I've tried F6 and F7B bios' and no difference unfortunately.

3dmark2000 is around 1300 while the mvp3 can do 1600, so not a massive difference but still not 100% I feel.

I may have to go back to a K6-2 non plus and see if it still underperforms.

Did you try the Ali AGP tool v1.4?
Try and set Frame Buffer Size to "ALL", enable Write allocation without hole (and maybe also set GAT to "0"). Then reboot and see if it works. I recommend using the Ali (ULI) driver version 2.13
This has always fixed the performance for me when using Voodoo cards (although not on GA-5AA/5AX, since I don't have them, but other Ali Aladdin V boards).

Thanks for the reply, yeah I've tried all that. I tried a V3 PCI too. I will have to test a Geforce at some point.

Just tested with the motherboard L2 cache disabled and same performance.

Looking at all my various benchmarks (in my mvp3) it basically performs the same as a K6-2 500 (non +), except in CPUMark where it gets 48 (a lot better than the non plus K6s).

Just tested 3dmark99 and also ~2300. So very much the same speed as RockstarRunner.

I've got an older V1.1 version of this board I might try.

I'd switch over to my mvp3 board but don't have the ps2 header for it and seems the com ports have issues, otherwise it is a very good board.

Damn, I really need to get my hands on one of these boards. The issues you guys are having with them are really intriguing. Me likey! 😁

Hah yeah that would be great 😁

bloodem wrote on 2022-02-10, 10:06:

Damn, I really need to get my hands on one of these boards. The issues you guys are having with them are really intriguing. Me likey! 😁

mwdmeyer wrote on 2022-02-10, 10:10:

Hah yeah that would be great 😁

Well, thanks to @RockstarRunner, I have been playing with the GA-5AA rev 3.2, and... man, what a ride! 😀
The board has a revision B1 South Bridge and revision G North Bridge with internal tag RAM (there is no external tag RAM present), so it can cache up to 512 MB of RAM, which I was able to confirm (there is no performance difference between 128 MB and 256 MB of RAM).

I first tried the board with a GeForce 2 MX and then a GeForce 2 Ti. Both of them work perfectly at AGP 1X with driver 7.76 (there are random freezes @ AGP 2X, but that's to be expected with most SS7 boards). Anyway, performance @ AGP 1X is identical to that @ AGP 2X, so no biggie.
The board works fine at FSB 110, 115, 120 (it might also work at 133... though I haven't tried it).

TEST SYSTEM:
CPU: AMD K6-2+ 570 OC @ 605 MHz (FSB110 x 5.5)
VGA: GeForce 2 Ti (OC @ Ultra speeds)
SOUND: NO SOUND CARD
ALI AGP DRIVER: 2.13
nVIDIA DRIVER: 7.76

ALI AGP TOOL CHANGES:

• AGP RATE: 1x
• GAT MODE: 2
• FRAME BUFFER SIZE: ALL
• WRITE ALLOCATION: Enable without Hole

BENCHMARK RESULTS:

3DMark99 default: 4836 3DMarks / 8418 CPU Marks
3DMark2000 default: 4328 3DMarks / 267 CPU Marks
GLQuake 640 x 480 x 16: 160.3 FPS
Quake 2 OpenGL 640 x 480 x 16: 93.2 FPS
Quake 2 OpenGL 3DNow! 640 x 480 x 16: 105.3 FPS
Quake 2 Software 3DNow! 640 x 480 x 16: 32.7 FPS
Quake 3 640 x 480 x 16: 61.1 FPS
MDK2 640 x 480 x 16: 61.67 FPS
Expendable 640 x 480 x 16: AVG 44.12 FPS / HIGH 76 FPS / LOW 31 FPS

These results are very solid for a SS7 system (but still lower than what I get with the Asus P5A rev 1.04).
The most important speed boost comes from the higher FSB, which seems to be a bottleneck for K6-2/3(+) CPUs.

So, obviously, the next thing I tried was a Voodoo 3 3000 AGP... and, boy, was I disappointed! As others have mentioned, performance with this video card is pretty bad, ~ 50% lower than on other ALI or VIA based boards:
3DMark99: 2538 3DMarks
GLQuake 3dfx Glide: 126 FPS
Quake 2 3dfx Glide: 54 FPS 😮

I literally tried everything humanly possible, tested every combination of BIOS settings, tested both the F6 and F7b BIOS ROMS (both support K6-2/3 "+" CPUs), tested every combination of AGP tool settings, multiple ALI AGP drivers, multiple Voodoo 3 drivers (including some newer unofficial ones)... nothing worked.
Then I also tested a PCI Voodoo 3 3000 to see if this was somehow related to an AGP bug or not... and, it's not. The PCI version has the exact same problem.

