the NVRAM patch is just to allow 512MB VRAM cards to work, they still need a correctly adjusted fCode ROM as I understand it
speaking of which guess im digging the Quadro FX 4500 back out of the G5 Quad I put it away in LOL
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Question: How powerful of a graphics card will work in a beige Power Macintosh G3?
- Thread starter LeoI07
- WikiPost WikiPost
- Start date
- Sort by reaction score
- Status
the NVRAM patch is just to allow 512MB VRAM cards to work, they still need a correctly adjusted fCode ROM as I understand it
speaking of which guess im digging the Quadro FX 4500 back out of the G5 Quad I put it away in LOL
So i added the patch to the nvramrc in the Sawtooth/OF 3, but the card still shows 256MB in System Info.
Likely the fCode-ROM on the card doesn't have "512MB enabled". But this is where i'm stuck. I know, i once had a ROM that showed 512 at least with a safe boot, but refused to boot normal. But this is very long ago and i don't know where i lost it.
Also with the ROM, i made with the ROM-Maker yesterday, the card showed up in OS9's System Profiler with 512MB of V-Ram, when booted to OS9 with an additional PCI-card. But it gave no signal to the display at all.
heres a OF Directory dump from my from my PowerMac3,6 1.42Ghz FW800 MDD hopefully it helps
I Dont THINK this word is present with the G3 Blue and White either, just trying it at an OF prompt gives me an unknown word, although no telnet package on that so I cant easily make a dump like I have just done with the PM3,6 sadly
TBH im impressed the 512MB Safe Boot ROM even produced an OpenFirmware image, going by how my 512MB Quadro just deadlocks the system if anything is plugged into its DVI ports
but yeah if you managed to find that ROM again it would be very interesting to see how it played with the NVRAMRC patches
It's like an and/or mask. So you can have the same strap, but it will not be the same in truth if it has different straps on the resisters.
Take the resister value and add this to it( append ), just read the resistor value( strap nothing ), ignore the value of the resistor and use the value in the strap alone.
The last one, I think was what I did to all the cards I flashed. I set the strap for the VRAM size and the strap for the timing table to ignore the resister value and only use the value I set in the soft strap( NVSTRAP ).
I just noticed 2 things
1: @DearthnVader already provided an MDD Directory dump, whoops my bad!
but
2: while studying my own dump I noticed that one of the words avliable on the PCI-Bridge I had installed is
alloc-mem-addr without the brackets mind, I dont know if this is the word your looking for if your look for specifically (alloc-mem-addr) as one of the general available words rather then a device specific one?
The card may need adjustments as well. Basically, you need a card that will work as 512MB in a Power Mac G5, otherwise it won't work in a pre G5.
There may be other areas that need changing.
heres a OF Directory dump from my from my PowerMac3,6 1.42Ghz FW800 MDD hopefully it helps
OF Dictionary dump is not just the list of words. It's everything that's stored in the Open Firmware dictionary (the compiled code for each word, devices, and properties). The process is described at
#137 and an example is described at #159 .
I have all the info I need for B&W G3 (I have one) and I'll work on a patch for that.
They are different words.
(alloc-mem-addr) is in the main list of words. The main pci root device (grackle/macio on Beige G3 with devalias pci) has a alloc-mem-addr word that calls (alloc-mem-addr).Here is the info from Beige G3 Open Firmware 2.0f1.
Code:
0 > dev pci ok
0 > see alloc-mem-addr
: alloc-mem-addr
my-addrs ^-7D86A8 ; ok
0 > -7D86A8 (see)
: ^-7D86A8
{1} ^-7F0738 max 1! 0@ ^-7D96C8 @ 3! 1@ ^-7F0F50 0@ ^-7D8700 2! 0@
^-7D96B0 @ 2@ 1@ + u< if
^-7D9280 ^-7D9310 ['] claim-real catch if
2drop 3@ 0@ ^-7D96C8 ! 0
else
else
2! 2@ ^-7D9280 ^-7D9438 2@ ^-7D9280 + 0@ ^-7D96B0 !
then
then
2@ 1@ + 0@ ^-7D96C8 ! 2@ ; ok
0 >Notice that many of the words are unnamed in 2.0f1, unlike 2.4.
