Here is the list of video cards that don't show a video signal in my motherboard:

1PCI  Dell 09629U   P0002            Nvidia TNT		
2AGP  VisionTek     NV996            Nvidia TNT2 M64
3AGP  Compaq 179997 P0016            Nvidia TNT2
4AGP  Dell 0G0170   P118             Nvidia Geforce FX5200
5PCI  GeCube        GC-R9250LPCI-C3  ATi Radeon 9250
6PCI  VisionTek     X1550 V/D/VO	    ATi Radeon x1550

The computer boots with these cards, there's just no video signal.

Just about finished with the benchmarking.

I thought I'd try out the fastest card that works with the motherboard to see how well it performs. I have a FireGL 9700 X1-128 and a busted up Radeon 9800Pro with a missing inductor and a broken fan.

I got the FireGL 9700 X1-128 running in Windows 98 using catalyst 6.2 with the 9700pro driver. Everything runs fin in dos windows except for 3dmark. 3dmark99 gives me about a dozen of the tiny arrow check boxes that the catalyst control center uses as bullet points and then locks. 3dmark2000 goes black for 5 seconds and then returns to the windows desktop. Hmmmm. Is it because of the driver? The driver with the FireGl card? The driver with the FireGl card and 3dnow and this bios? I upgraded to DirectX9c. No luck.

So I put a new inductor on the 9800pro and rigged a fan.

It looked great in a 2d desktop. The Text mode on the other hand ....

I tried 3dmark and got the same failures as I did with FireGL 9700 X1-128. Tried some quake 2 GL and it was running pretty well for about a minute before it had a burst of VGA garbage before going black. Looks like the 9800 pro needs some more recovery time.

I thought about trying the omega drivers with the 9700. For a second. And then I remembered why I hated rage cards back in the day.

But on the plus side, I was finally able to get ATTO benchmark to run on the innodisk CF. It only does UDMA 4 with the promise controller.

I'd really like to get that 9700 to run just to see what the best case scenario might look like, even though it seems like it might not make that much of a difference on a slow CPU.

I tried these patches: Re: 3dmark99 MegaThread Re: 3dmark99 MegaThread

I’ve always known that SGRAM was more desirable than SDRAM, but I had thought it was just more capable ram, as in it could hit higher frequencies. After looking at my benchmark, I can see that SGRAM out performs SDRAM by about 5-15% on 3D workloads at the same frequency and I started asking more questions. How could that be? Bulk fills and masked writes handled by the ram chips themselves? I see. Seems reasonable. Z occlusion built into the ram? My My. Very Interesting. I should have stopped there. I started looking into more advanced GDDR features . When I got up to bus and address inversion, my mind was blown like the first time I looked into MFM vs RLL or Row Hammer exploits. Time for bed. That’s quite the box of tidbits that I never knew about.

I tried a bunch of stuff like disabling AGP x2 in the BIOS & changing the AGP aperture window. No difference.

Turns out that going back to ATI Catalyst 4.3 did the trick. 3dmark 99 and 2000 are running now on the FireGL 9700.

https://vogonsdrivers.com/getfile.php?fileid=1567

douglar wrote on 2025-08-29, 18:55:

Here is the list of video cards that don't show a video signal in my motherboard: […]
Show full quote

Here is the list of video cards that don't show a video signal in my motherboard:

Copy code to clipboard

1PCI  Dell 09629U   P0002            Nvidia TNT		
2AGP  VisionTek     NV996            Nvidia TNT2 M64
3AGP  Compaq 179997 P0016            Nvidia TNT2
4AGP  Dell 0G0170   P118             Nvidia Geforce FX5200
5PCI  GeCube        GC-R9250LPCI-C3  ATi Radeon 9250
6PCI  VisionTek     X1550 V/D/VO	    ATi Radeon x1550

The computer boots with these cards, there's just no video signal.

I got the AGP TNT and the FX5200 cards to work in DOS! I wish I knew how though. PCI TNT is still a no-go. And I still don't have them working in Win98.

I also recapped an old 64 bit memory MX 4000 card because my working MX 4000 card turned out to only be 32 bit memory.
Here is is with Puffy Caps-- https://theretroweb.com/expansioncards/s/bfg-asl-mx4000
Curiously, I went by the name "Puffy Caps" when I was a slam poet. (<--- Joke!)

Repo Man11 wrote on 2025-08-25, 23:23:

When I've used Promise IDE cards with MVP3 motherboards they have always worked, but at some point in installing Windows 98 drivers Windows would get stuck when rebooting and I would have to move the card to another PCI slot. Once all of the drivers were installed this would no longer happen and I could leave it in the same slot from then on. Just an FYI.