So, now I'm starting to wonder if maybe this is a bug/incompatibility between the ALI "G" revision north bridge (or its implementation on these Gigabyte boards) and Voodoo cards.
To try and get to the bottom of this, I'd be interested to hear from other Gigabyte GA-5AA owners (or even GA-5AX owners) that are using them with Voodoo 3 cards:
- what are your 3DMark99 / GLQuake / Quake 2 scores (ideally, you should first use the ALI AGP tool and set FRAME BUFFER SIZE: ALL / GAT MODE: 2 / WRITE ALLOCATION: Enable without Hole, and then restart the OS )
- what motherboard revision you have
- what chipset (north bridge) revision your motherboard has.

CONCLUSION:
- The GA-5AA does not have any issues with K6-2/3 "+" CPUs
- It works perfectly fine (it's both fast and very stable) with GeForce 2 cards
- something kills the performance with Voodoo 3 cards, my initial guess is an incompatibility between these cards and the chipset (might only affect revisions "G" of the chipset)

Not sure if there are any GA-5AA boards with the revision "H" north bridge (I know there are some late GA-5AX rev 5.2 boards that come with this revision). It would be interesting to see how those perform with Voodoo 3 cards as well (although this revision apparently only fixes the "suspend to RAM" function).
In the upcoming weeks, I'll also test 2 x Voodoo 2 cards in SLI, I'm very curious to see if they suffer from the same bug.

Really good findings, bloodem, glad you were able to get somewhere with the board. Also glad that it wasn't just my experience either 😆 now I wonder what could be the issue between the chipset and the V3.

How does V2 perform?

RockstarRunner wrote on 2022-03-05, 12:20:

Really good findings, bloodem, glad you were able to get somewhere with the board. Also glad that it wasn't just my experience either 😆 now I wonder what could be the issue between the chipset and the V3.

How does V2 perform?

Yeah, at this point I have a hard time believing that there is an easy software solution to this problem. Based on what I've seen, everything points to a hardware compatibility problem. More data is needed to draw a conclusion, though.
I haven't tested the Voodoo 2 cards yet, but I will definitely do it either this weekend or the next.

Power delivery problem? Power delivery has been an issue with multiple early GB boards. But I'd imagine that'd show itself in the form of lockups and general instability. My GA-BX2000 even has a special Voodoo jumper.

http://www.plasma-online.de/index.html?conten … gabyte_tnt.html

GF2 Ti should be far more efficient with the 150nm process than a V3 or a regular geforce 2.

A001 wrote on 2022-03-05, 15:56:

Power delivery problem? Power delivery has been an issue with multiple early GB boards. But I'd imagine that'd show itself in the form of lockups and general instability. My GA-BX2000 even has a special Voodoo jumper.

http://www.plasma-online.de/index.html?conten … gabyte_tnt.html

GF2 Ti should be far more efficient with the 150nm process than a V3 or a regular geforce 2.

I know about the "Voodoo jumper", which basically uses the PSU's 3.3V rail to power the card directly, however the behavior in this case is not indicative of a power related issue (as you said, power issues should cause restarts, instability, lockups). I actually do have some SS7 boards with this exact issue.

Anyway, just to make sure, I tested a GeForce 3 Ti 500 on the GA-5AA board, and it works fine (and that card is more than twice as power hungry as a Voodoo 3 3000).
So something else is happening, but we need more data from other GA-5AA/GA-5AX owners in order to identify a clear pattern.

Thanks I tested with both a Voodoo 3 2000 PCI and also Voodoo 3 3000 AGP, and both had the same performance issues.

I haven't tried any other cards yet.

mwdmeyer wrote on 2022-03-06, 09:12:

Thanks I tested with both a Voodoo 3 2000 PCI and also Voodoo 3 3000 AGP, and both had the same performance issues.

I haven't tried any other cards yet.

Could you post a high res picture of your motherboard?
I'm specifically interested in the motherboard revision and chipset revision.

Yep, I'm just away at the moment but will in a week or so. I have 2, one v3.2 and one v1.1, only tested v3.2 so far.

mwdmeyer wrote on 2022-03-06, 11:01:

Yep, I'm just away at the moment but will in a week or so. I have 2, one v3.2 and one v1.1, only tested v3.2 so far.

That's awesome!
Testing both revisions could give some valuable clues as to what's going on here. Keep us posted! 😀

Here is a photo of the Rev 1.1 board. My 3.2 is in a machine at the moment. I will try and swap out soon and compare.

https://www.vogonswiki.com/images/b/b3/Gigaby … AA_rev_1.1.jpeg

mwdmeyer wrote on 2022-03-11, 04:24:

Here is a photo of the Rev 1.1 board. My 3.2 is in a machine at the moment. I will try and swap out soon and compare.

https://www.vogonswiki.com/images/b/b3/Gigaby … AA_rev_1.1.jpeg

Nice! As expected, Rev 1.1 has the older "E" North Bridge revision. Obviously, it also has the external tag RAM, so it can only cache up to 128 MB of RAM (which, IMO, is perfectly fine for SS7).