-7D86A8 is for (alloc-mem-addr).my-addrs keeps track of the next and max memory addresses.
Code:
0 > my-addrs d# 44 dump
FF833298: 85 00 00 00 FD 00 00 00 00 00 04 00 00 01 00 00 :................:
FF8332A8: 80 00 00 00 FE 00 00 00 FF 8D AF 80 FF 8D AF 00 :................:
FF8332B8: FF 8D B7 40 FF 83 35 A0 FF 83 36 30 :...@..5...60: ok
0 >
Code:
/n field >as.next-mem-addr
/n field >as.mem-addr-max
/n field >as.next-io-addr
/n field >as.next-bigio-addr
/n field >as.mem-base
/n field >as.io-base
/n field >as.mem-addrs
/n field >as.io-addrs
/n field >as.my-pkgs
/n field >as.'alloc-mem-addr
/n field >as.'alloc-io-addrmy-addrs.>as.mem-addr-max is set to cmembase/grackle-base:80000000 + 7d000000 (2000 MB) = fd000000. That means it never experiences the bug that Open Firmware 2.4 and Open Firmware 3 have. That might explain why it contains other incorrect code (I think the claim-real call is missing the address value on the stack) - the incorrect code never gets executed so the bug was not discovered.Incorrect code from Open Firmware 2.0f1:
Code:
if
\
\
\
const_951_h#10000000 \ h#10000000
const_94b_1c \ h#10000000 1c \ there's only 2 numbers here. where's the address?
['] claim-real \ h#10000000 1c 'claim-real.
catch \ error | ... falseThe correct code for this part from Open Firmware 2.4:
Code:
if
_my-addrs \ _my-addrs
>as.mem-addr-max \ _my-addrs.>as.mem-addr-max
@ \ @_my-addrs.>as.mem-addr-max
h#10000000 \ @_my-addrs.>as.mem-addr-max h#10000000
0 \ @_my-addrs.>as.mem-addr-max h#10000000 0
['] claim-real \ @_my-addrs.>as.mem-addr-max h#10000000 0 'claim-real
catch \ error | ... falseAre there examples of PCI cards that work on Open Firmware 2.0f1 (early Beige G3) that don't work on Open Firmware 2.4 (later beige G3)?
Open Firmware 2.4 initially allocates only 256 MB so
my-addrs.>as.mem-addr-max is 80000000 + 10000000 = 90000000. But a 512 MB BAR must begin at a multiple of 512MB (20000000) so (alloc-mem-addr) needs to loop 3 times (it allocates 256 MB at a time) - once to get from 90000000 to a0000000, then two more times to allocate the 512MB which should leave _my-addrs.>as.mem-addr-max set at c0000000.Since the incorrect code is never executed and the loop patch is not required, I don't think we need to do anything for Open Firmware 2.0f1 unless there's a problem with it reserving 2000 MB at the start.
Last edited:
Somebody remind me, did the GF6200 AGP have issues not working in a MDD?
I know some GF cards flashed to Mac had issues in the MDD, I just don't know if the GF6200 was one of them?
I have a few QuickSilver G4's, but they have a bug in OF with Telnet. You can make the connection, but you have no input in the Telnet session on the remote terminal.
I've never managed to figure out how to bypass this bug, every once in a while you get input, but there doesn't seem to be in rhyme or reason to it that I can find.
I also have a SP G5 1.6Ghz( PCI ).
Loading the ROM made from the ROM Maker in the G5 produced the VRAM,size of 0x20000000( 512MB ), but I get garbled display from the card. Som the ROM isn't quite correct.
