Resolved Saving a Bricked Radeon 9700 Pro Mac Edition

[z970]

Before I pack this endeavor up and hang my head in defeat after multiple days of stressing over this fiasco, I wanted to get all your opinions on what to do to try to resurrect this card, as I've seen many instances of our forum's members coming up with insanely clever solutions that have never occurred to others before.

Several months ago, I bought a Radeon 9700 TX for a dual processor Pentium !!! workstation I've got. It worked fine, and there were no issues. The trouble is, the 9700 was completely saturating its AGP 2x bus, and as a result achieving only marginally better speeds than the Radeon 9000 before it.

This past week, I came across a June 2003 Power Mac G5 2.0 DP queried to be scrapped. So I took it home, cleaned it out, and fired it up. It instantly boot right up, bong and all, to a loaded install of OS X 10.4.11 that looked like it wasn't touched since about 2010. So, I swapped out hard drives, installed a fresh copy of Tiger (at least until Fienix leaves beta), and everything was working fine.

It had its stock GPU installed, a Radeon 9600 Pro. Now, seeing as how the 9600 Pro was essentially a 9700 slashed in half, and that I had an actual 9700 that was never reaching its full potential, I decided to read up on flashing graphics cards. - Of course, not before confirming that the 9700 TX and 9700 Pro Mac Edition were technologically identical and only had differences of BIOS and clock speeds.

Long story short, I used FreeDOS 1.1 and ATIFlash 4.07, and not only had a 9700 Pro Mac ROM ready, but a reduced 64k 9700 Pro Mac ROM as well, just in case. After struggling with DOS for a while, I finally ended up starting ATIFlash and followed The Mac Elite's instructions on ATI card flashing, and all went well.

Now, here's where the trouble begins. I dumped the original ROM without issue, and began to flash the original 128k Mac ROM. But in that immediate moment, I wasn't paying attention to how the backup ROM (from the card) was a little bit smaller than 64k, meaning the 128k ROM was much too large (of course, without any indication during flashing). But as soon as I realized that, the machine was already rebooted and it was too late.

This was fine. My rationale was that if it didn't work when tested in the G5 (the G5 ended up refusing to boot with it installed), I can simply go back and do a blind flash, as the process was simple enough, and I already had the sounds from the machine (indicating checkpoints in the process) more or less memorized. And that's what I proceeded to do after being rejected by the G5.

I did NOT however, anticipate that the P!!! would refuse to boot too, eventually beeping whenever a key was pressed, meaning there was probably an inescapable message on screen that couldn't be displayed. So that means my only planned escape route was effectively cut off right before me, and the card was rendered (at least until further notice) unusable.

Afterward, I came across this article (https://www.overclock.net/forum/74-...cked-graphics-card-fix-failed-bios-flash.html). I tried its advice and used one of the two Mac PCI graphics cards I had in the G5's PCI-X slot (backwards compatible), and discovered that it works. So, I plugged in the half-done 9700 (w/ floppy power cable via an adapter) with the display connected to a Radeon Mac Edition (PCI version) I had in one of the PCI-X slots, and it again refuses to boot up, even ignoring a PRAM reset request.

As a fallback, I tried the same thing on the Pentium !!!. I switched its primary graphics adapter to PCI via the BIOS (on a known good card), plugged in one of the Mac PCI cards (this time a Rage 128), and installed the 9700 in its AGP slot. Still no display. I then learn that Open Firmware GPUs do not work the same way as EFI GPUs, and will not work in both a Mac and a PC, unlike Intel Mac cards.

So (for lack of any more transitioning words), then I start trying to find a PC BIOS for either the Radeon ME, or the Rage 128, so either one of them can temporarily be flashed to try to get the Pentium machine to ignore the 9700 and just properly boot. But no matter where I look online, I cannot find a BIOS for the Rage 128, which I then learn seems to be a rebranded Xclaim VR 128, and that I also fail to try finding a BIOS for. What about the Radeon ME? Well, after researching not only the card specs but BIOS as well (no luck there), it appears that the Radeon ME is a kind of mix of the Radeon 7200 and the 8500, and that to my knowledge, no real PC equivalent (and by extension ROM) should exist for. At least, that's more or less what I was able to gather.

Therefore, I am locked out of accessing my card due to a lack of compatible computers that will POST w/ the card installed, and also a lack of PC PCI graphics cards to try overriding the only compatible PC I have access to. And thus, my current situation.

