Thank you, I don't know what I'm doing. I've added this:

https://www.amazon.com/dp/B00066IO2O

to convert from molex to 4-pin cpu to plug it into the right place.

Continuing my adventures with the ISA855, I couldn't resist getting one of the knock-off "new" nVidia FX 5500 PCI cards to try with it.

By all accounts, the FX5500 was terrible when it came out, but it is said to be possibly the best PCI-based 3D accelerator with Win98 driver support and way better than most Win98 compatible games would expect. As this board only has PCI and they're fairly cheap, why not try it? In the UK these cards sell for "too much for me if it's DOA" from UK sellers - about £80, about £50 from ebay/aliexpress direct from China, and about £40 via US ebay, but still shipped direct from China. £40 was low enough to trick myself into buying it. It took far less time to arrive than estimated too - only a couple of weeks despite an estimate of far longer.

Unfortunately, when installed in either PCI slot, the FX5500 is completely "missing". The nVidia BIOS message doesn't show up, and the main BIOS just switches to the built-in video chip. I have various BIOS settings to try, but it's not looking promising right now. I thought it might be DOA, but I have a random newer motherboard sitting on my bench right now that POSTs and the nVidia VBIOS message showed up and the VGA port on the nVidia outputs the video signal. So, although I haven't confirmed the nVidia card is working fully, it's not DOA.

My speculation right now for why:
1) PCI version incompatibility. I assume not - the mobo claims 2.2, I can't find info on the (non-knock off) versions of the FX5500. I don't think there was too much different between 2.2/2.3 but I know very little about PCI.
2) Power lines. The mobo is obviously odd with power - it only takes 12V and must have regulators to provide the 3.3/5 voltages to the parts that need it. Perhaps it doesn't have everything wired up to the PCI bus that the card is expecting.
3) Power draw. As (2), but the card may be drawing more than the mobo can supply via the small 12V connector (or my old PSU can provide on the 12V rails). I think not on this, as I'd expect stability issues with the card plugged in if that were the case.
4) Just a regular incompatibility: BIOS weirdness, BIOS setting I haven't found yet. Perhaps some corner of the PCI spec the motherboard or card designer chose not to implement.
5) Something related to the ISA bridge chip, the other unusual thing about this board.

The fact that the ATI rage pro PCI card I have been using works tells me that the PCI bus on the mobo is at least semi-working.

As a side note: The £20 ATI rage pro 8MB that I acquired to improve DOS gaming was generally better under DOS than the built-in 855GM graphics (faster framerates in benchmarks, no graphics artifacts during VESA benchmarks). I had never got around to trying the built-in graphics under Win98, but while trying to get the nVidia working I went on a side-quest to see what that was like with Halflife. No benchmarks, just "feel". Direct3D/OpenGL performance of the ATI rage running Halflife is terrible (unusable at 1024x768), but the high-performance-for-the-era CPU in the ISA855 makes software rendering totally fine with the ATI rage at that res. The mobo's built in graphics chip (Intel's "Extreme Graphics 2" separate driver needed, not included part of the chipset driver) gave what look like excellent Halflife framerates with OpenGL and Direct3D drivers (and software rendering). I didn't do side by side comparisons of the output, but they all looked fine to me. I did notice some odd sound corruption with my ISA soundcard when using the 855M, so that's not great, though it might be something I haven't noticed before. Maybe because I had the non-functional nVidia card on the PCI bus (can't remember), or some BIOS setting I was trying to get the nVidia working has caused issues for the ISA bridge/card, or it might just due be to changing between graphics modes/drivers - several times I had to restart halflife when switching between software/opengl/direct3d. My current display only goes up to 1024x768 so I couldn't test higher resolution performance.

RetroFlump wrote on 2024-02-03, 10:56:

Unfortunately, when installed in either PCI slot, the FX5500 is completely "missing". The nVidia BIOS message doesn't show up, and the main BIOS just switches to the built-in video chip. I have various BIOS settings to try, but it's not looking promising right now. I thought it might be DOA, but I have a random newer motherboard sitting on my bench right now that POSTs and the nVidia VBIOS message showed up and the VGA port on the nVidia outputs the video signal. So, although I haven't confirmed the nVidia card is working fully, it's not DOA.

