tabm0de wrote on 2024-03-04, 06:47:

fluxrez wrote on 2024-03-04, 02:27:

I honestly don't see a reason the get anything more powerful running. If I understand this thing well enough, the CPU is the bottleneck. Only early 3D games are playable.

I will agree, *Llama in my view is first of all for dos and some early-late win95/early win98 gaming pc anything faster then that i would go for p3.

Suggestions in forums for all in one that do play dos to late win98 3d games has always been to have a 486 / early pentium and then a p3 whit 3dcard for more requiring old games.

Agreed but that's not going to stop me from trying. My experience with Radeons is that if the driver works the compatibility should be roughly the same.

I forgot and left the board on all night. My heart sank for a second when I realized it had been running all night with no fan but it's a tank and completely stable.

Aaron707 that I/O shield fit perfectly thank you!

The IO shield is great. I had to move the PS2 ports down about 1.3 mm on mine but it was very easy. I hadn't even used tinkercad before.

I printed this pfSense case that should work well with a 9250.

While I wait on the 9250, I have an RV100 based card but I can't tell what this connector is. Does anyone recognize it?

Dothan Burger wrote on 2024-03-05, 17:37:

I printed this pfSense case that should work well with a 9250.

While I wait on the 9250, I have an RV100 based card but I can't tell what this connector is. Does anyone recognize it?

Looks like VESA EVC or some weird P&D variant. Unusual thing.

A Radeon 9550 did not post. Just putting the card in the slot turned off the LED. I had to revert to the old card and mess with the board some to get the LED back on.

Someone mentioned trying a geforce 5200 but unless it has 64mb of RAM from what we know it shouldn't work. Because nvidia wasn't banking memory into a smaller MMIO space until the 7 series according to RLOEW's ptchnvz manual.

It's interesting that all the cards that work have PCI variants.

Dothan Burger wrote on 2024-03-06, 23:07:

A Radeon 9550 did not post. Just putting the card in the slot turned off the LED. I had to revert to the old card and mess with the board some to get the LED back on.

Someone mentioned trying a geforce 5200 but unless it has 64mb of RAM from what we know it shouldn't work. Because nvidia wasn't banking memory into a smaller MMIO space until the 7 series according to RLOEW's ptchnvz manual.

It's interesting that all the cards that work have PCI variants.

Also S3 Trio 3D/2X works, but it is not fast enough for 3d games. Currently I'm using Asus V3005, SiS 305 based card. It is quite good actually, speed is on a Geforce 2 MX level and dos compatibility seems fine to me. Haven't tested it too extensively yet. I'm limited to low profile cards because of my case, so I'm trying cards that fit. I have low profile Radeon 9200, Trident Blade 3D and Matrox G550 cards, that I might try later.

janih wrote on 2024-03-07, 11:04:

Dothan Burger wrote on 2024-03-06, 23:07:

A Radeon 9550 did not post. Just putting the card in the slot turned off the LED. I had to revert to the old card and mess with the board some to get the LED back on.

Someone mentioned trying a geforce 5200 but unless it has 64mb of RAM from what we know it shouldn't work. Because nvidia wasn't banking memory into a smaller MMIO space until the 7 series according to RLOEW's ptchnvz manual.

It's interesting that all the cards that work have PCI variants.

Also S3 Trio 3D/2X works, but it is not fast enough for 3d games. Currently I'm using Asus V3005, SiS 305 based card. It is quite good actually, speed is on a Geforce 2 MX level and dos compatibility seems fine to me. Haven't tested it too extensively yet. I'm limited to low profile cards because of my case, so I'm trying cards that fit. I have low profile Radeon 9200, Trident Blade 3D and Matrox G550 cards, that I might try later.

Dang even the Radeon 7000 I'm running isn't Geforce 2mx level. I'll be interested to hear how the other cards fair.

I want to test a geforce3 ti 200 because there was a PCI variant available.

Is there any benefit to moving that AGP slider to 2X? None of the other cards I've tested had the option.

Dothan Burger wrote on 2024-03-12, 01:25:

Is there any benefit to moving that AGP slider to 2X? None of the other cards I've tested had the option.

Now I know not to mess with the AGP settings in smartgart.

RV100 and RV280 work but not R100 and R200, the cards will install and game for a short time then crash randomly. I'm going to build a system to verify those cards are actually working properly.

Nvidia, in ~2004 had a stock of graphics cards that were natively AGP, as the PCIe standard started to become more prevalent.
They designed a PCIe to AGP bridge, named BRo2. https://en.wikipedia.org/wiki/Nvidia_BR02 Just out of curiosity, can these be sourced somewhere?

hilram wrote on 2024-03-24, 13:14:

Nvidia, in ~2004 had a stock of graphics cards that were natively AGP, as the PCIe standard started to become more prevalent.
They designed a PCIe to AGP bridge, named BRo2. https://en.wikipedia.org/wiki/Nvidia_BR02 Just out of curiosity, can these be sourced somewhere?

They used that chip both ways, 6600GT's were native PCI-E and that chip enabled AGP variants. I doubt they have any unused supply. Probably a 6200 AGP would be the cheapest way to attain one.

A PCI or PCI-E slot would have probably been a more compatible solution for this board. I don't think AGP was ever really a completely fleshed out standard.

What were you thinking?

