PowerBook Ti ATI Radeon 9000 vram replacement

[pbertolo]

I recently scored a wonderful PB Ti 1 GHz, top specs (1 GHz, Radeon 64), complete with everything, down to the plastic wraps.
I am glad I could reacquaint with this ultra-cool machine, as it reminds me of when I was (well… twenty years) younger and I used to own a wonderful 667 DVI unit which I wrongfully got rid of when its time had come.
Seller auctioned it as in “unknown“ conditions (generally it means: there’s a catch, but if I disclose it nobody will buy it) and unfortunately it seems as though the machine has bad vram, resulting in garbled stuff on the screen, unless you go down to 32k colors (OS 9) or gray scale (OS X). So, I guess there’s at least one bank which is faulty.
Apparently, not such an uncommon issue with these machines.
I am not particularly bothered by it, but I was wondering if it possible to fix it.
Now, everything is soldered, but by some quick check I would speculate that the vram chips could be reballed, assuming one can source a spare (by searching the web, similar modules are still available for cheap).
Finding some pro lab familiar with reballing of mobile phones, tablets etc shouldn‘t be an issue.
However, before going into this adventure, I was just wondering if any of you has any experience with that.
Thanks in advance for any piece of advice.

 

[GRudolf94]

That is based on the assumption that the issue is "VRAM is faulty". Spares should already come with solder balls, and as such, no reballing is needed - you get the old chip off, clean the GPU interposer, solder the new one on. Heck, ideally one would do both in one go in this case.

That leaves another issue: these were never meant to be reworked, the chips are probably glued to the interposer with epoxy (admittedly, I haven't looked at one of those ATi GPUs in a while and don't remember, really), the interposer is much easier to damage than a regular board - it's rather thin, and the GPU itself might be faulty, too. It's a potentially annoying repair, and I think your biggest issue here is finding someone who'll agree to it. Your best shot is someone who does PS3 repair (the GPUs on those were also MCMs like the Radeon on the TiBook is). A lot of the doubt and annoyance can be bypassed by sourcing a whole replacement GPU, if that's even possible.

 

[pbertolo]

The hardware test is returning the message “vram faulty”, with an additional code reading “disp/13/2”.

To me it makes sense, as the issue is visible at full color resolution and disappears as you go down to 32k colors in OS9 (OS X doesn’t support 32k colors, you have to go down to 256).

There seem to be 4 very accessible (at least visible) vram chips, each of them 128 Mbit: 2 installed directly on the discrete board, two (the optional 32 Mb) on the logic board.

By logic, I would say the internal display doesn’t need to map the extra 32 Mb, so my guess is that at least one of the two chips on the board is gone.

Again, it’s pure speculation, I have no direct experience, that’s why I am seeking for advice.

 

[Phipli]

By logic, I would say the internal display doesn’t need to map the extra 32 Mb, so my guess is that at least one of the two chips on the board is gone.

I wouldn't assume that.

The 9250 I upgraded from 128 to 256MB, in doing the upgrade, I also upgraded from 64 bit to 128 bit. Effectively the new VRAM was in parallel. When I had a bad solder connection on the new VRAM, it impacted the display even at low resolution.

If the setup is similar, the same might be true.

 

[pbertolo]

I see the point. Can’t really tell, my speculation was just based on the fact that the baseline version of the video card had only 32 Mb. So, I assumed the optional 32 Mb would be used to pilot a bigger external display, but could be whatever, it’s clear vram is not just about resolution.
Anyhow, the ATI 9000 on the PB is equipped with sgram Samsung K4D263238A - GC 45. Some other boards are equipped with Infineon HYB25D128323CL-4.5 Or Hynix HY5DV283222A. The Samsung chip has “low” access frequency, shall I assume faster chips will work the same?

 

[GRudolf94]

The hardware test is returning the message “vram faulty”, with an additional code reading “disp/13/2”.

To me it makes sense, as the issue is visible at full color resolution and disappears as you go down to 32k colors in OS9 (OS X doesn’t support 32k colors, you have to go down to 256).

What I'm saying is the hardware test doesn't know the difference between a physically faulty VRAM chip, a poor connection, or an issue within the GPU itself (say, a dead data line) that prevents correct access to the VRAM.

Anyhow, the ATI 9000 on the PB is equipped with sgram Samsung K4D263238A - GC 45. Some other boards are equipped with Infineon HYB25D128323CL-4.5

Both are rated at 4.5ns and therefore capable of the same access speed.

 

[pbertolo]

What I'm saying is the hardware test doesn't know the difference between a physically faulty VRAM chip, a poor connection, or an issue within the GPU itself (say, a dead data line) that prevents correct access to the VRAM.

Sure, assuming it’s indeed just one or more failed vram chip is a bit like shooting in dark, I agree. I will try to take some pics of the issue, maybe they can reveal a little more.

 

[pbertolo]

Just in case, here are 3 screenshots (made with Grab) showing how the issues goes with the color depth. When the issue is present (almost for any resolution / color depth combination, except the poorest ones), the issue is dynamic, in the sense that it is always different and it changes pattern as you move things around the screen. I assume Grab dumps the content of the vram, but surely if the issue is elsewhere (e.g. data　lines) the result wouldn't be any different. Interesting to note that in OS X it looks much worse than under OS 9. In OS 9 as long as you limit the color palette to 32k, there's no issue. Here I would speculate that the much more demanding graphic features of OS X (transparency etc) are using more VRAM, hence the issue appears at 32k colours, too.

