F2bnp wrote:

Sounds awesome 😀. You say that 4.5x130Mhz is actually slower than 5x120 with L3 Cache, I suppose that's on SuperPi? How about 3D games?

I should buy another K6-III+ CPU and see if it can hit 600Mhz, mine can do 575, but 600 is always unstable 😵 .

I need to go to bed now and the results are still on the socket 7 systems but tomorrow I can post some results.

But yes, its mostly SuperPi that really likes the onboard cache. With the cache off I got 4.52 as best at 5x120 MHz this time around. 4,5x130 gets about the same time while 4.5x135 got ~4m 42s the single time I ran it, I have high hopes for 4,5x140...

In 3dmark99 I know that 4.5x130 and 5x120 with motherboard cache produced the same 3dmark score but 5x120 with cache got a better CPU score. I think the two settings scored more or less the same in 3dmark2000 and 3dmark2001 and the difference in games wasnt large either.

I diddnt manage to finish all the game testing today as Im rerunning all tests with some new BIOS settings that is 0.1 - 0.3 FPS faster for the speed settings that run with the onboard cache disabled. I will wait with posting most of the results another day or two 😜

Here are two more scores that Im pretty happy with for now.

New personal SuperPi record again with this CPU and motherboard.

SuperPi 1M @ 5x125MHz with the onboard cache deactivated.

This is my best 3dmark99 CPU score with the K6-3+ so far, the 3dmark score is nothing special.

3dmark99 @ 5x125MHz with the onboard cache deactivated, Voodoo III 2000 PCI @183/183

Very nice, iirc my k6-3+ 600mhz was around 8000 cpu marks so that is a nice jump!

But you said it: 3dmarks nothing special. How is that possible? on a k6-2 450 (p5a, 128mb) i scored 2293 3dmarks with a v3 pci 2000 at stock settings and 1024x768.

V3 is still bottlenecked but i expected a bigger jump anyway with a 175mhz higher clocked cpu with cache and an overclocked v3 at lower resolution.

3dmark99 hangs a lot on 3dnow! maybe try running sandra or everest (or both).

btw, p2-450 and stock v3 2000 pci 1024x768 scored 3252 3dmarks so get up there, start using ice! 🤣 (p3-450 scores even higher with sse)

meljor wrote:

Very nice, iirc my k6-3+ 600mhz was around 8000 cpu marks so that is a nice jump! […]
Show full quote

Very nice, iirc my k6-3+ 600mhz was around 8000 cpu marks so that is a nice jump!

But you said it: 3dmarks nothing special. How is that possible? on a k6-2 450 (p5a, 128mb) i scored 2293 3dmarks with a v3 pci 2000 at stock settings and 1024x768.

V3 is still bottlenecked but i expected a bigger jump anyway with a 175mhz higher clocked cpu with cache and an overclocked v3 at lower resolution.

3dmark99 hangs a lot on 3dnow! maybe try running sandra or everest (or both).

btw, p2-450 and stock v3 2000 pci 1024x768 scored 3252 3dmarks so get up there, start using ice! 🤣 (p3-450 scores even higher with sse)

I think the rather bad 3dmark score is because of the combination of the Gigabyte GA-5AX and the PCI Voodoo III.

Phil wrote that in his testing the Gigabyte GA-5AX with a Voodoo III was slower than all other Socket-7 boards he tried. It seems the board is good for CPU benchmarks but sucks when it comes to 3D, especially with a Voodoo III. In DOS for some reason he got more competetive scores so the performance issue seems limited to Windows.

The driver Im using is perhaps not the fastest either but its the same driver I used with my Intel reference system. I get pretty bad 3d scores in all benchmarks and games with the combination of the PCI Voodoo III and the Gigabyte GA-5AX.

Another thing that perhaps isnt optimal is that Im using Direct X 8.1a, again this was what I used with my Intel reference system which wasnt exactly fast but at least more in line with my expectations.

I have benched Sandra 99, PCMark 2002 and CPUmark 99 and the scores I got were great so the CPU performance isnt the issue, neither is the memory bandwidth.

Edit

Well its not the driver, I tested one of the last WHQL drivers for the Voodoo III and it was much slower than "Amigamerlins" driver pack.