It's hard to remember from work I did in 2004/5/6, but I do remember that with the ROM from the ROM Maker, the only two changes I made to the pre-made ROM Maker ROM was to the strap for the timing table( 0-7 ) and the strap for the VRAM,size.
That made the card work prefect in my 2004 G5 1.8Ghz ( PCI ), but I only ever made two of the cards. I think they were XFX cards, but one of them refuse to work with the Cube when I sent it to the user. The other one I never tested outside of my G4.
Both cards were AGP, and no one ever asked for the 512MB PCI cards to be flashed by me, I don't think they even made any at that time. The 256MB cards worked perfect in Tiger( 10.4.3+ ) in the Yikes and B&W.
I don't remember the exact result in the G5 with the 512MB AGP cards before I overwrote the VRAM,size strap, only that the cards need it to work correct. Could have been with the straps copied over from the PC ROM without modification only 256MB worked, or could have been the garbled video I am seeing now I n the G5 with the 512MB PCI card.
I really just don't recall.
I know some GF cards flashed to Mac had issues in the MDD, I just don't know if the GF6200 was one of them?
I have a few QuickSilver G4's, but they have a bug in OF with Telnet. You can make the connection, but you have no input in the Telnet session on the remote terminal.
I've never managed to figure out how to bypass this bug, every once in a while you get input, but there doesn't seem to be in rhyme or reason to it that I can find.
I also have a SP G5 1.6Ghz( PCI ).
Loading the ROM made from the ROM Maker in the G5 produced the VRAM,size of 0x20000000( 512MB ), but I get garbled display from the card. Som the ROM isn't quite correct.
It's hard to remember from work I did in 2004/5/6, but I do remember that with the ROM from the ROM Maker, the only two changes I made to the pre-made ROM Maker ROM was to the strap for the timing table( 0-7 ) and the strap for the VRAM,size.
That made the card work prefect in my 2004 G5 1.8Ghz ( PCI ), but I only ever made two of the cards. I think they were XFX cards, but one of them refuse to work with the Cube when I sent it to the user. The other one I never tested outside of my G4.
Both cards were AGP, and no one ever asked for the 512MB PCI cards to be flashed by me, I don't think they even made any at that time. The 256MB cards worked perfect in Tiger( 10.4.3+ ) in the Yikes and B&W.
I don't remember the exact result in the G5 with the 512MB AGP cards before I overwrote the VRAM,size strap, only that the cards need it to work correct. Could have been with the straps copied over from the PC ROM without modification only 256MB worked, or could have been the garbled video I am seeing now I n the G5 with the 512MB PCI card.
I really just don't recall.
attached is my attempt at making an OF Directory of my FW800 MDD, im not sure if this completed successfully or not
it did not just end at an OK prompt, once it got to the end of what you see in the text document, after the initial OK you see at the end of the text file it then looked it returned to an OF prompt but was replaying each line into it if that makes sense?
so this could be a load of garbage!
So this has been complete misunderstanding on my side!
I have never seen a card (AGP or PCI) work in a PowerPC Mac fully booted with more than 256MB! Not even in a G5. All i said concerning a card showing 512 in safe boot has been in a G5! And i tend to think now, it was only the Gainward Bliss, that did. For this i am 100% sure, i had such a ROM. No so sure for the 6200.
But, as far as i remember, there where very few edits to be made. No more than 6 or 8 bytes. So at this point, the first (and may be bigger) problem would be to get a card working with 512MB at all.
So question is, is anybody here at the moment, having any (non PCIe!) card, meeting this condition:
Working with full 512MB in a Power Mac G5? Would be glad to hear of and even more to test such a ROM, if it happens to be a 6200.
Last edited:
and heres a BootROM dump via your text guide, from my MDD both as a .ROM and .TXT file, I think this one did work!
but you will have to confirm that
thats a good point you make the 512MB Gainward never worked even in a G5, despite MVC's (Rubytuesday was his user name back then IIRC?) best efforts, as you say it only ever worked in Safe Mode
although its worth noting @joevt specifically checked for 512MB BAR support on a G5 Quad a PCIe machine, rather then an older PCI/AGP Based G5, maybe its one of the revisions apple made specifically for the PCIe G5?