Apparently this is also a good strategy for PCI video cards too. If you card doesn't initialize in PCI slot 3, try moving it to PCI slot 2. Now it works. If it isn't working in slot 2, try slot 3. It must reset the PCI Bus Enumeration or something like that.

Anyway, I have a surprise leader in the benchmarks right now. PCI Creative Labs 3D Blaster Banshee (CT6750)

edit -- Although it looks like it just locked hard in 3dmark 2000 =(
edit -- It was HOT! Seems to be running stable now that I put a fan on it
edit -- Moving the card around fixed some cards, but the Radeon PCI 9250 & X1550, Riva TNT PCI , and Intel i740 AGP are still not liking this board.
edit -- Intel i740 AGP works if I disable AGP 2x features in the BIOS

The full results are below--

Seems like the Banshee might be best for earlier games and the MX4000 might be better for later games, but I guess that all comes down to drivers

What are some good games to test that would be picky about the cards and drivers ?

Notes:

• The Chris3d benchmark gives a pretty clear rating of AGP vs PCI
• Could not get the PNY MX4000 to run doom fast, even though it had some of the best 3d mark scores
• I ran 3d Bench from an 800x600x16bit desktop using the default test button. That's 800x600x 16bit for 3DM99 and 1024x768x16bit for #DM2K.
• The cards slower than the i740 had serious 3d rendering issues. The i740 itself had pretty good rendering quality
• 3d Bench 99 seemed CPU limited on everything faster than the i740
• 3d Bench 2000 seemed like there was a bandwidth limitation and a driver CPU limitation.
• In the end, PCI vs AGP didn't matter all that much. In the Voodoo3 case, the PCI card had SGRAM that offset any advantages that AGP had.
• While the Fire 9700 is clearly the fastest card here, the K6-III 400 CPU wasn't able to feed it fast enough, allowing cards like the MX4000 and Voodoo that have lower CPU requirements in the drivers to out pace it.

I rearranged some of the rows in the table above and I found another NV5 card yesterday and I added it in the chart.

I think the noticeable things are that --

• Doom scores pretty much plateau after the GPU's got to 66Mhz, except for the later Nvidia cards when DOS became lower priority
• Even though the three cards in the Rage 128 category has PCI, AGP and faster AGP, the 3dmark scores don't really change, suggesting that even this card
• Unclear why the pair of NV18 cards do so well on 3dmark 2000. Do they cheat some how?

OK , I see that I got the wrong date on the Rage Pro Ultra. Should be 1999-09.

Working through the cards on an Nforce 2 Biostar-m7ncg-400 motherboard to see how they do with an excess of CPU.

Pretty sure the nvidia cards expect me to change drivers at least 3 times before they work just to show that I care. And I'm still stumped on some of the cards that benchmarked fine of the K6-III. Didn't get any easier. The 2d stuff is fine in windows & DOS, but either both 3D mark programs won't run or sometimes it's just 3d mark 2000. I probably need to strip down the install and start over again.

On the hunt for a good AGP universal board in my storage bins that can test all the cards, since I can't test the AGP 2x cards in the new system.

And of course that Riva TNT PCI that wouldn't do a thing in the MVP3 board works just fine in the much newer board. I guess it's nice that it works. Well, maybe "works just fine" is a bit of an over-statement. All of the PCI cards have been doing somewhere between below expectations to just plain poor in the Nforce board. Could it be running PCI at a slower clock rate?

Cool stuff, nicely compiled info.

SiS boards are supposed to be good for generally broad AGP compatibility, though their CPU<>RAM performance is a few percent down on other options.

I was able to finish up the TNT, Geforce 2, and Geforce 4MX benchmarks after a lot of effort. Wiping and reinstalling windows from scratch didn't do it. My order of battle for Win9x is usually to install windows --> chipset drivers --> direct x --> video drivers. Something was going wrong. I also tried installing windows --> direct x --> chipset drivers --> video drivers because the forceware drivers seemed to want that. I eventually I went outside my favorite driver bundles and tried the 6.18 nvidia drivers from phils computer lab and that 1) broke the forceware control panel and 2) made 3dmark 2000 work. Could those two things might be related? Not motivated to keep poking at that anymore. That took a long time.

I would have really liked to get a Geforce4 4200 to test, but the prices have jumped so high that I'm hesitant to spend that much on a card from a family that's got a rep for sudden BGA death. Those MX cards are not the best thing in the world, but they cost very little, are more than enough for a K6 CPU and will probably still be working after I'm gone!

I should have a working universal AGP board next week with a Duron 900. I'll use that to round out my CPU scaling.

