Question: How powerful of a graphics card will work in a beige Power Macintosh G3?

[LightBulbFun]

Open Firmware 3 1.1f4 (B&W G3)​

: $R BRpatch ;
: & get-token drop ;
defer mr 8BA & to mr
: m5 { m } dup 1 1c << mr 1 1c << + m 4+ ! 0 ;
8E3 & dup 3c + ' m5 44 + swap $R a8 + ' m5 c + $R
1e 8E6 & 1c + code!

F*ck yeah! pleased to report after knuckling down and manually typing it out into my G3 blue and white im very pleased to say it works perfectly!

and has gotten the Quadro FX 4500 to work with my Janky PCI to PCIe setup in it (I must of messed something out when I initially tried this with my FW800 MDD and its patch)

its completely bottlenecked to buggery but its so fun to see such a powerful video card working perfectly with such a system

 

[flyproductions]

Hope you are able to snag it, would be cool to see how fast this AGP system really could be.

Sadly always the same with these boxes! At a closer look to one of his pictures i saw something a bit strange in the lower rigtht corner of the CPU-cover. So i asked the guy to take a look behind and if possible some pictures. And there it was: some leakage at ohne of the tubes! But he said he had it running today, so most likely no damage so far. I told him not to fire it up again until this is fixed. And also, that i would still be willing to pay 50 for it. But, considering it will cost me another 50 for gas to get it, no more. So he wants to wait another week, if somebody is willing to pay more and if that doesn't happen PM me again.

So no plug-and-play for the fastest AGP G5! 😟 But it could be an easy fix. I have this one pum LCS of a quad here. Removed the CPU-boards today and it looks absolutely perfect. Pump is working. Only seems to have some air in it. So might need a refill, which shouldn't be a problem. Only question is: Will it fit?

 

[DearthnVader]

I don't know what's going on.

I wonder if it's the conflicting information like the sub-vendor and sub-system-id in the PCI header?

We know OF reads this information from the bios, even if it is x86 and starts to build some .properties base on this.

 

[DearthnVader]

I wonder if it's the conflicting information like the sub-vendor and sub-system-id in the PCI header?

We know OF reads this information from the bios, even if it is x86 and starts to build some .properties base on this.

That was the issue, changing 4238 02b4 to de10 4d00 in the PC bios part of the PC/Mac .rom fixed it.

This should correspond to the subvendor-id and subsystem-id( 0x3842 0xb402 ).

So the script should change them in the PC Bois part of the ROM to match what they are in the Mac ROM, or the ROM's made by the script won't work with OF.



The card now "functions" in a G5 and a QS, but I don't have a BAR1 patch for these Macs.

The QS has a bug where you get no input in the telnet session, and the G5 seems to hang when trying to read/dump the BootROM?

Didn't @dosdude1 make a modified version of linux FlashROM that could dump the 1MB BootROM of a PowerMac?

The GF6200's do not function in a MDD with a Mac FCODE ROM on them, neither the 256MB nor the 512MB PCI, and I think this was an issue with the flashed AGP cards too. I don't think any GF6200 ever functioned in a MDD.

Maybe this is something we can debug and fix?

Loading the FCODE ROM from a file does work in the MDD, and the cards do function properly, it's just if you flash them with the FCode ROM's they pull the "no show" on the PCI/AGP bus, we'll have to look into it.

Also, @joevt would you patch the Mac512.rom for Old World support, please?

Linked in post #646 .

 

[joevt]

The QS has a bug where you get no input in the telnet session, and the G5 seems to hang when trying to read/dump the BootROM?

We don't need input. Just output.
" enet:telnet,192.168.0.111" io dumpbytes

Show the steps you use for the G5 and show the telnet output. If you can't dump the rom then dump the dictionary.

 

[joevt]

Also, @joevt would you patch the Mac512.rom for Old World support, please?

Test7/1ROMs/Mac512.rom

 

[DearthnVader]

Test7/1ROMs/Mac512.rom

Opps!

Take care when you plug your PCI cards into a PCI-E to PCI bridge that it's properly plugged in.

I killed the 512MB card.

I ordered another, project on hold on my part until I get the new card.

 

[flyproductions]

I killed the 512MB card.

Ohh noooo!

Was it this PNY 512 MB PCI with the strange EEPROM? Don't know how it is in the US (or where you live). But over here in Germany would be nearly impossible to get a replacement. 😢

 

[DearthnVader]

Ohh noooo!

Was it this PNY 512 MB PCI with the strange EEPROM? Don't know how it is in the US (or where you live). But over here in Germany would be nearly impossible to get a replacement. 😢

EVGA NVIDIA GeForce 6200 (512P1N402LR), fairly common here on US eBay.