Has running 512MB-cards (like the Quadro 4500 or 7800 GTX) in a PCIe-G5 ever been a problem? As far as i remember not.
Sorry for editing after you answered! ☺️
Sure the Quadro works with it's full V-Ram. Also the 7800 GTX does. And so should do any other 512MB-PCIe-card, as long as there is some fCode-ROM available for it.
But this doesn't help much for the AGP/PCI cards. For those, thanks for confirming my memories: That's what i remember too! That not even Rubytuesday (in this case more Arti!) got the Bliss to work with more than 256MB. Also i have a bunch of ROMs for the Gainward from the Aquamac-Site. And also none of them got the 512MB showing up. So as far as i know, it's just a no go.
Last edited:
So just booted the Sawtooth to OS 9 with the PNY-card, and guess what...
Don't know, if the V-Ram can really be „used“ in OS 9 or is just shown in the Profiler...
...or if it's just the Mac, trying to ask us something: Do we really „need“ beige G3s, having more V- than DRAM? 😝
*For sure, it's not a Club3D but a PNY. This is just what i once did before to the the ROM i used, to check if this nameing can be randomly altered as long as the length is kept.
Don't know, if the V-Ram can really be „used“ in OS 9 or is just shown in the Profiler...
...or if it's just the Mac, trying to ask us something: Do we really „need“ beige G3s, having more V- than DRAM? 😝
*For sure, it's not a Club3D but a PNY. This is just what i once did before to the the ROM i used, to check if this nameing can be randomly altered as long as the length is kept.
Gonna poke in here since it might help me out, especially in the sea of low-level technical discussion.
I have this Radeon 9250 sitting around. What's necessary to do to throw it into my Beige? First page says "no shot" but there are subsequent pages that indicate "there's a chance", and even deeper I see a weird solder job to disable half of the VRAM, but even later (recently) there's talk of 512MB cards and I'm just lost.
I have this Radeon 9250 sitting around. What's necessary to do to throw it into my Beige? First page says "no shot" but there are subsequent pages that indicate "there's a chance", and even deeper I see a weird solder job to disable half of the VRAM, but even later (recently) there's talk of 512MB cards and I'm just lost.
Your FW800 has Apple PowerMac3,6 4.6.0f1 BootROM built on 02/20/03 at 13:52:27heres a OF Directory dump from my from my PowerMac3,6 1.42Ghz FW800 MDD hopefully it helps
It's a good dump with everything from
@startvec (the first thing in the dictionary) to dumpbytes (the last thing you added in the dictionary) captured.You only included the txt but that's all I needed. It looks like a good dump but I need to figure out how to get the device fcode images that exist after the main fcode image.
Your MDD has Apple PowerMac3,6 4.4.8f2 BootROM built on 09/30/02 at 10:24:31
Since it's different than FW800, I'll be wanting the dictionary dump and rom dump to compare and to create the patch.
Maybe the PCIe cards would work with a PCI to PCIe riser in PCI Macs (with ROM modifications for problems like we found in Nvidia 6200).
I'll try looking for vram sizing stuff in the 6200 OF code to see how it can enable 512MB after I've done making the nvramrc patches.
What Mac model and Mac bootrom version and Mac OS X version does it work with?
It has the
(alloc-mem-addr) name but it is marked as headerless which means it is not visible while fcode-debug? is false.What does the following show?
Code:
' (alloc-mem-addr) (see)
180a get-token drop (see)Something went wrong with the dictionary dump:
The Mac appears to have crashed at the end.