-

Apologies for the wall of text... Further ideas are of course most welcome.

 

[LightBulbFun]

Question

exactly which ROM did you use?

if you flashed the original Radeon 9700 Mac ROM to a PC 9700 then all your going to get is a VGA output from either of your ports

when flashing a PC Radeon 9700, you need to use a special ROM, which is a 9800 ROM modded to work with PC 9700's

it can be found here under the modified ATI PPC ROM section

Wikidownloads2 - The Mac Elite

themacelite.wikidot.com

(also the dumped PC ROM size is not an indication of the cards EEPROM size)

although sadly going by the fact your G5 refuses to POST, sounds like a bad flash (as with a 9700 PC card flashed with an MDD 9700 ROM, it should still POST just the PC 9700 card wont output any DVI output)

 

[z970]

I used the OEM Radeon 9700 Pro ROM from the Original ATI PPC ROMs.

But isn't the 9800 physically different from the 9700? There are more internal differences at play than the clock speed, aren't there?

There are multiple 9800 ROMs. Which one is the specially modified one?

Another reason I'm thinking the EEPROM's size is less than 128k is that there were several small chips on the board labeled either "24k" or "27k" (I can't remember). Certainly, that is less than even 64k, but that's supposed to be the smallest size available.

The G5 did actually POST, as the bong was there. What's funny is that it never got farther than that, even ignoring keyboard commands.

Could there be any other ways to get into a working system with the card attached?

 

[sparty411]

I used the OEM Radeon 9700 Pro ROM from the Original ATI PPC ROMs.

But isn't the 9800 physically different from the 9700? There are more internal differences at play than the clock speed, aren't there?

There are multiple 9800 ROMs. Which one is the specially modified one?

Another reason I'm thinking the EEPROM's size is less than 128k is that there were several small chips on the board labeled either "24k" or "27k" (I can't remember). Certainly, that is less than even 64k, but that's supposed to be the smallest size available.

The G5 did actually POST, as the bong was there. What's funny is that it never got farther than that, even ignoring keyboard commands.

Could there be any other ways to get into a working system with the card attached?

The only other differences there might be would be pixel pipeline count, ROP's, TMU'S, (no unified shaders on cards that old) and maybe the memory bus size. Check Techpowerup for a side by side comparison.

 

[z970]

The only other differences there might be would be pixel pipeline count, ROP's, TMU'S, (no unified shaders on cards that old) and maybe the memory bus size. Check Techpowerup for a side by side comparison.

Right, those were what I was talking about.

I thought these qualities were baked in, irrespective of the firmware?

 

[sparty411]

Right, those were what I was talking about.

I thought these qualities were baked in, irrespective of the firmware?

I do believe they are physical characteristics that cannot be modified by any type of firmware/BIOS. If the 2 cards are physically identical though, I can't see any reason why you shouldn't be able to use a 9800 firmware on your 9700. Definitely do a side by side comparison on Techpowerup.

 

[z970]

ATI Radeon 9700 PRO Specs

ATI R300, 325 MHz, 8 Pixel Shaders, 4 Vertex Shaders, 8 TMUs, 8 ROPs, 128 MB DDR, 310 MHz, 256 bit

www.techpowerup.com

ATI Radeon 9800 PRO Specs

ATI R350, 380 MHz, 8 Pixel Shaders, 4 Vertex Shaders, 8 TMUs, 8 ROPs, 128 MB DDR2, 340 MHz, 256 bit

www.techpowerup.com

Looks like the only meaningful difference is that the 9700 has 110m transistors as opposed to the 9800's 117m transistors.

So... that's what I thought. Unavoidable physical differences.

 

[sparty411]

ATI Radeon 9700 PRO Specs

ATI R300, 325 MHz, 8 Pixel Shaders, 4 Vertex Shaders, 8 TMUs, 8 ROPs, 128 MB DDR, 310 MHz, 256 bit

www.techpowerup.com

ATI Radeon 9800 PRO Specs

ATI R350, 380 MHz, 8 Pixel Shaders, 4 Vertex Shaders, 8 TMUs, 8 ROPs, 128 MB DDR2, 340 MHz, 256 bit

www.techpowerup.com

Looks like the only meaningful difference is that the 9700 has 110m transistors as opposed to the 9800's 117m transistors.

