• Updated 2023-07-12: Hello, Guest! Welcome back, and be sure to check out this follow-up post about our outage a week or so ago.

Mach 5 G3

waynestewart

Well-known member
I have a few G3 upgrades but one of them would usually freeze up during the boot process or immediately after. It's a Sonnet 450/1mb. I was tempted to toss it but I thought it may be useful as parts. Today I had my 8600/300 out and thought I'd try the card. It seems to work fine with this one machine, even ran at 500mhz until I installed the Sonnet software. Anyone heard of G3 cards that only run in Mach 5 machines?

 

Solvalou

Well-known member
It's nothing unusual that some accelerators work better in some machines then in others, even when the box tells you it's compatible. Glad to hear it works for you though. :)

 

Unknown_K

Well-known member
On PCI Powermacs you are supposed to remove the CPU cache SIMM for them to work correctly. The 8600/300 has the cache on the CPU and not on the motherboard so it works correctly.

 

waynestewart

Well-known member
So you're saying the empty cache slot in the 7300 and 8500 is causing the Sonnet 450 to not work? Sonnet 400/1mb cards work in my 9600 and 9500 but the 450/1mb doesn't.

On PCI Powermacs you are supposed to remove the CPU cache SIMM for them to work correctly. The 8600/300 has the cache on the CPU and not on the motherboard so it works correctly.
 

ClassicHasClass

Well-known member
No, the opposite. The cache slot should be empty so that it does work. I've used Sonnet 400/1 and 500/1 G3 cards in my stable of 7300s with no problem, even under NetBSD. It makes me wonder if there's a fault with the card that the 8600's architecture is glossing over. trag might know.

 

trag

Well-known member
It sounds like there is something marginal about that CPU card. The presence/absence of a cache can affect timings/capacitance/loading of signal lines, so that's a good thing to consider.

The 9500 and the original 9600 have the cache soldered to the logic board and it is not removable, so the presence of that cache could be why it doesn't work in the OP's machines, unless he's talking about a 9600 Enhanced, which is also a Kansas/Mach 5 machine.

On the 9500, removing resistor R31 (IIRC, better do a search first) will disable the on-board cache. The Umax S900 was civilized enough to include a jumper position for this.

Other than the lack of a cache there really aren't any other substantial differences between the x500 CPU slot and the Kansas/Mach 5 slots. I've never actually traced out the slots for the Kansas machine (as I did for the other x500) but the fact that third party upgrade cards work in both means that they must be almost identical. My guess has always been that there's simply more 3V (or maybe 12V) supply lines in the Kansas slot.

If you have an ohmmeter handy, the other thing you can do is check the CLKID pins on your upgrade cards to see if they're all biased the same. This is a little farfetched, but could be a factor if timings are tight. And there were all those upgrades that made us think that logic boards wouldn't work properly above 45 MHz because they had the CLKID pins set wrong, way back when....

You can find the identity of the CLKID pins here: http://www.io.com/~trag/Apple_pinouts/CPU_Slot_Pinout They're either tied to ground, or (IIRC) floating. In any case, check for connections to GND, 5V or nothing and compare amongst the cards.

If the card works until you install the Sonnet software, then most likely the card has a faulty backside cache chip. The software for G3 upgrades activates the cache (among other things) and the most common problem one sees, if the card crashes when the software loads, is that something is wrong with the cache chip.

I once had a NuBus G3 upgrade from NewerTech with that problem, and when I pulled the heat sink, I found that the assembler had dribbled heat sink grease across the cache chip pins.

If you're really ambitious, compatible cache chips are available from Digikey for $15 - $20 each the last time I checked. I think they're 100 pin chips.

 

Byrd

Well-known member
You might also want to check your RAM in your 9600/9500 machines - noting that accelerator cards work better with 50ns, 2K refresh chips (I might be wrong on these numbers - noting I am at work, can't check anywhere!). You may find some of the RAM in your other machines is rated to 60ns, 4K.

JB

 

waynestewart

Well-known member
I tried the card on 4 different machines 7300, 8500, 9500, and 9600 before I happened to dig out the 8600 Mach 5. It shouldn't be a cache issue since the 7300 and 8500 don't have cache cards. I used to get machines that were too old to sell from a local used Mac store. At the time that was B&W G3s and older. Quite a lot of the 75-9600 machines had G3 cards and extras like 128mb RAM, SCSI cards, ATA cards, Video cards etc. A lot of stuff I passed on to a local school and it's students, mostly G3s. The one thing I kept was all the 128mb RAM from the 75-9600s. While I do have some 60ns 128mb RAM I sorted that out and keep it separate from the 50ns. In this case all the RAM I'm using is 50ns. All the machines I tried the Sonnet 450/1mb in already ran fine with slightly slower Sonnet G3 cards, mostly 400/1mb cards. I don't have any other cards faster than 400mhz though I do have some Powerlogic and XLR8 ZIF cards. I guess I could try a higher speed with one of them and see what happens.

 

trag

Well-known member
IIRC, some of the upgrades which work in both kinds of PCI Mac (Kansas & non-Kansas) had a jumper or switch to set depending on whether it is installed in a Kansas machine or not. Could your upgrade have such a switch which is set to the Kansas position?

 

waynestewart

Well-known member
No jumpers on the card.

IIRC, some of the upgrades which work in both kinds of PCI Mac (Kansas & non-Kansas) had a jumper or switch to set depending on whether it is installed in a Kansas machine or not. Could your upgrade have such a switch which is set to the Kansas position?
 
Top