• Please review our updated Terms and Rules here

Anyone have any ideas on how to get this bottom cover off my g3?

Evelyn Rumti

New Member
Joined
Jun 1, 2026
Messages
2
I have been struggling to get this bottom panel off my 1998 bondi blue imac g3 for over 5 days straight. The square part that's on top of the cd drive and near the screen is the only part that's stuck. I've tried using ifixit wedges and flathead screwdrivers in the seams trying to pry it lose. I've tried wiggling and prying while my sister pulls the panel from the handle upwards. I've tried thumping it and pushing the panel inwards and nothing. I even got my dad to pull on it upwards as hard as he can while I wiggle and pry multiple flathead screwdrivers in the seams. And we also tried pushing and thumping it while he pulled. I went through countless guides online for my g3 revision. Text tutorials, video tutorials, user manuals, online forums. And even forums about my exact issue. And everyone seems to be able to remove the panel at one point. But I've been trying for days straight. And making little to no progress. It seems to be most jammed in this one spot (You can see in attached pictures). It's a miracle we haven't cracked anything yet. But I'm hoping that you guys maybe have something else I can try to finally remove this cover?1000160091.jpg1000160097.jpg
 
That part should be fairly easy to remove, I've certainly never had too much trouble in the past.

Sometimes it can get stuck, but I would usually just grip the whole plastic cover from the top and gently twist/wiggle while pulling upward. Pressing inward on the lower square/rectangular part may help.

There are a handful of plastic tabs around the upper part of the cover. Some of them are (I think) just rectangular guides to aid in correct placement, but some sort of snap in (friction fit?).

Try to make sure the top part is loose/free as well, before trying to pull it off.

If the CD drive has shifted internally or taken a whack from the front, it might be putting pressure on the shell. Can you set the machine upright in the proper orientation to use it and post a picture?
 
Last edited:
Well, I did actually manage to get it open last night. Turns out, all I had to do was pull it outwards instead of just pulling up. Only to be greeted with loads of battery acid cheese and rust. And capictors just falling off. So, that board is toast.
 
Well, I did actually manage to get it open last night. Turns out, all I had to do was pull it outwards instead of just pulling up. Only to be greeted with loads of battery acid cheese and rust. And capictors just falling off. So, that board is toast.

Pictures?

Damaged and totally, utterly beyond repair are different things.
 
It seems to be most jammed in this one spot (You can see in attached pictures). It's a miracle we haven't cracked anything yet. But I'm hoping that you guys maybe have something else I can try to finally remove this cover?

FWIW, I actually kind of had to be amused/amazed at this, because, well.... A: the plastics of these machines have a really unfortunate reputation for getting brittle as they age, but more personally, B: I was given one of these machines by a rabid Apple fanboy/all-around-tech fiend friend when it was only about two years old (*). When I got it it already had broken hinges on that door that covers the ports on the side, and it was *really easy* to get the "pants" off it (that's what I called the piece that's giving you trouble) because it'd already been... thoroughly loosened.

(* Got it in late 2000; he dumped it on me because he and his wife had come to an agreement about how many computers he could have in the house at any one time and he wanted a new toy. Ended up installing OS X Public Beta on it the week it came out. Also got a few old Unix workstations from the same source, and they'd likewise clearly experienced some... tough love. And I mean it when I say "love", I mean that part, the guy had a huge passion for these machines that apparently extended to a burning desire to see what was inside. At all costs.)

I'd say definitely be proud of yourself that you were able to figure it out without going full "hulk smash" on it, I guess. ;)

Only to be greeted with loads of battery acid cheese and rust. And capictors just falling off. So, that board is toast.

Since you said battery I assume the cheesed board is the motherboard. My condolences. If it's of any comfort these... weren't great machines anyway. I actually ended up with the motherboard from a *second* one of these from someone who'd planned to repackage it in a cute box after the monitor died (and the monitor *does* always die on these, it started going flaky on my first one after only a year or so), and I half-heartedly poked at the project for a while before purging myself of the whole lot a few years later. They're just not very good computers. (Too slow/limited memory for OS X, they're really picky about hard disks... they're just objectively too limited to be of much use beyond running OS9.)
 
Last edited:
You can actually put up to 512 MB of ram in the Rev. B model, which is the preferred model anyway because it has 6 MB of VRAM instead of just 2 MB.