Where were the directions on dumping the ROM?
thats exactly what I Have been doing in my PowerMac G4 Fun thread
PowerMac G4 Fun! Part 2!
things are still waiting their turn :) when I do have updates to post dont worry ill post em :) I just have very limited space, so when I move onto something else other projects are forced to take a backseat (its for example why I have not done any soldering projects in a fair while as that...
forums.macrumors.com
GeForce 6600 and ATI Radeon X1900 work, but a 512MB Quadro FX 4500 does not
the device when headless will load its fCode ROM and build properties just fine (in a NWR Mac), but if you try and boot with any display plugged in the system deadlocks (although the G3 BW will at least output some garbage before freezing)
if you boot headless into OS X the NVIDIA drivers just panic the system
I have wondered given the 6600 works nicely with the G3 Blue and white if it too would work in a G3 Beige with a suitably edited ROM
but the G3 Beige wont even boot with an x86 PCIe GPU, let alone any fCode one
Last edited:
@joevt I think I got the dictionary dump correct this time:
BTW did you ever figure how to decompress a 1MB New World ROM?
There are a few FCODE ROM's for video cards only contained in the BootROM I'd like to be able to extract.
BTW did you ever figure how to decompress a 1MB New World ROM?
There are a few FCODE ROM's for video cards only contained in the BootROM I'd like to be able to extract.
Maybe a buggy telnet?
In the text file "Dumping the New World Rom.txt" of the attachment at #137
Why is the stack starting at 2 instead of 0 ?
2 > : dumpbytes ( start size ) okLooks like this is a good dump. It has everything up to the end of
dumpbytes which is the last word you added to the dictionary.Yes. It looks like every rom does it slightly differently. I find the parts (one for early new world Macs, multiple for later new world starting from G4 MDD) using offset and length in the header, extract the parts, use a lzss command compiled from Apple Open Source code to do the decompression, catenate the parts together (if there's more than one), then pass it through my DumpMacRom.sh script for disassembly of the precompiled part, detokeniing for the fcode images, and converting the detokenized fcode images to forth.
I catenate the parts together because some of the fcode images may use words defined in other fcode images. My detokenzer goes through all the fcode images and remembers words from each.
from the same FW800 MDD I get
Code:
0 > ' (alloc-mem-addr) (see)
(alloc-mem-addr), unknown word
ok
0 > 180a get-token drop (see)
:
^ff87.50a1
{1} ^ff85.207f max {1!} {0@} ^ff87.398a @ {3!} {1@} ^ff85.638d {0@}
^ff87.5039 {2!} {0@} ^ff87.39b9 @ {2@} {1@} + u< if
{0@} ^ff87.39b9 @ ^ff87.408b 0 ['] claim-real catch if
" catch occured" ff875088 ^ff85.cf0d 3drop {3@} {0@} ^ff87.398a ! 0
exit else
{2!} {2@} ^ff87.408b ^ff87.3e52 {2@} ^ff87.408b + {0@} ^ff87.39b9 !
then
then
{2@} {1@} + {0@} ^ff87.398a ! {2@} ; ok
0 >should I set fcode-debug? to true and re-run the commands, or is this what your looking for?
now thats interesting! so is that outputting an image in OpenFirmware etc or is that a via VNC or such?
That's good. The
get-token method should work for the patching. Note that Open Firmware since MDD can use 4 digit fcode numbers.The code looks like a cross between the Open Firmware 2.4/Sawtooth/Gigabit version and the Open Firmware 4 (Quad G5) version. It is missing the loop that Open Firmware 4 has and that my patch will add.
The
see and (see) command doesn't work well with patched words - but it should at least show that a change was made and approximately where it was made.You can try if you're curious. I don't think it will add any more info for myself. At least you'll see an increase in the number of words listed by the
words command. Also, it will allow ' to find words that were previously invisible so ' (alloc-mem-addr) (see) will work.
I don't think I understand. I haven't even tried putting the card in. I got it out of a box of random graphics cards a long, long time ago. Unless you mean the Mac? It's a bummy v1 ROM model found on the first revision Beige MTs.
No. It just works. Unaccellerated, of course.
Also works fine in OS X. But only showing 256MB there.