I think Im going to give up trying to get the Gigabyte GA-5AX to perform good with the Voodoo III PCI and try one o my M577 PC-chips boards instead. I will still post the results I got with the Gigabyte GA-5AX though, just not tonight. The boring thing with the M577 is that it dosnt support any FSB over 100 MHz.

I managed to shave off another couple of seconds and set a new best SuperPi 1M time again.

K6-3+ @ 4.5x135 Mhz, Gigabyte GA-5AX, 256 MB PC133 CL3 @ 135 CL2.

Still no luck with running SuperPi @ 4.5x140 as the memory module the board likes best isnt really capable of 140 MHz CL2, at least not in my hot apartment. It can wait until the cold season.

I diddnt bother with running Sisoft Sandra at 4.5x135 but that is just as well as the setting isnt 100% stable, the PCI bus runs a bit to far out of spec for the Voodoo III. Here is a screenshot with the memory bandwidth at 4.5x130 which is 100% stable, not too bad.

Here are some rather bad FPS numbers with the PCI Voodoo III, the Gigabyte Ga-5AX really seems to hate the card.

K6-3+ 5x100, 256MB CL2, Voodoo3 "3500" PCI.

640*480 16bit

GLQuake: 58.5 FPS
Quake II standard OpenGL driver 640*480: 37.5 FPS
Quake III: 34.7 FPS
Unreal Gold: 44.7 FPS

K6-3+ 4.5x130, 256MB CL2, Voodoo3 "3500" PCI.

640*480 16bit

GLQuake: 76.3 FPS
GLQuake with miniGL 1.47: 89.3 FPS
Quake II standard OpenGL driver: 48.2 FPS
Quake II 3dnow! standard OpenGL: 53.4 FPS
Quake III: 43.3 FPS
Unreal Gold: 55.5 FPS

K6-3+ 5x125, 256MB CL2, Voodoo3 "3500" PCI.

640*480 16bit

GLQuake: 73.4 FPS
GLQuake with miniGL 1.47: 90.9 FPS
Quake II standard OpenGL driver: 47.8 FPS
Quake II 3dnow! standard OpenGL: 53.2 FPS
Quake III: 43.6 FPS
Unreal Gold: 56.5 FPS

K6-3+ 5x120 256MB CL2 Voodoo3 "3500" PCI, Onboard cache on.

640*480 16bit

GLQuake: 76.2 FPS
GLQuake with miniGL 1.47: 94.2 FPS
Quake II standard OpenGL driver: 49.3 FPS
Quake II 3dnow! standard OpenGL: 56 FPS
Quake III: 44.3 FPS
Unreal Gold: 58.0 FPS

The Quake II numbers with 3Dfx miniGL 1.47 was 2-3 FPS better than the ones with 3dnow! + the standard OpenGL driver, at best 59 FPS.
All FPS numbers posted here are with the games running at 640*480 but the scores at 1024*768 were not much worse.

I really have no idea why this system performs this bad with the PCI Voodoo III. I have found the screenshots I thought I lost with some results from last year when I benched the exact same hardware down to the memory module with a Geforce 2 MX and a Geforce 3 ti200. Those two video cards performed as I would have expected, the PCI Voodoo III did not.

Oh lord, this again! I've encountered the exact same behavior from a Voodoo 3 PCI on my GA-5AX. As soon as I installed a V3 AGP performance was far better. I have no clue as to what is going on exactly. You're saying 130MHz should be stable enough? I'd love to give it a try.

F2bnp wrote:

Oh lord, this again! I've encountered the exact same behavior from a Voodoo 3 PCI on my GA-5AX. As soon as I installed a V3 AGP performance was far better. I have no clue as to what is going on exactly. You're saying 130MHz should be stable enough? I'd love to give it a try.

Yea I did not have any problems with 130 MHz FSB as long as the onboard cache is disabled.

Higher FSB needs more voltage though, this CPU runs 6x100 with 1.8 - 1.9V but I know it needs at least 2.0V for 5x120 and 4.5x130, I used 2.1V during my testing at these speeds just to be on the safe side. I used 2.2V for 6x125 which was stable when the ambient temp was below 25C but yesterday when the ambient was 28-30C the CPU needed 2.3V for 100% stability. Im using an Intel cooler for a Coppermine CPU and those are not very large but the cooler never got hotter than lukewarm.