 

[MacUp72]

Yeah, Apple and ATI, a history of failings
I once tried the oven baking method, it worked but not for very long..I dont know if a Pro-reballing is worth it?

 

[pbertolo]

First off, the PRAM battery had to be replaced, as totally dead. Power management behaving totally erratically. Curiously, the optical unit was also affected.
Original item: Panasonic CGL3032, Li ion 3.7V rechargeable battery.
Impossible to source an exact spare, seems like it has never been available for retail. Some shops on eBay offer working spares (including the board), but at hefty prices. Additionally, shipment is not always granted due to risks connected to Li ion batteries.
Luckily, where I live (in Japan) there's an online shop that is selling a custom made spare they get produced on purpose in China.
Still on the expensive side (around 25 USD), but all in all more than acceptable.

 

[Phipli]

First off, the PRAM battery had to be replaced, as totally dead. Power management behaving totally erratically. Curiously, the optical unit was also affected.
Original item: Panasonic CGL3032, Li ion 3.7V rechargeable battery.
Impossible to source an exact spare, seems like it has never been available for retail. Some shops on eBay offer working spares (including the board), but at hefty prices. Additionally, shipment is not always granted due to risks connected to Li ion batteries.
Luckily, where I live (in Japan) there's an online shop that is selling a custom made spare they get produced on purpose in China.
Still on the expensive side (around 25 USD), but all in all more than acceptable.

View attachment 70645

Sure they're not cheaper somewhere?

Locally, we have these :

Panasonic CGL3032 3.7V, LIR-3032, LIR3032, KL3032, L3032, ELIR3032, KLIR3032, | eBay

The pins are for horizontal mounting on a PCB. It results in significantly higher energy density than NiCd and NiMH batteries. Have proved useful in refurbishing Fossil PDA watches and Apple Power Book PRAM battery assembly (Apple part number: 922-6714, 922-6711, 922-4631 etc).

www.ebay.co.uk

Tabbed LIR3032 batteries aren't a unique part to Apple.

 

[pbertolo]

Yes they are cheaper and I saw that, but the ad reads they won't be shipped to Japan.

 

[pbertolo]

Anyhow, still behaving totally erratically. The main battery is dead, but there's no way I can power it up using only the external power supply (65W).
After starting up, the machine goes on sleep continuously, as if it was operating on a dead battery.
Already tried to reset PRAM and PWM, to no avail.
I am almost sure that back at the time my 667 DVI could run just on the power supply, like a desktop.
Or am I missing something?

 

[pbertolo]

Eventually, I replaced the motherboard with an 867 Hz / Radeon 32. In fact, in the original board with 64 Mb of VRAM, there's one memory chip which is just very partially in contact with the heat dissipator. It also looks a bit "toasted", with the lettering somehow darkened when compared to the cooled chip. So, my guess is that it's probably the faulty component.

 

[helmetguy]

Just in case, here are 3 screenshots (made with Grab) showing how the issues goes with the color depth. When the issue is present (almost for any resolution / color depth combination, except the poorest ones), the issue is dynamic, in the sense that it is always different and it changes pattern as you move things around the screen. I assume Grab dumps the content of the vram, but surely if the issue is elsewhere (e.g. data　lines) the result wouldn't be any different. Interesting to note that in OS X it looks much worse than under OS 9. In OS 9 as long as you limit the color palette to 32k, there's no issue. Here I would speculate that the much more demanding graphic features of OS X (transparency etc) are using more VRAM, hence the issue appears at 32k colours, too.

Interestingly, I've seen the desktop version of the ATI Radeon 9000 do the exact same thing when it gets really hot.

 

[pbertolo]

Interestingly, I've seen the desktop version of the ATI Radeon 9000 do the exact same thing when it gets really hot.

I don’t have the equipment nor the skills for attempting the replacement of the vram chip on my own, but if I find a lab that could do that for me I would like to give it a try and replace the ”toasted” chip. All the clues are leading to it.

 

[pbertolo]

Anyhow, getting close to regretting having bought this machine…
So far:
- replaced the PRAM
- replaced the internal drive (absent) with SSD
- replaced the motherboard, due to the faulty video card
- replaced the optical drive (prone to failure and it failed)
Next will be refurbishing the main battery (this would be just for fun, so to speak).
Fundamentally nothing was still in working order after 20 years. Definitely not a machine that has aged gracefully, gorgeous look aside.
As for the optical drive, this is something that drove me mad.
The unit was behaving erratically, sometimes it would work, sometimes it would not. Some disks were OK, some others were not.
After replacing the motherboard, it just stopped working at all, no disc was recognized.
So, I sourced a spare sold as certified working.
But when I dropped it in, the same: no disc could be read.
I had to hard reset the PRAM a few times and then it started working again. I suspect the original unit is in fact OK.
Very picky machine it is...

 

[jeremywork]

I realize this is no longer the question at hand, but for what its worth the 64MB Mobility Radeon 9000 in the TiBook will limit each display's frame buffer to 32MB, though the total can exceed this (3840 x 2160 can run at 32bit/Millions [~31.6MB] while the onboard panel is at 1280 x 854 / Millions [~4.2MB; total ~35.8MB], but 3840 x 2400 is reduced to 16bit/Thousands even when the onboard display is at 640 x 480 / 256 [total would be ~35.5MB if 3840 x 2400 worked at Millions, so not hitting a total limit].)

The VRAM chips are almost certainly addressed in parallel as @Phipli suggested, but it is interesting to see 32MB chosen as a per-buffer limit (perhaps just coincidentally.)