Does it show up in OS as a device?
I haven't tried any external cards beside ATI Rage, but my guess would be:
1) Set "Internal VGA" to disabled
2) Set "Init Display First": PCI

Paul_V wrote on 2024-02-03, 23:38:

Does it show up in OS as a device?

No - nothing I can see to indicate the card is active on the PCI bus at all. At this point, still not sure if this problem is electrical/physical or logical/timing/software, but I'm leaning towards the former right now.

For example, I tried the pcicfg DOS command to dump all the PCI devices. Nothing shown for this card on the ISA855 mobo, but it's obviously listed on other mobos: nVidia vendor ID, VGA card.

Methodology: I dug out various other PCI motherboards, 4 of them, pre-dating and post-dating the 855 chipset - Athlon 1400, P4, Athon dual core, Celeron D - and the FX5500 worked immediately in all of them. Booted them all to DOS and used the card in benchmarks. The nVidia BIOS message always shows up before the BIOS screen. These were mostly also using Award/Pheonix Modular BIOS , which makes me assume this isn't a pure BIOS compat issue.

pcicfg spits out the PCI status register for the card on a working mobo, and I did wonder about its "66 MHz <something something>" bit being the culprit as that wasn't shown for the ATI graphics card or anything else PCI I encountered by default. So, I dug out a few other "known good" PCI cards - a 3COM ethernet and two Adaptec cards, one SATA and one SCSI. At least one of those Adaptec cards also listed the 66 MHz bit, and being drive controllers they also have their own extension BIOSes, so while I don't have other VGA cards to try, I figured if they didn't work in the ISA855 this might tell me something. So far, every other PCI card I've tried in the ISA855 has joined the PCI bus .

One of my other hunches above was "power", since I figure there's a limit to how much that ISA855 mobo will get through its two +12V connections (as I'm using the ATX 4-pin CPU connector to power it). If a mobo was a "dumb" design and the nVidia was pulling too much, I'd expect stability issues or the PSU to trip out. Inspecting the physical PCI pins on the ATI vs the nVidia, AIUI my ATI card is physically wired to report > 7W on the 2 pins used for this, while the nVidia is > 25W (the max the PCI spec apparently allowed for). I dug around online and I couldn't tell how these pins are actually used by a mobo - I wondered if the motherboard might electrically isolate the card if it required too much. No conclusion so far - I need to get my multimeter and magnifying glass out.

I even dug out Intel's 855GM design manual online and I think those pins should get tied down on the mobo side (so no clever isolation...?), but I'm a software professional, not hardware/electronics, so things from me start to get even more hand-wavy as I go down this rabbit hole. I did find the 855GM PCI reference schematics, so I may try to check how the PCI connectors are wired to see if I can find any that are obviously disconnected - it seems possible to me that the mobo designer decided not to bother connecting some pins if they're only used by a small minority of cards. That said, some signal pin not being connected currently seems unlikely to me - I assume there's not a lot of space on connectors for "maybe don't bother connecting these" pins. Something to do with power requirements seems more likely - it's currently the main way I can think of that a graphics card would differ from the other cards I've tried; a drive controller like the ones I've had would presumably also need high-ish bandwidth and exercise similar PCI transaction capabilities, though I'm just guessing.

FTR - I also tried the nVidia graphics card in both ISA855 PCI slots. I still need to try removing the ISA sound card just in case there's some kind of conflict there, though I did make sure all the PCI IRQ BIOS settings are set to auto, so it's mostly for completeness.

Paul_V wrote on 2024-02-03, 23:38:

I haven't tried any external cards beside ATI Rage, but my guess would be:
1) Set "Internal VGA" to disabled
2) Set "Init Display First": PCI

Thanks for the suggestions, but yes: first thing I had to do. My initial install was just to pull out the ATI graphics card and put in the FX5500, expecting it to "just work". I did have internal VGA off, and the mobo did its "no graphics card" beep. Turning on the built in VGA after restoring the ATI got me back to a booting system, but no closer to a working nVidia card. The BIOS is apparently not even "seeing" the nVidia card, so something very fundamental electronically is happening, or the card is being rejected early in the boot with something like PCI bus card enumeration.