Good to know that Im not the only one experiencing performance issues with the combination of a PCI Voodoo III and the Gigabyte GA-5AX.

Have you tried exploring PCI settings with something like TweakBIOS? There MUST be a logical explanation for all of this.

alexanrs wrote:

Have you tried exploring PCI settings with something like TweakBIOS? There MUST be a logical explanation for all of this.

Not really, Im not going to use this system for anything so Im not willing to spend much more time smacking my head against the wall.

Here is a screenshot from last year with the same system running GL-Quake @ 600 MHz with a Geforce 3 ti200. This is with the motherboard cache deactivated and not using what I now know is the fastest BIOS settings.

GL-Quake 1024*768 32bit. K6-3+@600, 256MB, Geforce 3 ti200.

A K6-3+ can be pretty speedy, just not running in this motherboard with a PCI Voodoo III, at least not without more tweaking. If I am going to use this system as a gaming rig sometime in the future I will probably use a Geforce 3 ti200.

In the benchmarks i did on a k6-2 450 and p5a (got 3dmark99 numbers for more than 20 cards) the v3 pci was 20% faster compared to the agp. the 3000 pci was a little faster even but the 3500 agp could not match that.

Do you happen to have a voodoo4 agp? It was the fastest card on this platform, even outperforming v5 5500 pci with the same driver. All at 1024x768.

But between the v3 cards the pci was faster (but banshee pci was even above that). In a real game the v4 4500 was still the fastest of all but it was matched by the banshee pci, also at 1024x768.
The matrox g400 agp was a good performer also on this board. Every nvidia was a little slower and all scored about the same, including gf3 ti200.

On slower systems (k6, p2-450) scores were all over the place with some cards. Much better scaling on a p3-700 and higher.
For example: The highest scores (game and 3dmark) on a p2-450 came from a tnt2-ultra and g400, even beating a geforce4 ti at1024x768.

meljor wrote:

In the benchmarks i did on a k6-2 450 and p5a (got 3dmark99 numbers for more than 20 cards) the v3 pci was 20% faster compared t […]
Show full quote

In the benchmarks i did on a k6-2 450 and p5a (got 3dmark99 numbers for more than 20 cards) the v3 pci was 20% faster compared to the agp. the 3000 pci was a little faster even but the 3500 agp could not match that.

Do you happen to have a voodoo4 agp? It was the fastest card on this platform, even outperforming v5 5500 pci with the same driver. All at 1024x768.

But between the v3 cards the pci was faster (but banshee pci was even above that). In a real game the v4 4500 was still the fastest of all but it was matched by the banshee pci, also at 1024x768.
The matrox g400 agp was a good performer also on this board. Every nvidia was a little slower and all scored about the same, including gf3 ti200.

On slower systems (k6, p2-450) scores were all over the place with some cards. Much better scaling on a p3-700 and higher.
For example: The highest scores (game and 3dmark) on a p2-450 came from a tnt2-ultra and g400, even beating a geforce4 ti at1024x768.

I do not have a Voodoo III or 4 AGP or a Banshee PCI.

If both F2bnp and I experience the same performance issue with a Voodoo III PCI running on the Gigabyte GA-5AX I think we at least can be sure of that there is an issue with the combination and I will leave it at that.

Yes but i just find it strange that on the p5a the opposite happens, since it uses the same chipset. On other platforms the pci and agp voodoo3's are on par. Only with real fast p3 cpu's the agp version becomes faster.

Edit: geforce 2mx pci and agp both score the same on the p5a, so it must be a voodoo ''thing''

meljor wrote:

Yes but i just find it strange that on the p5a the opposite happens, since it uses the same chipset. On other platforms the pci and agp voodoo3's are on par. Only with real fast p3 cpu's the agp version becomes faster.

Edit: geforce 2mx pci and agp both score the same on the p5a, so it must be a voodoo ''thing''

These days I like things that just work. 15-20 years ago I had unlimited patience for quirks and actually thought it was fun trying to figure out why things diddent work as expected but now when Im working in a field where I do that all the time I do not have the endurance to do it at home as much anymore. I will probably stumble on an AGP Voodoo III sooner or later and then I will see if I can get some more reasonable results. For now I will think of the Gigabyte GA-5AX as a Nvidia video card only motherboard.