Hoping it has the same 256k EEPROM, as we have made a Mac/PC ROM that allows the card to function in both.

I was going to do some testing in my Beige to see if there was any advantage to the extra 256MB of VRAM over the 256MB card I have in it now.

Likely the PCI Bus is just going to hamper things to those ends, but I was going to see if there was any use case at all that favored the 512MB card.

Sadly the 6200 couldn't drive Dual-Link DVI, but at least we unlocked Core Image for Old World PCI Mac's, so that's something.

Open a few images in Core Image FunHouse and watch the VRAM usage.........

I was also going to test what happens with the 512MB BAR1 if the system has less than 512MB of RAM. I'll have to see where I misplaced the manual to the EVGA 512MB card to see if it list the minimum system RAM for a PC.

I've never seen a PCI or AGP card with more than 512MB of VRAM, I think there is a limitation there, tho when a tested the card with PCI Passthough to Qemu-system-ppc64 the nouveau drivers reported 1024MB of VRAM and a 1024MB BAR1. Tho that was just likely a bug in the way that nouveau reads the Soft Straps.

The card did function proper with Debian PPCLE on the TTY but when trying to get X running there were issues. I had the same issues with X on the GF6600 PCI-E with Qemu-ppc64 a few years back.

Not sure if I just don't have Xorg configured proper or it's an issue with PCI Passthrough.

 

[Amethyst1]

I've never seen a PCI or AGP card with more than 512MB of VRAM, [...]

ATI Radeon HD 4670 AGP

 

[flyproductions]

ATI Radeon HD 4670 AGP

Bet, @joevt could get this to work in a G4 too! 😉

 

[flyproductions]

EVGA NVIDIA GeForce 6200 (512P1N402LR), fairly common here on US eBay.

You seem to be in a better situation over there. More recent PCI-cards have vanished nearly completely. And if they appear, it's usually with "collectors"-price-tags.

I was going to do some testing in my Beige to see if there was any advantage to the extra 256MB of VRAM over the 256MB card I have in it now.

Likely the PCI Bus is just going to hamper things to those ends, but I was going to see if there was any use case at all that favored the 512MB card.

I doubt there would be any noticeable advantage. Even 256MB ishould be a massive amount of VRAM for a machine which only supports up to 768MB of DRAM.

Sadly the 6200 couldn't drive Dual-Link DVI, but at least we unlocked Core Image for Old World PCI Mac's, so that's something.

Yes, this is really sad! A 30"-capable Cube would be a nice thing to have.

 

[joevt]

I was also going to test what happens with the 512MB BAR1 if the system has less than 512MB of RAM. I'll have to see where I misplaced the manual to the EVGA 512MB card to see if it list the minimum system RAM for a PC.

BAR is separate from RAM but you probably know that and are just wondering if BAR is allowed to be bigger than RAM. I don't see why not. The problem would be not that the BAR is too big. It would be that your Mac has low memory for Mac OS X.

ATI Radeon HD 4670 AGP

But does it have an Open Firmware driver? Or Mac OS X PowerPC driver?

The GF6200's do not function in a MDD with a Mac FCODE ROM on them, neither the 256MB nor the 512MB PCI, and I think this was an issue with the flashed AGP cards too. I don't think any GF6200 ever functioned in a MDD.

Maybe this is something we can debug and fix?

Loading the FCODE ROM from a file does work in the MDD, and the cards do function properly, it's just if you flash them with the FCode ROM's they pull the "no show" on the PCI/AGP bus, we'll have to look into it.

Maybe the straps (bytes 0x50-0x104) of the fcode ROM don't match with what's in the GPU's ROM? When you load the fcode from file, the GPU is still using straps from the GPU's ROM.
Unless the GPU's ROM has the fcode ROM but you're disabling it with pci-probe-list or whatever?

 

[Amethyst1]

Or Mac OS X PowerPC driver?

Not to my knowledge.

Sadly the 6200 couldn't drive Dual-Link DVI,

Yes, this is really sad! A 30"-capable Cube would be a nice thing to have.

Dual-link DVI is not required for 2560×1600... if you and your 30" monitor can live with 38 Hz refresh rate.

 

[flyproductions]

Dual-link DVI is not required for 2560×1600... if you and your 30" monitor can live with 38 Hz refresh rate.

Would happily do! But as far as i remember OS X didn't offer this combination in the settings. It just switches to 1280 x 800 if a 30" is attached to a Single Link card. Maybe SwitchRes works?

I'll test later.

 

[Amethyst1]

Would happily do! But as far as i remember OS X didn't offer this combination in the settings. It just switches to 1280 x 800 if a 30" is attached to a Sngle Link card. Maybe SwitchRes works?