The only obvious reason I can think of for it to be "picky" about hard disks, as long as they're real spinning media is that the onboard controller only supports ATA-3. And 80 GB or less is a safer bet than 128 GB+.

Based on the table at:

I would guess that the drive needs to support CHS, 22-bit LBA, or 28-bit LBA since 48-bit LBA only came along with ATA-4.

And of course it needs to be compatible with the formatting tools included with OS installer.

I don't recall the source of the issue (OpenFirmware, Mac OS, other hardware), but you generally want the operating system installation to stay inside the first 8 GB of the drive on these systems
 
Last edited:
You can actually put up to 512 MB of ram in the Rev. B model, which is the preferred model anyway because it has 6 MB of VRAM instead of just 2 MB.

You can put 512MB in a Rev A as well, they have the same Grackle memory controller. (For that matter if you want 6MB of VRAM in a Rev. A you can fix that, too, it's on an SGRAM DIMM. The only set-in-stone difference is the Rev. B has an ATI Rage Pro instead of a Rage IIc; this same difference exists between older and newer revs of the Beige G3.) The reason a lot of people think that 256MB is the limit on either one of these is because the vast majority of 256MB SO-DIMMs you'll find lying around were pulled from laptops newer than the iMac G3 and have 256MB in a single bank, which the Grackle can't take. (You'll also see "384Mb" quoted as a limit on these machines because for a while the only 256Mb dual-bank SO-DIMMs that were available were twice as tall as the standard low-profile ones, and a tall one will only physically fit in the slot on the upper side of the CPU card. Googling around I guess there's a bunch of people that cite Low End Mac as a source for a claim that Rev. As only support 128MB in one of the slots, but if we know anything now it's take *anything* that ever appeared on Low End Mac with an iceberg-sized grain of salt.)

The most reliable source for DIMMs that actually work are from early high-end Pentium III laptops that were new enough to be worth maxing out but old enough to still have 440BX-era motherboard chipsets. (Laptop chipsets based on the 815 and later support the denser DIMMs, and, like I said, those machines vastly outnumbered the older ones even in the early 2000s.) I had 512Mb in mine, and I would still call it not enough to be useful, although I guess I'll be fair, OS X was a wallowing pig on these machines for all the reasons. The difference between 256 vs 512mb of memory was hardly the straw that broke the camels' back.

The only obvious reason I can think of for it to be "picky" about hard disks, as long as they're real spinning media is that the onboard controller only supports ATA-3. And 80 GB or less is a safer bet than 128 GB+.

Have you actually ever owned one of these machines? All of these older Macs (I'm going to call anything predating the AGP G4 towers "old") are weird and picky about a *lot* of things. Both the iMac G3s and my B&W G3 tower would be just plain weird about certain hard disk brands or models. Off the top of my head I recall the iMac was perfectly fine with Quantum drives (I think that's what the OEM one was), but it absolutely refused to behave with either Fujitsu or IBM Deskstar disks, and with either Maxtor or Western Digital you could flip a coin whether it'd work or not. (All of these were drives under 128GB and worked fine in other machines.) The B&W was "less" pickey but still hated Deskstars universally, and I also ran into an odd situation where it worked fine with a Western Digital 80GB to install OS9, but if you tried to install OS X on that particular drive it'd go through the motions but then greet you with a big full-screen (/) NOPE symbol trying to boot. Swapped it for a same-size Seagate, no problem.

I mean, FWIW, maybe other people had much better luck. Even within brands it would mostly be hit and miss, it could well be that I just kept drawing the jokers out of the deck. But I have heard similar complaints from other people about these older Macs.

don't recall the source of the issue (OpenFirmware, Mac OS, other hardware), but you generally want the operating system installation to stay inside the first 8 GB of the drive on these systems

The tray-loading iMac G3s (and I guess it sounds like the first-gen iBooks?) were the last machines to have this issue. And it's another reason why they suck for OS X, given the OS basically consumes that much. I actually hacked up the Netinfo database on the OS X install I had on the iMac so it'd put the user directory directly on the next partition, and likewise relocated the Applications directory, but it was a huge hassle that a normal mortal isn't going to do.