I still have 3 more K6-3+ CPUs I never even tested but I seriously doubt they will be better than the first one so Im thinking that perhaps its time to take a short break from tinkering with Super Socket 7 and see what I can do with an Intel 440LX motherboard and then return to the Gigabyte GA5-AX to see if I can improve last years results with a Nvidia card. I still like the Gigabyte GA-5AX, in CPU benchmarks the board is really fast.

It is still a VERY nice overclock. The ga5-ax is the best board for that. I think the asus tops out at 115-120fsb and most other boards simply do 100fsb and stop there.

It's a shame i found a ga5-ax that was dead, i would love to play with one.

There are a couple of undocumented settings for 124MHz and 133MHz FSB for the P5A and P5A-B boards.

https://web.archive.org/web/20050212201337/ht … imh/p5a-oc.html

I just tested my FIC VA-503+ with my V3 2000 PCI just to see if the PCI bus made any difference.

If you remember, I hit 86.9 FPS with the K6-3+ at 550 with a V3 3500 AGP (183Mhz).

My V3 2000 locked up in the console when OC'ed to 183MHz, but when I ran it at 166MHz, it scored 81.0 FPS. Even at stock 143MHz, it scored 70.7 FPS. Something tells me that if I could get it to run at 183MHz (not really interested in trying), it might be even faster than the V3 3500 AGP because of all the SS7/AGP weirdness.

I find it hard to believe that the VIA and Ali chipsets are THAT far apart, speed-wise.

boxpressed wrote:

I just tested my FIC VA-503+ with my V3 2000 PCI just to see if the PCI bus made any difference. […]
Show full quote

I just tested my FIC VA-503+ with my V3 2000 PCI just to see if the PCI bus made any difference.

If you remember, I hit 86.9 FPS with the K6-3+ at 550 with a V3 3500 AGP (183Mhz).

My V3 2000 locked up in the console when OC'ed to 183MHz, but when I ran it at 166MHz, it scored 81.0 FPS. Even at stock 143MHz, it scored 70.7 FPS. Something tells me that if I could get it to run at 183MHz (not really interested in trying), it might be even faster than the V3 3500 AGP because of all the SS7/AGP weirdness.

I find it hard to believe that the VIA and Ali chipsets are THAT far apart, speed-wise.

Neither do I and as meljor had better performance with his Asus board with Ali chipset I think we can be sure of that this problem is limited to the Gigabyte GA-5AX, and perhaps not even all GA-5AX but at least both my board and F2bnp's board.

I thought that I would post some benchmarks where the K6-3+ really shines 😀 I will also post Pentium III scores as a reference.

First out is CPUMark 99

Pentium III Katmai 6x100 MHz, The slow cache really hurts the performance.

Pentium III Coppermine 6x100 MHz, The fast cache makes the Coppermine perform well.

Pentium II coppermine 4.5x133 MHz, More FSB equals more performance.

K6-3+ 4.5x130 MHz, Pretty much the same clock for clock performance as the Coppermine.

K6-3+ 5x125 MHz, Without onboard cache the clock for clock performace is a little worse but still @625 MHz it beats the Coppermine@4.5x133.

Edit added the K6-3+ @ 625 MHz

I installed a Geforce 3 ti 500 in the Gigabyte GA-5AX system, its hell of a lot faster tha the PCI Voodoo III.

The only issue is that the power mosfet for the AGP port gets slightly warm... like the surface of the sun and it heats the board which makes the cache chip alot warmer so it wont run at 120 MHz 😜. Im pretty sure this is the reason I benched with the cache disabled when using Geforce AGP cards last year... I will need to find a solution to this issue... later.

For now the the system runs fine at 6x120 MHz without the onboard cache but 4.5x130 MHz did not want to run 3dmark 2001. Im pretty sure 4.5x130 worked just fine with a Geforce 3 ti200 last year but I guess this ti500 card isnt liking the 87 MHz AGP bus very much.

Nice score on the cpu mark. I am REALLY curious how it scores in the cpu tests of everest, can you run those for me please? Especially interested to see what the fsb does for the fpu tests, as it is usually a lot slower as p2 in these.

If you want to run these: On the lavalys page there are older versions, please use 2253 beta. (the program now has the name aida, older versions are everest)

Shame about the ti500, still would like to see how it goes compared to the v3 pci.