Seems like PCI cards cause the integrated chipset graphics to wake up on my Nforce 2 based Biostar-m7ncg-400. I'm temped to see if I can revive my nforce 2 ultra while wait for parts.

That little rack for your cards is really cool, where'd you get it?

I made them. Can you see these?

https://www.tinkercad.com/dashboard/collectio … /cXmIKfS7SBp/3d

https://www.tinkercad.com/things/5mAiN2849sr- … d-stand-25-gram
https://www.tinkercad.com/things/iYex3uzUq3u- … d-stand-23-gram
https://www.tinkercad.com/things/16pMpNOqcl8- … d-stand-30-gram
https://www.tinkercad.com/things/gHMSxvgj37Y- … d-stand-33-gram

https://www.tinkercad.com/things/1jjy3PxmTaZ- … o-rack-54-grams

I wanted to do some CPU scaling tests and here we are 4 weeks later and I think i finally got a motherboard working with a Universal AGP port. I pulled 4 boards out of my closet. All were dead. I recapped. I swapped CPU's. Socket 370. Socket 462. No luck. I got a duron 900 on an ECS K7S5A SiS 735 off ebay for $30. Motherboard worked for about 60 seconds. Then it too went nighty-night. What terrible luck.

Eventually I started digging through the wall of salvaged PCs. Turns out they were all earlier than 2000 or after 2005 except for a Dell XPS with a i815 motherboard, that proprietary ATX power supply and a PIII 800EB. It's as dusty as they come, but it's working.

Every Win98 build I've been doing lately, I've been trying to see if my Innodisk ICF 4000 CF will work with it. The answer is always no. Writes to the device are slow. Gets about 50% of the way through a Win98 install and locks up. On paper, the stats look great. I wish I hadn't bought 4 of them. Looks like they were used in an embedded WinXP install in the past. Didn't get any better luck with the i815 chipset than I did with the MVP3 or the Nforce 2.

1) So many weird quirks. There was a good two hours this morning where I had to lean an 80mm fan against the AGP card to get the system to boot. Not required after 2 hour though. It got better after the green AGP clip snapped off. https://theretroweb.com/motherboards/s/intel- … easton-dell-oem
2) Gotten pretty good at identifying driver issues vs Direct X issues. Started adding quake benchmarks, so I'm looking forward to troubleshooting Rage OpenGL issues in the future.

this thread sums up my experience with the Soyo 5EHM 1.3 1MB L2

i have never been able to run AGP stable on this motherboard nor the IDE in UDMA mode

the strange thing is i remember using for about a year, in 2000, the Aopen AX59Pro + Aopen PA3000 (TNT2 m64) + K6-2 500Mhz and that worked flawlessly

its funny how your brain acts but i swear that combination worked better, in some instances, than the actual one (k6-2+ @ 600Mhz, 512MB, Voodoo 3 2000 PCI, SB Live!, SSD @ PCI IDE card) or maybe i wasnt that strict with the performance

I finally found my Diamond S60 Radeon 7000 PCI card that I hadn't seen since covid. It was in the original box inside a box of original boxes tucked under the foosball table. Crazy.

What's crazier is that got me looking for more cheapo "cockroach" Radeon cards to round out the tests. I was looking on retroweb for Radeon 9200 cards and I noticed the picture of one had my basement rug as a background. <Keanu Reaves voice> Whoah! </Keanu Reaves voice> Took about 30 minutes to find it. I had put it in a shuttle cube and stacked it up on a top shelf.

The 7xxx to 9xxx Radeon's don't properly display mode 10h Chris's 3d bench. I think that's pretty well known issue. Bench mark still seems to work though. And curiously enough the bug was fixed on my radeon x700.

Things I've learned about what to do when 3d mark won't start--

• First try disabling vendor control panels and widgets and then reboot
• If rebooting didn't fix it, try reinstalling Direct X
• If reinstalling Direct X didn't fix it, try reinstalling the same driver from the setup program
• If reinstalling the driver didn't fix it, try reinstalling Direct X one more time
• If that didn't fix it, try running a different 3d benchmark first like Quake 3 before running 3dmark
• And only once I reach this point, consider spamming different driver versions.

I added in some CPU Scaling. I re did things with more consistent driver versions.