(I *really* went to town with breaking NetInfo on Public Beta; I had the machine sitting in my cube at the office, and one boring afternoon I actually succeeded in making it possible for people to log into it using their NIS accounts. It's really fun how much of the old NeXTStep OS was still hiding under the covers in early releases of OS X. I also experimented with running OS X on the NextStep UFS filesystem, it was an option, which turned out to be a BAD IDEA. The OS itself would work for the most part with a true case-sensitive filesystem, but applications broke *all the time*, including a depressing amount of the early open-source ports. Fun times.)
 
Last edited:
I haven't used them much in some years, but I do have a Bondi iMac G3 as well a couple G4 Macs and a few earlier machines (LC II, LC "III", Power Macintosh 6100).

The iMac G3 isn't exactly stock any more, though, as I ended up putting a faster CPU module (333 MHz) into it that probably came from one of the later "fruit flavor" units. And I previously updated the firmware on the previous CPU module.

I know there are plenty of Macs with quirks, to say the least, just don't recall having any real problems with this model.

Never really had any inclination to put Mac OS X on it though. My elementary school had iMac G3 systems in their computer lab, so I have a certain nostalgia/fondness for the classic Mac OS.
 
Never really had any inclination to put Mac OS X on it though. My elementary school had iMac G3 systems in their computer lab, so I have a certain nostalgia/fondness for the classic Mac OS.

I will say in all fairness to these machines they *do* seem to be a lot less problematical if you just keep the classic OS on them. I specifically mentioned it with the B&W incident, but I think I saw the same thing with the iMac, IE, sometimes these drives that would refuse to boot or even install in OS X would seem to be fine if all you did was slap OS 8.6 or 9.0 on the machine. FWIW they also seemed a lot more forgiving under Linux, but Linux very much suffered from a "why am I running it on this, again?" problem.

(Mac-On-Linux was pretty cool, though.)

They are still pretty awful computers in terms of build quality as well, though. If your iMac's monitor is still on the original flyback transformer it is a genuine miracle. On the bright side, if you pull those plastic "pants" off all you need to do is unscrew the little pigtail that goes to the internal monitor and replace it with a multisync Mac-to-VGA adapter. With 6MB of VRAM you could push as high as 1600x1200, you weren't locked into the limited range of the internal monitor. Adapting the slot-loader iMacs after their monitors died was a *lot* harder.

Ultimately I guess it just seems like these machines aged like the finest milk. They were legitimately kind of impressive for five minutes when they came out in 1998 and may well have saved Apple Computer, but those pretty shells hid a lot of pent-up technical clunkiness. But I suppose if you compare them to a contemporary Pentium II or something it's not like one of those was much use three years later either. This period from the late 90's through the first half of the aughts might actually mark the peak of rapid obsolescence in the PC industry; the only saving grace that's likely to apply to the Pentium II is you could probably swap out its motherboard/video card/etc. It was a particularly bad time to be a proprietary All-In-One.
 
Last edited:
Depending on the exact motherboard and which chipset was used, a Pentium II system could potentially have been upgraded to a Pentium III or it's Celeron equivent and a decent chunk of memory.

That would have served well enough to squeeze out a few more years running Windows XP, even if it started life running Windows 95.

Realistically the iMac G3 would probably have become a decorative paperweight by comparison.
 
I would guess that the drive needs to support CHS, 22-bit LBA, or 28-bit LBA since 48-bit LBA only came along with ATA-4.
I don't think LBA-48 requires ATA-4.

Disk block addressing methods are limited by the driver and the hard drive. The ATA controller shouldn't care. It just needs to pass the register values correctly to the ATA device.

I have CHS/LBA notes regarding Open Firmware at
https://forums.macrumors.com/threads/cant-boot-from-firewire-drive.2478368/post-34454141
Basically, if you remove the driver limits, then you should be able to connect a LBA48 drive to any Mac that has IDE.

I created a patched Mac OS X Heathrow/OHare kext to support LBA48 but haven't tested it yet.
I think Mac OS 9 supports LBA48 already.

You can put 512MB in a Rev A as well, they have the same Grackle memory controller
Grackle supports up to 1 GiB of RAM but I think you need 4 slots for that.

The DingusPPC emulator for Gossamer machines allows connecting two 512 MiB DIMMs or one 1024 MiB DIMM because the HWInit code during POST accepts the SPD info that DingusPPC provides. The real Grackle can't handle DIMMs of those sizes.