You will need SwitchResX to add a custom timing, and the monitor needs to accept that custom timing. I'm not confident Apple's 30" does.

If you use an older version of SwitchResX that can't calculate CVT-RBv2 timings, the maximum refresh rate you can get via single-link DVI is 37.2 Hz (at 164.5 MHz pixel clock using CVT-RB timings).

 

[flyproductions]

...and the monitor needs to accept that custom timing. I'm not confident Apple's 30" does.

I'm not 100% sure, but i seem to remember this in fact beeing the problem: These Apple alu things not accepting anything but their actual native frequency.

 

[Amethyst1]

I'm not 100% sure, but i seem to remember this in fact beeing the problem: These Apple alu things not accepting anything but their actual native frequency.

Yeah. But other 30” monitors may be more forgiving.

 

[flyproductions]

Well, i tested: Installed SwitchResX, added the custom resolution i wanted, saved and rebooted. And this is what i get:

Not installed! All advice i could find online for this problem is "Disable SIP". Disable SIP on a PPC? 🤔

 

[joevt]

Well, i tested: Installed SwitchResX, added the custom resolution i wanted, saved and rebooted. And this is what i get:

View attachment 2102739

Not installed! All advice i could find online for this problem is "Disable SIP". Disable SIP on a PPC? 🤔

Apple 30" displays only support 2560x1600 60Hz and 1280x800 60Hz. I tried now with mine and 30Hz only produces black screen.

 

[Amethyst1]

Not installed! All advice i could find online for this problem is "Disable SIP". Disable SIP on a PPC? 🤔

I've run into this as well on some of my setups. Sometimes, it was "fixed" by registering SwitchResX.

Apple 30" displays only support 2560x1600 60Hz and 1280x800 60Hz. I tried now with mine and 30Hz only produces black screen.

As I feared. I wonder if more modern 30"s (e.g. Dell U3011 or U3014) do accept lower refresh rates at their native resolution.

 

[flyproductions]

Apple 30" displays only support 2560x1600 60Hz and 1280x800 60Hz. I tried now with mine and 30Hz only produces black screen.

Yes! This confirmes what i had to experience yesterday.

So no 30" @ full resolution on a G4 Cube...

...unless you cramp a 6600 GT into it, which, if you get it in at all with VRM-replacement and all that stuff and get this one of 50 boots it fully boots on a 2x AGP, will never run inside the stock enclosure with the power the Cube's powerbrick / stock VRM can provide.

Edit:
This could be another "once true, but not anymore"-thing, that these 6600 GTs in 95% of the boots hang at system boot / switching from grey to blue screen while in the last 5% boot fine in 2x AGP systems like the Sawtooth and the Cube. We all have seen miracles happen in the last two months! 😃

 

[LightBulbFun]

As I feared. I wonder if more modern 30"s (e.g. Dell U3011 or U3014) do accept lower refresh rates at their native resolution.

for what its worth my Dell U2715H will happily take 2560x1440 @ 30Hz through its HDMI port from a Radeon 9200

obviously its not an EDID exposed option sadly, but it will happily sync to it once forced with SwitchResX

...unless you cramp a 6600 GT into it, which, if you get it in at all with VRM-replacement and all that stuff and get this one of 50 boots it fully boots on a 2x AGP, will never run inside the stock enclosure with the power the Cube's powerbrick / stock VRM can provide.

if all you care about is 30 inch Cinema display-ness, then surely finicky boot issues aside, you could take a dual link 6600 GT, and under-volt/under-clock the crap out of it it to get its power and thermals in check so it can go into a stock G4 cube without cooking itself/the cube?

even when severally kneecap'ed in that manner, its probably still better then a 6200!

 

[flyproductions]

if all you care about is 30 inch Cinema display-ness,...

It's not that i "needed" this. But it's more or less the last thing which does not work with the maxed-out-Cube.

But, even with this 2x AGP-thing out of the way, the 6600 would not be an option. First i have no room left anymore to replace the VRM, as where it would have to go, now sits the pocket router to give reasonable WIFI-speed. Also, even undervolted/-clocked to the max, it wold need a massive active cooling on it, which there is also no room for inside the case. The Radeon 9700/-800, GeForce 66-/6800 class of vidcards are just no match to the Cube. An sadly no low-power, low-profile, 2x AGP-keyed-cards mor "modern" than the 6200 do exist. So most likely GF 6200 is the end of that road.

 

[DearthnVader]

Test7/1ROMs/Mac512.rom

Doesn't seem to work with the BAR1 patch for the Beige 2.0f1. I get the same screen corruption I get without the BAR1 patch.

It maybe I have just mistyped the patch?

Here is with the nvram patch applied:

I'm not sure line 5 of the patch, is that ' before m5?

And without the patch: