @ 2.4V (testing motherboard itself was unstable at 140 MHz FSB)

In the end I went with efficiency and 600 MHz @ 1.7V (Prime95 stable). I was also surprised to have these voltages stable, I was suspecting that the board (GA-5AA) feeds higher voltage, but no - I later confirmed the results with GA-5AX.

^see haha they have a unicorn chip. ( I want one 🙁 )

Yeah I would like to have one as well...

Jasin Natael wrote on 2022-09-03, 16:54:

Sphere478 wrote on 2022-09-03, 04:18:

the 1.6v chips are the same cores as other + chips, it's just because of the low clocks they can take less volts. what's the hig […]
Show full quote

Jasin Natael wrote on 2022-09-02, 13:43:

Probably one of the 1.6v chips. I doubt that it will go much further than 600 even with higher voltages.
Still like to have one of the 1.6v chips to play with though.

the 1.6v chips are the same cores as other + chips, it's just because of the low clocks they can take less volts. what's the highest clocked 1.6? like 450mhz? I'm sure there was binning involved also.

edit:

elcrys wrote on 2022-09-02, 16:55:

I wrote about the CPU here (a standard 2.0V CPU): Re: SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod […]
Show full quote

I wrote about the CPU here (a standard 2.0V CPU):
Re: SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod

In short, with air cooling:
660 MHz @ 1.9V (stable in Prime95)
690 MHz @ 2.2V (unstable in Prime95)
700 MHz @ 2.4V (testing motherboard itself was unstable at 140 MHz FSB)

In the end I went with efficiency and 600 MHz @ 1.7V (Prime95 stable). I was also surprised to have these voltages stable, I was suspecting that the board (GA-5AA) feeds higher voltage, but no - I later confirmed the results with GA-5AX.

^see haha they have a unicorn chip. ( I want one 🙁 )

Yeah I would like to have one as well...

I have one, but the (modified to 256k L2) K6-2+ 570 I have will run at 620 MHz, where my 1.6 volt K6-3+ 400 will not, even at 2.4 volts with a good sized cooler.

Repo Man11 wrote on 2022-09-03, 16:59:

Jasin Natael wrote on 2022-09-03, 16:54:

Sphere478 wrote on 2022-09-03, 04:18:

the 1.6v chips are the same cores as other + chips, it's just because of the low clocks they can take less volts. what's the highest clocked 1.6? like 450mhz? I'm sure there was binning involved also.

edit:

^see haha they have a unicorn chip. ( I want one 🙁 )

Yeah I would like to have one as well...

I have one, but the (modified to 256k L2) K6-2+ 570 I have will run at 620 MHz, where my 1.6 volt K6-3+ 400 will not, even at 2.4 volts with a good sized cooler.

I meant a near 700mhz specimen.