So... that's what I thought. Unavoidable physical differences.

A quick Google search yields a bunch of different stories from people with success stories. I don't think the transistor count should matter. Your call though. The 9700 Pro is quite a rare card, as many of them have already bit the dust, due to inadequate cooling solutions. I have a pile of dead 9700 and 9800 cards sitting on a shelf lol.

 

[LightBulbFun]

I used the OEM Radeon 9700 Pro ROM from the Original ATI PPC ROMs.

so im guessing you never read the appropriate buyers guide then?

Flasher's Buying Guide - 95/9700 - The Mac Elite

themacelite.wikidot.com

where they link the appropriate ROM to use

or the ROM chip directory to determine your EEPROM chip size? http://themacelite.wikidot.com/rom-s

please do your research before attempting to just haphazardly trying flash a card with the wrong ROM, as you can very easily brick something to a difficult to recover state

(NB I found ATI flash in DOS to be a bit finicky for flashing PC cards to Mac, I highly recommend you use ATI Flasher in OS X to flash your ATI cards http://thomas.perrier.name/graphiccelerator.html )

 

[z970]

@LightBulbFun Alright, my fault... I thought it should have been a lot more straightforward than that, and evidently, I was wrong.

One way or another, it will eventually be recovered. The way I see it, if the fan still spins and nothing's fried, it's just a step away from factory state.

Thanks for the links...

 

[AphoticD]

@z970mp, do you have access to a G4 tower? It might be the G5 is a bit picky about things.

I had a G5 that refused to boot with any AGP card installed. After trying (buying) several different GPUs, I figured out it was the logic board at fault...

 

[z970]

@AphoticD Not a functional one, no.

(Unless of course anyone is willing to loan me a QS PSU...)

I'm still working on it. Attempt after method after theory after error...

 

[z970]

The Radeon lives again!!

Without any immediate logical or reasonable explanation, something in me told me to tape pins 3 and 11, thereby disabling AGP 8x mode. Total gut decision out of nowhere. My best guess is that something about the OEM ROM coupled with this card variant doesn't bode well for 8x mode, thus the refusal to function properly. (http://themacelite.wikidot.com/pins-3-and-11)

Thankfully, as The Mac Elite has said, this card doesn't seem to mind being in 4x mode on a G5. So, although it's unoptimized, and still flashed incorrectly, it is at least entirely functional and usable in this state. DVI appears to still be DVI, and VGA likewise.

But even when limited to 4x mode, it can still pump out much more than the stock 8x 9600, meaning in all likelihood, it isn't bottlenecked at all.

I can also confirm that TFF webpage scrolling is noticeably smoother now, too (even with no GPU acceleration and all). Of course, foxPEP helps here.

Once correctly booted, I tried using Graphiccelerator to flash the correct ROM (modified 9800 one from Mac Elite), but upon the prompt to flash the card in Slot 1 (AGP slot), I was given this dialogue.

The best I could gather via aggregated research is that Graphiccelerator doesn't "like" the ROM that already lives on the card, and refuses to go any further when it sees it. Safe mode does not change this behavior.

In any case, this will do the job fine until some means to flash the card properly eventually comes my way. In the meantime, ATIcellerator II is doing a fine job of underclocking the card so it isn't getting so hot, and thus, wearing on the circuitry (it has also already been cleaned and repasted).

Thanks for the help, all. My week-long torture trial is over!

 

[andysa1966]

Good news,

not totally off topic, i have a radeon 9700 Pro ME, I forgot to unplug the power cord when pulling it and managed to touch terminal with the screw driver and zapped it, blew the cinema screen, now the card still works with a green or pink tinge with the ADC output, works fine with DVI.

Wish i could fix the problem, may only be a blown resister

 

[z970]

@andysa1966 My condolences.

 

[LightBulbFun]

it is at least entirely functional and usable in this state. DVI appears to still be DVI, and VGA likewise.

colour me surprised!

do you have a photo of the actual card itself?