I didn't test that in the DingusPPC emulation of Yosemite or Yikes machines because they have 4 DIMM slots that can take 256 MiB DIMMs. Actually, DingusPPC has a 512 MiB DIMM option for those and BondiBlue machines.

The SPD of these DIMMs in DingusPPC are set like this:
Code:
        case 512:
            this->eeprom_data[3] = 0xC; /* 12 rows    */
            this->eeprom_data[4] = 0xB; /* 11 columns */
            this->eeprom_data[5] = 0x2; /* two banks  */
            break;
        case 1024:
            this->eeprom_data[3] = 0xC; /* 12 rows    */
            this->eeprom_data[4] = 0xC; /* 12 columns */
            this->eeprom_data[5] = 0x2; /* two banks  */
The grackle aka MPC106 documentation does not list those options as valid.
I think MPC107 added an extra bit or two to allow one or both of those.
 
There were 3rd party CPU upgrades released for the tray loading iMac G3s, but not the later slot loading models, since the CPU was soldered to the logic board on those.

You could get at least a 600 MHz G3 CPU upgrade for the tray loading model, but it is of limited usefulness due to the crappy ATI Rage integrated graphics. Very early models had the "mezzanine connector", which was basically a glorified PCI slot with a different connector. The only known video upgrade was a Voodoo2 board with 8 MB, and they're exceptionally rare today.
 
Grackle supports up to 1 GiB of RAM but I think you need 4 slots for that.

Yep. That's why the Beige G3 supports 768MB (three slots) and the B&W a gig.

My B&W was maxxed out to that as well as having a G4 ZIF+the necessary firmware patch to unlock it, making it essentially a Yikes in blue overalls. It also had a *pair* of Radeon 7000s in it to support dual monitors. It ran Tiger... surprisingly well, I guess, other than the video being a bit sluggish. I did try that patch to enable Quartz Extreme over PCI, but the gains were... negligable, and it might have made other things worse. (The Grackle just has too limited bandwidth; this isn't a north/southbridge setup, it basically just has a single PCI "lane" that's split up with PCI-PCI bridges between the onboard peripherals and the slots, so if you enable texture transfers from system RAM to video that's going to clobber everything else contending with the same resources.) Still, it definitely proved the longevity value of being able to make a few strategic upgrades.
 
As far as I know, PCI is fundamentally a shared bus setup anyway.

So the only way to really increase performance beyond using a single card would be boosting the clock speed or adding a second PCI bus (which is basically how AGP works).

-----

I have to wonder if some sort of separate memory bus that runs between cards would have been useful to supplement PCI. Exchanging data between main memory and a PCI device is probably the biggest bandwidth hog.

Hilarious idea -> squeeze two graphics cards onto one card...
 
As far as I know, PCI is fundamentally a shared bus setup anyway

It is, but by the time the B&W came out PC chipsets were moving the PCI slot control to a Southbridge chip that had a faster private bus to the Northbridge memory controller. This would let them, for instance, hang the video card slot directly off the Northbridge so texture access didn’t contend directly with the pipe going to the regular PCI slots and whatever greedy peripherals like disk controllers you also had embedded in the SB.

In short the Grackle was just badly obsolete by 1999; in terms of total system bandwidth it’s comparable to pretty ancient Pentium chipsets like the Neptune. The only saving grace was the version in the B&W could run that one PCI segment shared between the video card and the I/O chips on the motherboard at 66mhz instead of 33mhz.
 
While PCI is a parallel drop bus, you're not required to have multiple PCI devices on the bus, meaning that all bandwidth can be dedicated to a single device.

This is what HEDT workstations and servers used to do. They'd either have a single slot, or at most 2-3 slots on a PCI bus controller, and have multiple PCI bus controllers on the motherboard, each for their own group of slots. This got around another problem of the PCI bus having to operate at the lowest common denominator. So if you had 66-133 MHz bus devices and slammed in a 33 MHz card, it'd reduce the entire bus speed for all of the cards. Same for 66-100 MHz devices, you don't want them on the same bus as 133 MHz devices to avoid performance loss.