System 1:
K6-III 400
DFI P5BV3+/E (REV.:B) https://theretroweb.com/motherboards/s/dfi-p5bv3-e-rev-b
256MB SDRAM @ 100Mhz
8GB San Disk Ultra on a Promise Ultra 100 TX2 https://theretroweb.com/expansioncards/s/prom … se-ultra100-tx2
3dMark CPU Score ~6000

System 2:
Intel PIII 800EB
Intel D815EEA (Easton) https://theretroweb.com/motherboards/s/intel- … easton-dell-oem
256MB SDRAM @ 133Mhz
16GB SD on Sintechi Reader
3dMark CPU Score ~12000

System 3:
AMD Athlon XP Barton 2500+
Biostar M7NCG 400 Ver: 7.2 https://theretroweb.com/motherboards/s/biostar-m7ncg-400
256MB DDR @ 333 Mhz
32GB Samsung NGFF Sata Device on a Jmicron Bridge
3dMark CPU Score ~37000

Software used :

• Win98se & Direct X9c redistributable
• Drivers from the retroweb for the old stuff
• Catalyst 4.3 on all the Radeon stuff except for the the x700 that needed Catalyst 6.2
• Detonator driver on the TNT cards and the Gforce GTS card
• The MX cards got forceware 71-84, The FX5200 got forceware 81-98

Notes:

• The PCI cards that I have often seem to have boost of some sort, either a clock speed boost or SGRAM instead of DRAM
• The PCI vs AGP difference is very small at 400 Mhz
• The PCI vs AGP difference gets much larger on faster boards, and not just because the AGP gets faster. The PCI seems to be slower. But maybe the question isn't "how do I fix this?" but instead it is "Why are you running PCI graphics on this AGP 8x system"
• There are a couple spots of regression in the chart where things are slower with the faster CPUs which are curious, like on the Permedia 3dMArk 2000 test, or the Doom tests on the MX cards.
• The RV200 card wasn't stable on the AGP 2x system or the AGP 8x system. Curiously, this vaguely awakes old memories of forgotten frustration for that one weekend when I thought I try out a bargain Radeon card that I picked up on close out and then promptly returned. The R280 cards are much nicer.

Here is the list of cards:
https://theretroweb.com/expansioncards/s/diam … d-2000-8-ram-ic
https://theretroweb.com/expansioncards/s/a-trend-atc-2345b
https://theretroweb.com/expansioncards/s/ati- … -xpression-plus
https://theretroweb.com/expansioncards/s/ati-xpert-work-agp
https://theretroweb.com/expansioncards/s/eon-lung-hwa-dp335
https://theretroweb.com/expansioncards/s/stb- … -max-2-g-16-agp
https://theretroweb.com/expansioncards/s/gain … cardex-9804-12a
https://theretroweb.com/expansioncards/s/stb- … ocity-128zx-agp
https://theretroweb.com/expansioncards/s/ati- … -wonder-128-pci
https://theretroweb.com/expansioncards/s/ati- … -wonder-128-agp
https://theretroweb.com/expansioncards/s/ati- … 28-pro-ultra-gl
https://theretroweb.com/expansioncards/s/visiontek-tnt2-m64
https://theretroweb.com/expansioncards/s/nvid … -p0002-0000-b01
https://theretroweb.com/expansioncards/s/dell-0040u
https://theretroweb.com/expansioncards/s/diamond-viper-v770
https://theretroweb.com/expansioncards/s/nvidia-p0016
https://theretroweb.com/expansioncards/s/crea … -banshee-ct6760
https://theretroweb.com/expansioncards/s/3dfx … odoo-3-2000-pci
https://theretroweb.com/expansioncards/s/3dfx-voodoo-3-2000
https://theretroweb.com/expansioncards/s/nvid … -32mb-reference
https://theretroweb.com/expansioncards/s/nvidia-p73cr
https://theretroweb.com/expansioncards/s/msi- … 000-t64-ver-200
https://theretroweb.com/expansioncards/s/pny- … 00-64mb-ddr-pci
https://theretroweb.com/expansioncards/s/bfg-asl-mx4000
https://theretroweb.com/expansioncards/s/nvidia-p118-vga
https://theretroweb.com/expansioncards/s/ati- … on-7000-ddr-agp
https://theretroweb.com/expansioncards/s/pcpa … -ddr-v-d-vo-pci
https://theretroweb.com/expansioncards/s/ati- … der-radeon-7500
https://theretroweb.com/expansioncards/s/cele … a-radeon-9200se
https://theretroweb.com/expansioncards/s/gecu … gc-r9250lpci-c3
https://theretroweb.com/expansioncards/s/ati- … deon-9200-128mb
https://theretroweb.com/expansioncards/s/powercolor-r96-lc3
https://theretroweb.com/expansioncards/s/ati- … egl-9700-x1-128
https://theretroweb.com/expansioncards/s/sapp … 00-hybrid-256mb

Edit - On Review, I think that STB Glyder card has a 128bit memory pathway or a faster memory clock or both.