I was under the impression that when a Radeon 9700 PC is flashed with the Mac ROM you only get VGA out of the 2 ports

because the Mac Radeon 9700 has 2 DVI (ADC counts as a DVI port in this context) ports one via an external TMDS transmitter so this means the Mac 9700 ROM's port mapping/framebuffer personality does not match up with the PC 9700's port layout

but clearly DVI seems to be working for you with your 9700 and the Mac 9700 ROM

interesting that it refuses to work in AGP 8x mode, not overly surprising given the original card/ROM was made for an AGP 4x computer only, but still interesting

I had wondered if the MDD 9700 ROM was incompatible with the G5's Firmware

if the 9700 MDD ROM does work fully with your card, then I recommend you stick with that ROM rather then the modified 9800 ROM

 

[z970]

@LightBulbFun

It's identical to this one.

When I am within the ability to, I'd like to at least try the 9800 ROM. Keeping Graphiccelerator compatibility and AGP 8x are things I'd rather have than not. If at least for slightly better G5 compatibility.

I'm not surprised that the 9700 MDD ROM eventually worked out with the G5 firmware. MDD '03 and G5 Rev. 1 turned out to be very, very similar under the hood, as discovered by SkyCapt at Macintosh Garden. Amazing it took so many years for the community to figure that out...

 

[LightBulbFun]

MDD '03 and G5 Rev. 1 turned out to be very, very similar under the hood, as discovered by SkyCapt at Macintosh Garden. Amazing it took so many years for the community to figure that out...

No, no they are not...

 

[z970]

No, no they are not...

I'm still trying to find that article I read detailing this. Its arguments and conclusions were very well read, researched, and vetted. Thereby, I found it EXTENSIVELY plausible to believe. But for the life of me, it's disappeared off the face of the Web.

I think it's OK to revisit historical information as long as it's in the name of accurate, honest truth.

 

[LightBulbFun]

I'm still trying to find that article I read detailing this. Its arguments and conclusions were very well read, researched, and vetted. Thereby, I found it EXTENSIVELY plausible to believe. But for the life of me, it's disappeared off the face of the Web.

I think it's OK to revisit historical information as long as it's in the name of accurate, honest truth.

im sorry but what?!

this came up one time before on this forum and I know I disproved it, so im quite surprised to see it come up again!

the PowerMac G5 and PowerMac G4 MDD have nothing in common on a motherboard architecture level

heres a FW800's block diagram

and heres the block diagram for a First generation G5

 

[MagicBoy]

I'm still trying to find that article I read detailing this. Its arguments and conclusions were very well read, researched, and vetted. Thereby, I found it EXTENSIVELY plausible to believe. But for the life of me, it's disappeared off the face of the Web.

I think it's OK to revisit historical information as long as it's in the name of accurate, honest truth.

Er. No.

The whole premise of the G5 was high bandwidth throughout the system which is why they started again using the HyperTransport bus and a processor architecture that IBM were using in RS/6000 and AS/400 mid-range enterprise systems.

I'm all in favour of accurate, honest truth. Saying the MDD '03 and G5 Rev. 1 are similar under the hood is false. The Block diagrams lightbulbfun posted prove this.

 

[AphoticD]

I'm still trying to find that article I read detailing this. Its arguments and conclusions were very well read, researched, and vetted. Thereby, I found it EXTENSIVELY plausible to believe. But for the life of me, it's disappeared off the face of the Web.

I think it's OK to revisit historical information as long as it's in the name of accurate, honest truth.

I've read quite a few posts across the 'net from a user named SkyCapt (also the user "whichkraft" at https://arstechnica.com/civis/viewtopic.php?f=19&t=1329095 which is written in the same style), they mention a G5 motherboard and memory controller under the hood of the MDD.

No harm in having an opinion of course, except for when it gets mixed in with fact and people run with the misinformation!

 

[z970]

I see... Maybe it was its own, special respin on previous boards? In the long run, I still think the performance that our friend SkyCapt managed to pull out of his MDD deserves closer and more careful examination, because the results really are intriguing.

In any case, thanks for the correction all.

 

[MagicBoy]

There's some comedy gold in that Ars thread! They seem to share the scepticism...

"Does your toaster blow when the wind comes from the southwest?"

"Have you considered tweaking CONFIG.SYS to get your low memory availability over 600K? That helps."

?

 

[MagicBoy]

I'd start with examining whatever SkyCapt had been smoking. Bunnspecial might know where to find a mass spectrometer...

Looking at at the other threads from that poster it's likely another case of an "interesting" person on the internet with irrational obsessions that override facts, like good old Rabidz on here. The thread on the DVD player application is illuminating.