The PowerMac G3 B&W and the G4 Yikes! IIRC had a 66 MHz PCI slot for just the video card, and another PCI bus for the other three PCI-X slots.

1780525609585.png
 
The PowerMac G3 B&W and the G4 Yikes! IIRC had a 66 MHz PCI slot for just the video card, and another PCI bus for the other three PCI-X slots.
Grackle based Power Macs have one PCI bus hosted by Grackle (MPC106).

On the B&W G3 the PCI bus is 32-bit, 66 MHz. That's weird because MPC106 is supposed to only support 33 MHz. You see in the lspci dump below that MPC106 has the 66MHz flag not set. The B&W G3 Developer Note says the 66 MHz PCI slot cannot support 33 MHz PCI cards. The B&W G3 schematics for the slot show the M66EN pin is pulled high to force 66 MHz for the PCI card.

This 66 MHz bus has the graphics PCI card and also a PCI bridge for the 64-bit 33 MHz PCI slots. These slots are not PCI-X. PCI did include 64-bit and 66 MHz options. PCI-X adds 133 MHz and higher options.

The Open Firmware PCI device tree for the B&W G3 looks like this:
Code:
/pci@80000000
  /pci-bridge@d
    /mac-io@5
    /firewire@0
    /pci-ata@1
    /pci-bridge@2 - slot J11
    /pci104c,8025@3 - slot J10
    /ACARD,6280M@4 - slot J9
    /usb@6
  /ATY,BugsyParent@10 - slot J12

It doesn't show info for the grackle PCI host device at device number 0.

Here's an lspci dump of the Grackle host, the PCI bridge, the GPU, and FireWire (using my "lspci for Open Firmware" scripts at https://forums.macrumors.com/thread...gans-to-18gb-and-beyond.2482947/post-34613425 )
Code:
00:00.0 Host bridge [0600]: Motorola MPC106 [Grackle] [1057:0002] (rev 40)
	Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
	Status: Cap- 66MHz- UDF- FastB2B+ ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR+ INTx-
	Latency: 0, Cache Line Size: 32 bytes

00:0d.0 PCI bridge [0604]: Digital Equipment Corporation DECchip 21154 [1011:0026] (rev 02) (prog-if 00 [Normal decode])
	Control: I/O+ Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR+ FastB2B- DisINTx-
	Status: Cap+ 66MHz+ UDF- FastB2B+ ParErr- DEVSEL=medium >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
	Latency: 64
	Bus: primary=00, secondary=01, subordinate=02, sec-latency=32
	I/O behind bridge: 00001000-00002fff [size=8K]
	Memory behind bridge: 80800000-80afffff [size=3M]
	Prefetchable memory behind bridge: 0000000080800000-00000000807fffff [disabled]
	Secondary status: 66MHz+ FastB2B+ ParErr- DEVSEL=medium >TAbort- <TAbort- <MAbort+ <SERR- <PERR+
	BridgeCtl: Parity- SERR+ NoISA+ VGA- VGA16- MAbort+ >Reset- FastB2B-
		PriDiscTmr+ SecDiscTmr+ DiscTmrStat- DiscTmrSERREn-
	Capabilities: [dc] Power Management version 1
		Flags: PMEClk- DSI- D1- D2- AuxCurrent=220mA PME(D0-,D1-,D2-,D3hot-,D3cold-)
		Status: D0 NoSoftRst- PME-Enable- DSel=0 DScale=0 PME-
		Bridge: PM- B3+

00:10.0 VGA compatible controller [0300]: Advanced Micro Devices, Inc. [AMD/ATI] RV280 [Radeon 9200] [1002:5961] (rev 01) (prog-if 00 [VGA controller])
	Subsystem: Advanced Micro Devices, Inc. [AMD/ATI] RV280 [Radeon 9200] [1002:5961]
	Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
	Status: Cap+ 66MHz+ UDF- FastB2B+ ParErr- DEVSEL=medium >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
	Latency: 16 (2000ns min), Cache Line Size: 32 bytes
	Interrupt: pin A routed to IRQ 255
	Region 0: Memory at 90000000 (32-bit, prefetchable)
	Region 1: I/O ports at 3000 [disabled]
	Region 2: Memory at 80b00000 (32-bit, non-prefetchable)
	Expansion ROM at 80b20000 [disabled]
	Capabilities: [50] Power Management version 2
		Flags: PMEClk- DSI- D1+ D2+ AuxCurrent=0mA PME(D0-,D1-,D2-,D3hot-,D3cold-)
		Status: D0 NoSoftRst- PME-Enable- DSel=0 DScale=0 PME-
	00008010:90000008.f0000008 Region 0: Memory at 90000000 (32-bit, prefetchable) [size=256M]
	00008014:00003001.ffffff01 Region 1: I/O ports at 3000 [size=256]
	00008018:80b00000.ffff0000 Region 2: Memory at 80b00000 (32-bit, non-prefetchable) [size=64K]
	00008030:80b20000.fffe0001 Expansion ROM at 80b20000 [disabled] [size=128K]