However, if my assumption is correct, this 1.1 revision should work fine with a Voodoo 3 (although you might still need to do the aforementioned fine tuning with the Ali AGP tool, in order to achieve best performance).

Tested TNT2 Ultra AGP in Rev 3.2 board and performance issues are gone. 43fps Quake 3 (has sound card installed), 1752 3dmark 2000, 3651 3dmark99.

mwdmeyer wrote on 2022-03-12, 04:14:

Tested TNT2 Ultra AGP in Rev 3.2 board and performance issues are gone. 43fps Quake 3 (has sound card installed), 1752 3dmark 2000, 3651 3dmark99.

Yes, don't think any nVIDIA cards have issues. There's something specific going on with the Voodoo 3, my guess is most other cards will work fine (maybe even the Voodoo 2?).
Still haven't managed to test 2 x Voodoo 2 cards in SLI, but will definitely do it (hopefully before the end of the month).

bloodem wrote on 2022-03-12, 05:53:

mwdmeyer wrote on 2022-03-12, 04:14:

Tested TNT2 Ultra AGP in Rev 3.2 board and performance issues are gone. 43fps Quake 3 (has sound card installed), 1752 3dmark 2000, 3651 3dmark99.

Yes, don't think any nVIDIA cards have issues. There's something specific going on with the Voodoo 3, my guess is most other cards will work fine (maybe even the Voodoo 2?).
Still haven't managed to test 2 x Voodoo 2 cards in SLI, but will definitely do it (hopefully before the end of the month).

Im going to get up this thread because i have the same issues with a ga-5ax 4.1 , south 1543c A1 and north 1541-A1.

i get with a voodoo 3 2000 no more than 2400 points in 3Dmark, with the k6-2 450 2.2V version 5740 (i didnt saw info but i assume is correct)

Tried the agp utility to force the agp + reg values and im going to test what you said (3dfx tools reports agp 2x, 3dmark report pci bus and dxdiag repport agp support not available, cpu z report agp 1.0 2x) i assume the correct info is coming form 3dfx tools & cpuz.

By the way, i have some stutters with half life, places with lights or effects even 17 fps-20fps, others games i tried for example the last one is turok in glide, perfect, the other one was doing some stutter is unreal.

EDIT2: testing a lot of bios options and im getting even less points, now 5500 cpu and 2165 for gpu.
without getting more points, less performance easy to reach.
EDIT3: swaped ide driver to win98se default ones "5229"(hdd detected as generic) + restart bios defaults and configure graphics aperture size etc = cpu 5806 and 2507 gpu
EDIT4: 2542 and 5840 with the 2.13 drivers suggested above, i dont understand even how i can install it?, i have read the readme i didnt saw my chipset there,anyway is working, thanks for the recomendation.

im starting to think, can be this ralated to the poor cpu cache has my cpu?

Done a clean OS install 6000 cpu & 2450 gpu.

This ide drivers 1.56 clone my dvd rom drive x7, i dont understand why.

I tried another one that comes with the package integrated driver but is not compatible or something because it dosnt apply the drivers to the bus & ide channels, still the 98se fabric ones.

This win98 honestly, im going to do something waiting for just install the drive or whatever and work, but it always gives me a stop hand, no man!, (remember jurassic park 1, the locked computer to enable the electricity from the fences), is the 90% of the times looking for workarounds, it's worth it for the games, but the operating system I've never seen anything this bad in my life.

By the way, i found a important data, aparently (i cant confirm the info) this ga-5ax 4.1 rev was made for fujitsu siemens:

http://www.motherboards.org/forums/viewtopic.php?t=72614

EDIT: gate 0 makes a lot of difference here, from around 5700 to 5919, and something like 2350 to 2435, without the agp utility "is better" still there a loss on the cpu enabling it, this one is pretty poor, so maybe is just a lack of l2 cache or whatever.
EDIT2: 2505 and 6109 reached after disabling passive release, isa buffer and delay transaction (no isa card installed) and usb port, some of those make a difference.
EDIT3: Data merge makes again a lot of difference, 2555 and 6218, i tried this option yesterday and probably i changed another thing at the same time because i had a lot less performance, and discarted it but is good to enable it.
EDIT4: Disable VGA frame buffer in this case is not good, i read somwhere on the net "sometimes is not working good", in this case that is not true with a voodoo 3, im going to test spread spectrum because is the first board i can put my hands in this option 😀 ,modern ones, at least the ones i had this option is not in the bios, It is only available if the bios is modified, I want to try it.