I'm sure there are more interesting projects to build with that chip 😀

Such as a way to connect any AGP cards into a modern system via PCIE
Or plug a newer PCIe card into an old AGP system (Geforce 8, RTX4090, whatever you prefer)

Deksor wrote on 2024-03-24, 16:44:

I'm sure there are more interesting projects to build with that chip 😀

Such as a way to connect any AGP cards into a modern system via PCIE
Or plug a newer PCIe card into an old AGP system (Geforce 8, RTX4090, whatever you prefer)

The BGA stencils look to be available. It would just be so tedious to remove the HSI from working cards. Add to that the absurd eBay pricing right now, I don't see a viable way to do something like that.

Dothan Burger wrote on 2024-03-24, 16:25:

They used that chip both ways, 6600GT's were native PCI-E and that chip enabled AGP variants. I doubt they have any unused supp […]
Show full quote

hilram wrote on 2024-03-24, 13:14:

Nvidia, in ~2004 had a stock of graphics cards that were natively AGP, as the PCIe standard started to become more prevalent.
They designed a PCIe to AGP bridge, named BRo2. https://en.wikipedia.org/wiki/Nvidia_BR02 Just out of curiosity, can these be sourced somewhere?

They used that chip both ways, 6600GT's were native PCI-E and that chip enabled AGP variants. I doubt they have any unused supply. Probably a 6200 AGP would be the cheapest way to attain one.

A PCI or PCI-E slot would have probably been a more compatible solution for this board. I don't think AGP was ever really a completely fleshed out standard.

What were you thinking?

I was thinking of a way to enable full AGP compatibility from the PCIe to AGP. With full subset support, and also to bump it from AGP 1x to AGP 2x or possibly AGP 4x support. AGP is never really just AGP. I used to work in computer store retail from 1995 to 2005, and the transition from AGP 2x to 4x to 8x was in no way straightforward.

@hilram

Eivind got AGP 1x working on a PCIe to PCI bridge chip with his ITX llama project.

Duffman wrote on 2024-03-25, 06:55:

@hilram

Eivind got AGP 1x working on a PCIe to PCI bridge chip with his ITX llama project.

Think he meant supporting features like AGP fast writes and AGP texture acceleration and of course, higher levels like 2x / 4x.

I could be wrong, but it seems like a lot of AGP features are software driven. Like code in the bios for AGP aperture and such.

Alright, so I was able to pack everything (ITX llama, Voodoo 3000 and a SATA) in the mini ITX case (bought off ebay and attached the picture).
To do that, I had to use an AGP riser from MODDIY. But man, that was then the nightmare began. With the riser I was unable to get the card working. Then I looked into the signal condition that was causing the issue. So, I shortened the cable to the size that I wanted and soldered all the wires back. That did not help of course and I still did not have the card working. Then I cleaned the cable to have the solder residue removed. That made the card work but it was not stable. Then I wrapped the cable with copper tape (also between the the ribbon cables) and finished it off with Kapton tape. That did the magic and everything is working very well. I had build a custom IO shield and have everything nicely put together. I have two Noctua fans ready to be placed to make the entire thing look amazing. Here is a sneak peek.

Dothan Burger wrote on 2024-03-25, 16:43:

Duffman wrote on 2024-03-25, 06:55:

@hilram

Eivind got AGP 1x working on a PCIe to PCI bridge chip with his ITX llama project.

Think he meant supporting features like AGP fast writes and AGP texture acceleration and of course, higher levels like 2x / 4x.

I could be wrong, but it seems like a lot of AGP features are software driven. Like code in the bios for AGP aperture and such.

Correct.
I believe the reason that Voodoo-cards, and old ATI-cards like Rage128 works without problems on this (ITX Llama) AGP-slot, is that none of these utilize any of the AGP features that AGP provides. These are essentially PCI chips re-wired to support the AGP connector, to take advantage of AGP 1x's double bandwith of 266 MBps compared to PCI's 133MBps.

ahmadexp wrote on 2024-03-27, 06:00:

Alright, so I was able to pack everything (ITX llama, Voodoo 3000 and a SATA) in the mini ITX case (bought off ebay and attached the picture).

That's awesome! I've tried AGP risers myself, never could get them to work. Looks like there might have been too much signal interference, as you pointed out.

hilram wrote on 2024-03-27, 09:56:

I believe the reason that Voodoo-cards, and old ATI-cards like Rage128 works without problems on this (ITX Llama) AGP-slot, is that none of these utilize any of the AGP features that AGP provides. These are essentially PCI chips re-wired to support the AGP connector, to take advantage of AGP 1x's double bandwith of 266 MBps compared to PCI's 133MBps.

AGP is a superset of PCI, therefore as long as the graphics driver doesn't assume or require AGP-only features, PCI communication works fine over both PCIe, PCI and AGP.
Obviously to go between PCIe and PCI, you'll have to do a signal translation (which is what the PI7C9X118 bridge chip does), but the software protocol is still PCI.

For all of you who longs for a faster graphics card solution, I don't really think we'd gain much even if the Vortex86EX natively supported AGP because the CPU is just too weak.
A simpler solution than mucking about with fancy bridges would probably be to just expose the single x1 gen.1 PCIe lane directly as a PCIe slot and put a more modern card in that. That however (depending on the card) might require drivers that might be too new for Win98, and the card would need to function with just a single lane - not sure if that's common or not.

I suppose early Nvidia PCIe cards might work fine, but again we're back to being CPU-throttled.

If DMP gets to releasing their EX3, I'm looking forward to seeing what it can do. I know it has a slightly beefier PCIe setup planned.