01:00.0 FireWire (IEEE 1394) [0c00]: Texas Instruments PCILynx/PCILynx2 IEEE 1394 Link Layer Controller [104c:8000] (rev 02) (prog-if 00 [Generic])
	Subsystem: Apple Inc. Device [106b:001c]
	Control: I/O- Mem- BusMaster+ SpecCycle- MemWINV+ VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
	Status: Cap- 66MHz- UDF- FastB2B- ParErr- DEVSEL=medium >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
	Latency: 16 (750ns min, 1000ns max), Cache Line Size: 128 bytes
	Interrupt: pin A routed to IRQ 0
	Region 0: Memory at 80a02000 (32-bit, non-prefetchable) [disabled]
	Region 1: Memory at 80a40000 (32-bit, non-prefetchable) [disabled]
	Region 2: Memory at 80a30000 (32-bit, non-prefetchable) [disabled]
	Expansion ROM at 80a20000 [disabled]
	00010010:80a02000.fffff000 Region 0: Memory at 80a02000 (32-bit, non-prefetchable) [size=4K]
	00010014:80a40000.ffff0000 Region 1: Memory at 80a40000 (32-bit, non-prefetchable) [size=64K]
	00010018:80a30000.ffff0000 Region 2: Memory at 80a30000 (32-bit, non-prefetchable) [size=64K]
	00010030:80a20000.ffff0001 Expansion ROM at 80a20000 [disabled] [size=64K]

The PCI bridge has a single pin for the 66MHz flag of both the primary and secondary status registers. It's set to 66MHz to support the 66 MHz primary link.
All the slots of the secondary bus have M66EN pulled low to force 33 MHz. The PCI bridge has a pin pulled high for the primary bus to enable 66 MHz and a pin pulled low for the secondary bus to disable 66 MHz. These pins are different than the one used for the status registers. The status register pin is more of a capability indicator rather than a status indicator.
 
Here's an lspci dump of the Grackle host, the PCI bridge, the GPU, and FireWire (using my "lspci for Open Firmware" scripts at https://forums.macrumors.com/thread...gans-to-18gb-and-beyond.2482947/post-34613425 )

Do the Paddington I/O chip and the UDMA controller not show up because they don't have the necessary metadata? (Just curious, since they are also technically PCI devices sitting behind the 21154 bridge at the same level as the firewire controller and the three 33mhz/64 bit slots.) Here's a diagram, just for the LOLs.

PowerMacG3-8.gif


I've always wondered quite what Apple was thinking when it comes to the 64 bit slot connectors. Are there *any* 64 bit cards that actually work in the machine? (... other than apparently there was an AHA-29160 variant that *might* have been the card included by the factory in the "Server" variant of the B&W? I've never seen one in person.) For several years I was actively keeping an eye out for something to try just for the lols, but every 64 bit PCI/PCI-X card I ever ran into at work was keyed to only work in 3.3v slots, while Apple keyed the B&W's slots for 5v. (Except, confusingly enough, the oddball 66mhz video slot, which is 3.3v.) Considering the whole point of having the 66mhz bus segment was an attempt to *partially* negate the disadvantage of not having AGP I'm really not quite sure why they bothered with 64 bit slots on the 33mhz segment, but... I dunno, maybe it made some kind of difference for that one disk controller?