Even though the ISA855 doesn't have an AGP slot, it does mention AGP aperture size in the BIOS, and I tried all those specifically because was mentioned second hand in another Vogons thread that Phil's Computer Lab once did a video where that made a difference for an FX5X00 card.

I've been through all likely (and unlikely) -looking BIOS settings that I think might even slightly affect bus timing, graphics or PCI behaviour, and not found one that makes any difference so far. I might as well not have the card in the PCI slot. Next step is the multimeter until more information surfaces or I "cut my losses" or something more important crops up in my life.

This is, so far, a fun mystery that forces me to explore things I know very little about - PCI bus electronics and firmware are closed book to me.

Aside: Last night, benchmarking the FX5500 on systems that have AGP slots allowed me to compare the FX5500 DOS performance with a couple of other old graphics cards, and a couple of built-in chipsets from newer boards. I know DOS performance wasn't an important metric by the time this card came out, but they weren't kidding when they said that this graphics chip was poor. The PCI bus is probably a large part of the bottleneck, and I know this is a knock-off, so perhaps has its own distinct issues. It does seem generally more compatible with the dos graphics benchmarks I've tried (fewer visual glitches) than some of the built-in chipsets. Those built-ins were very "post DOS" era, so I don't suppose too many chipset designers cared about DOS/BIOS graphics behaviour correctness or performance - I imagine they were all focused on the DirectX driver. Also, coming from a world where I expect built-in graphics to suck compared to discrete cards, it's odd to see things like DOS Quake perform significantly better on built-in graphics than when using a dedicated card. I assume this might have more to do with the built-in graphics being on a faster bus to the CPU and/or the fact that even sucky built-in graphics beat dedicated graphics chips from a couple of generations prior. That, and generally the focus being on 3D performance, not how fast you can push pixels to a framebuffer from the CPU.

Update on mine, the VGA went wonky. It stopped outputting anything for most of my monitors. One monitor shows text mode garbled. One monitor shows text modes fine but I had a working adlib tracker II screen res, now it comes out garbled on that monitor too. The seller is ignoring my request to take advantage of the three year warranty so I am going to just try to find a ATI rage XL as it was mentioned as working in this thread.

moon wrote on 2024-05-07, 18:22:

Update on mine, the VGA went wonky. It stopped outputting anything for most of my monitors. One monitor shows text mode garbled. One monitor shows text modes fine but I had a working adlib tracker II screen res, now it comes out garbled on that monitor too. The seller is ignoring my request to take advantage of the three year warranty so I am going to just try to find a ATI rage XL as it was mentioned as working in this thread.

You can try chipset downclocking.

1) Change "MGMT Core Frequency" in BIOS from "AUTO Max 266Mhz" to "400/266/133/200" (FSB/MEM/GFX LOW/GFX HIGH core freq)
2) Check the presence of a jumper cap (see the picture below) and remove it if present. This will change FSB speed from 133Mhz (533) to 100Mhz (400)

The reason is, this MoBo defaults to same BIOS / CMOS settings, but actual components on them differ from batch to batch.
In your case, the culprit may be GMCH. It can be 852GM, 855GM,855GME or 852GME.
Only the latter supports full bus speeds this motherboard can provide. For others default settings and 133mhz bus mean overclocking beyond stable operation and often causes aftifacts and issues, similar to what you describe.
https://www.intel.com/content/www/us/en/suppo … 8/graphics.html

I have already acquired a strange new ATI Rage XL PCI card from Amazon https://www.amazon.com/dp/B0BT222JZN that curiously all say they are compatible with "towing trailers". A purchaser does say that the PCI voltage selection pins are wrong on the card but it does boot fine on this board and it resolved the VGA issue in most cases. I'm having a hard time always figuring what is Adlib Tracker II problem or board problem, I'll be making a separate thread for that.

However I will happily try your change, I want to see if it resolves some of those other problems before I make that thread.

Anyway here is a video of it playing OPL3 module with an Orpheus II LT. The sound is just wonderful and all is right in the world for a little while.

https://youtu.be/rAoXJ-0oQ7o