• Hello MLAers! We've re-enabled auto-approval for accounts. If you are still waiting on account approval, please check this thread for more information.

Beige G3 onboard cache

stynx

6502
Gossamer's Tale
The quest for onboard cache

I have been looking around for hints on how to install the 2nd-level cache slot on the beige G3. There is unpopulated component, most likely a socket, that is labeled cache and is connected to the Grackle bridge. The pinout is in the schematics.
Sadly, i could not find a source for the cache slot component. But i think i have found the origin of the cache socket and cache-dimm.
s-l1600-4.jpg
There is a cache module from the IBM RS6000 7043 p-series that would fit. It has the VCC and GND pins at the correct locations and is (at least) 66MHz capable. The cache size is 1MiB and would perfectly fit into the timeframe. The Grackle controller, the VRM and the CPU-ZIF socket are also identical to the Gossamer board.
The IBM cache-socket is a bit longer and has an addition that is seemingly not used. The cache-socket also lacks the second notch, which the cache-module has. Maybe IBM had intended to use a variety of different cache modules.
s-l1600-6.jpg
Since i live in Germany, any attempt to buy a RS6000 board to get my hands on a cache-socket have failed due to enormous prices the sellers slap on for shipping ($100 for the board with CPU and cache, $700 shipping). Im still looking into getting a cache module to (maybe) manually solder it in. It would make a comparison between G3 and 604e more interesting and maybe the onboard-cache could be used with the G3? There are hints, that a 4MiB cache module from IBM has existed but maybe that's what the extension is for on the socket.
 
Here is a PDF from Motorola about the cache module
edit: The cache described in the PDF is not the same as the IBM cache. The PDF describes a 178 lead package while the IBM cache is a 182 lead package (like the Gossamer socket)

512KB and 1MB BurstRAM

Secondary Cache Modules for

PowerPC and PReP/CHRP Platforms
 

Attachments

Last edited:
The socket part number is
Burndy ELF182JSC–3Z50

Here is a comparison between the cache module pinout and the Gossamer pinout:

Bildschirmfoto 2025-04-04 um 12.39.21.png

Bildschirmfoto 2025-04-04 um 12.46.16.png
 
Last edited:
Gossamer's Tale
The quest for onboard cache

I have been looking around for hints on how to install the 2nd-level cache slot on the beige G3. There is unpopulated component, most likely a socket, that is labeled cache and is connected to the Grackle bridge. The pinout is in the schematics.
Sadly, i could not find a source for the cache slot component. But i think i have found the origin of the cache socket and cache-dimm.
View attachment 85181
There is a cache module from the IBM RS6000 7043 p-series that would fit. It has the VCC and GND pins at the correct locations and is (at least) 66MHz capable. The cache size is 1MiB and would perfectly fit into the timeframe. The Grackle controller, the VRM and the CPU-ZIF socket are also identical to the Gossamer board.
The IBM cache-socket is a bit longer and has an addition that is seemingly not used. The cache-socket also lacks the second notch, which the cache-module has. Maybe IBM had intended to use a variety of different cache modules.
View attachment 85182
Since i live in Germany, any attempt to buy a RS6000 board to get my hands on a cache-socket have failed due to enormous prices the sellers slap on for shipping ($100 for the board with CPU and cache, $700 shipping). Im still looking into getting a cache module to (maybe) manually solder it in. It would make a comparison between G3 and 604e more interesting and maybe the onboard-cache could be used with the G3? There are hints, that a 4MiB cache module from IBM has existed but maybe that's what the extension is for on the socket.
The RS6000 looks a bit like the unreleased Starmax 6000 board. I guess it’s no accident they’re both numbered 6000.

By the way, wouldn’t a cache socket from a PM7300-9600 machine physically fit the Beige’s cache slot footprint?
 

Attachments

  • IMG_3477.png
    IMG_3477.png
    1.8 MB · Views: 19
Last edited:
I mean, they did use the same socket as the Beige/B&W G3s. That's the source of the higher clocked 604 ZIFs. Probably the leftover developments from whatever CHRP work they must have helped with, I'd assume.

I've looked at the cache in those RS boards too and they 'seem' similar, but I didn't want to take the chance of frying something, if it managed to physically fit.
 
By the way, wouldn’t a cache socket from a PM7300-9600 machine physically fit the Beige’s cache slot footprint?
No, the older cache slots are 5V while the G3 is 3.3V
 
Interesting, I hope you can get some parts to test. They do look like different pins/spacing than these:

68kmla.org/bb/index.php?threads/6360-6400-6500-l2-cache-modules.38450/
 
Interesting, I hope you can get some parts to test. They do look like different pins/spacing than these:

68kmla.org/bb/index.php?threads/6360-6400-6500-l2-cache-modules.38450/
The cache in the G3 has 182 "pins" while PCI mac cache has 160 "pins". The G3/RS6000 cache module is a bit bigger.
 
Back in those prep/chrp days, gossamer and other mainboards like yellowknife or long trail etc. were designed with both 604e and 750 in mind. RS/6000 had 601 and 604 and some RS/6000 were based on CHRP. Thus, a L2 cache would be beneficial for 604e based boards. I worked with a Tatung 6000 prototype board and with 604e CPUs in its socket, a L2 was more or less mandatory. But the operation manual clearly stated that with a 750, the L3 should me removed. However, when later G4s got faster and the bus interface (601 or max bus) did not really develop further, Apple introduced L3 caches on first on some higher end G4 quicksilver/DA machines and on all MDDs to offset the gap between L1, L2 and the comparatively slow 133/167 Mhz maxbus to Ram. However, with the much slower early G3s and a thus much slower gap between CPU, L2 and Ram, the L3 might be creating more overhead than benefit.
 
Mouser has a variant as well: -- correction -- It is possibly not the correct slot since it has 41-48pos while the needed slot has 42-47pos
I have bought two slots and they fit. It is the same as the Perch slot and the same as the PPC PDS slot. I will try to solder the slot in at the weekend.
Ok, i just tried fitting one of the slots into the Gossamer board. It did not fit. The number of pins is ok, the spacing is ok BUT the pin arrangement is wrong. The ins are opposite to what they should be. I have begun to "shift" the pins one pin "up" but that is much work and the amount of pins i have bent beyond repair is also too high.
 
Last edited:
DON'T TRY THE L3-CACHE MOD!

I have soldered in the socket after painstakingly assembling it and after installing the L2-cache module....
The module released smoke and a hissing noise....

I instantly pulle the plug but the G3 would not boot without the L2 cache module, just a blank screen.

I tinkered a bit around and found out that only the Rom-card was dead.

That was a close one...

The cache module has seemingly some pins different than the Apple slot although i had compared the VCC and GND with the Gossamer schematic. I seem to have overlooked something.
 

Attachments

  • IMG_1597.png
    IMG_1597.png
    16 MB · Views: 18
  • tempImagerXx9Li.png
    tempImagerXx9Li.png
    13.5 MB · Views: 23
Last edited:
Here is the damages cache module.

I just had a thought:
Maybe the engineers at Apple f'cked up and did the pinout on the socket wrong and never really had to correct the issue since the cache slot was never used?
I have only found reserved pins in the effected area that are pulled high (3.3V).
 

Attachments

  • IMG_1601.png
    IMG_1601.png
    16 MB · Views: 21
  • IMG_1602.png
    IMG_1602.png
    15.8 MB · Views: 19
  • IMG_1603.png
    IMG_1603.png
    16.2 MB · Views: 11
  • IMG_1604.png
    IMG_1604.png
    16.5 MB · Views: 10
Back
Top