Just for contrast, here's the diagram for the original Beige G3s; basically all the stuff that's behind the PCI-PCI bridge is directly tree'ed off the Grackle and it runs the whole mess at 33mhz. This is why I compare the Grackle to a typical Pentium chipset; I guess since it has SDRAM support it's equivalent to a pretty new one like a 430TX, but it's the same concept, IE, the memory bus is the only "privileged" component under the CPU, everything else is on a shared PCI segment.
Screenshot 2026-06-04 at 12.38.58 PM.png
An original iMac is basically this except the physical PCI (and Perch, which is a PCI slot with extra steps) slots are deleted and a USB controller is tacked on.

Screenshot 2026-06-04 at 12.46.46 PM.png
This diagram points out one of the kinda amusing oddities of the iMac G3; that little CPU card with the memory on it piggybacked onto the motherboard is technically the "core computer" all by itself, the whole motherboard is basically a very complicated PCI peripheral.
 
Huh.

I knew the CPU, RAM, and ROM were on the daughterboard, but never would have thought the Memory and PCI controller were too.

Guess that means someone could make themselves a whole new motherboard if they really wanted, since most of the key components are on a removable card..

Wouldn't necessarily be an improvement, but still an interesting thought.
 
Do the Paddington I/O chip and the UDMA controller not show up because they don't have the necessary metadata? (Just curious, since they are also technically PCI devices sitting behind the 21154 bridge at the same level as the firewire controller and the three 33mhz/64 bit slots.) Here's a diagram, just for the LOLs.
pci-ata@1 is the UDMA ATA controller.
mac-io@5 is the Paddington I/O chip. It has the slower ATA buses, audio, ADB, Ethernet, etc.

I included only a subset of devices in the lspci output as examples for the primary and secondary buses. I've attached a full listing of the device tree and the parsed output of the "lspci for Open Firmware" output.

The Two zip files contain info for 3 different setups for the same B&W G3 (different devices connected to a StarTech PCI to PCIe bridge). Since the PCI to PCIe bridge works in the 33 MHz slot and the 66 MHz slot, it means that it does support 66 MHz, since the 66 MHz slot of the B&W G3 does not support 33 MHz like normal 66 MHz slots should. The dumps of the B&W G3 connected to the Netstor NA255A PCIe expansion chassis were created by my fork of pciutils and my pcitree.sh script (and GitHub and GitHub gists). I created an nvramrc script to make Open Firmware probe device numbers ≥ 0x10 behind the PCIe-PCIe bridge of the Netstor NA255A.

Considering the whole point of having the 66mhz bus segment was an attempt to *partially* negate the disadvantage of not having AGP I'm really not quite sure why they bothered with 64 bit slots on the 33mhz segment, but... I dunno, maybe it made some kind of difference for that one disk controller?
Most or all GPU PCI cards are 32-bit. Running the GPU at 66 MHz makes it faster. The PCI-PCI bridge separates this fast bus from the slow bus because PCI cards can't run at 66 MHz if any other card on the same bus is 33 MHz. The bridge chip supports 64-bit so Apple included the extra bits in case anything wanted to use the extra bits to get the same speed with 33 MHz 64-bit as the 66 MHz 32-bit slot. Apple chose 5V to be compatible with older PCI cards? Universal PCI slots to support 3.3V and 5V would have been nice but that would require more current for the limited 3.3V power supply? Apple chose 33 MHz for the secondary bus because 66 MHz to multiple slots is more difficult than to a single slot?

Just for contrast, here's the diagram for the original Beige G3s; basically all the stuff that's behind the PCI-PCI bridge is directly tree'ed off the Grackle and it runs the whole mess at 33mhz. This is why I compare the Grackle to a typical Pentium chipset; I guess since it has SDRAM support it's equivalent to a pretty new one like a 430TX, but it's the same concept, IE, the memory bus is the only "privileged" component under the CPU, everything else is on a shared PCI segment.
Original Beige G3s don't have a PCI-PCI bridge. Grackle is the memory controller and the PCI host controller. The B&W G3 overclocks Grackle PCI to 66 MHz.

Non-grackle based Power Macs may have two or three PCI host controllers (one of them may be AGP).

This diagram points out one of the kinda amusing oddities of the iMac G3; that little CPU card with the memory on it piggybacked onto the motherboard is technically the "core computer" all by itself, the whole motherboard is basically a very complicated PCI peripheral.
The PCI bus from the grackle seems to be a convenient bottleneck at which to separate the two parts of the iMac.
 

Attachments

Back
Top