PowerPC overclocking and limitations

[jessenator]

Maybe we could compare boards and see what's populated on each variant?

That seems like a good place to start. I'll snap some shots of my 3 boards tomorrow (hopefully a 4th, but I'm positive it'll be the same as my 4400/200 board). I have that 4400 board, a 5000/225 (originally), and a 3000/200 StarMax boards.

 

[jessenator]

Here's the vanilla Power Macintosh 4400/200 board sections: https://imgur.com/a/jHZDyuf

The Motorola StarMax 3000/200 board sections: https://imgur.com/a/1l7rPTj

The Motorola StarMax 5000/225 board sections: https://imgur.com/a/V9BLfwE

 

[Franklinstein]

You've already got the basic 603e models covered, so I'll just get a couple shots of my 160 and 200MHz 604e-based models (ComJet PowerCity DT, aka Starmax 4000 DT) as soon as I can dig them up.

 

[jeremywork]

I've got a running 6500/300, in case it would be useful to reference anything from it. I'd imagine it won't answer nearly as many questions as a Starmax 5000/300 would, but feel free request any pictures you may want of its board/cpu.

 

[jessenator]

I've got a running 6500/300, in case it would be useful to reference anything from it.

Thanks for offering! Unfortunately, the Gazelle board architecture is different from the Tanzania :/  But, I mean, all these mods are really doing is adjusting the clock multiplier.

I actually looked it up and you can do similar overclocking techniques on the Gazelle boards, however, they are not nearly as user-friendly—if you can call soldering pin headers on the Tanzania boards "user-friendly." The resistors involved are directly under where the 603ev BGA is soldered on, and are extremely close together.

 

[jeremywork]

Ouch- glad I have no reason to mess with a natural 300.

The fact that you can fit a standard jumper header on the Tanzania board is pretty cool. You could even add in something like a SCSI ID selector switch on the case panel, so you could set the target speed between startups (if you could run at 125-300MHz stable, you'd have one machine which could mimic *most* of the performance range between the 601 and G3.)

 

[Franklinstein]

I haven't been able to upload pictures but I did investigate my Starmax 4000 and PM 4400. The differences, other than the CPU OPTION block which seems to only set CPU multiplier, seem to be in the CPU ID section. I think this is what configures the base system IDs and functional devices and doesn't have anything to do with the processor directly. 

On the Starmaxes, both your 603 and my 604 models, R255, R256, R231, R236 are set.

On the 4400, R254, R252, R231, R236 are set.

So there are two resistor sets that are different. The only real differences between these Starmaxes and the 4400 is that these use VGA instead of DB15, they have PS/2, and they have 3 PCI slots with no CSII. I'm not sure if the CPU ID block controls any of this or not or if they just set a gestalt ID or something. I'd be interested to find out if changing any of them will enable the integrated Ethernet. Sure it's only 10bT and you'd still have to solder in the jack and a bunch of discrete components, but it could be useful in freeing up a PCI slot. I guess we'd need to look at a board that has its integrated Ethernet to find out.

 

[jessenator]

On the Starmaxes, both your 603 and my 604 models, R255, R256, R231, R236 are set.

On the 4400, R254, R252, R231, R236 are set.

So there are two resistor sets that are different.

Interesting. I looked it up and while the 4400 (both 160 and 200 versions) have a Gestalt ID of 515. I couldn't find one for the StarMax online, but my guess is that they're the same.

Did you see any differences on the FEATURE OPTION jumper section between your boards and mine?

FWIW, every time I hooked up any of the boards I have with my 7.6.1 CF drive it would just say it's a Power Macintosh 4400. I can't be sure, but I think by 7.5.3 the model-specific icon and name was phased out of Macs, but the vanilla 7.6 StarMax CD does have its own in the "About this Macintosh" window, which I assume is handled by the registration/license INIT that comes in the System Folder of Motorola install media. I don't recall what it says in Apple System Profiler (I've installed a later version on my CF drive, which is where the above ID of "4400" shows).

I would further hazard a guess that the VGA and on-board ethernet is more of the board components, but I too would like to see a board with it.

There are also two board P/Ns I've seen and wondered if there were more. My boards are either 820-0881-A (4400, 3000) or 820-0880-A (5000).

 

[Franklinstein]

Ah, the FEATURE OPTION block. I missed that one; I'll have to check it soon. Which board is your picture from? It's white lettering so I would guess the 4400. That's probably where the Ethernet and other features are enabled, if I had to hazard a guess. If you had the necessary parts to solder in you could probably also enable MFM floppy drive support from there too if you really wanted to.

So then I wonder what CPU ID sets. I mean it's the same setting on both of my Starmaxes and the one you have, and they're not only different models (4000 and 5000) but also running different processors (603ev and 604e). The bus speed is set sort of between the PSX chip and the VRAM module, so it's not for that. Too bad these things don't have much documentation.

 

[jessenator]

It's white lettering so I would guess the 4400.

Yes. Here's all of them. Strange that the StarMax boards are the same, though they're different models (and different board numbers.. so much information to string together without documentation... I feel like a conspiracy theorist with a strung cork board  :lol: )

And it's doubly confusing as there's no chart, at least that I could tell, like there is for the CPU PLL (at least on the StarMax boards)—nothing to indicate what it would opt for. Did any Tanzania clones, that you know of, use MFM floppies? Or is it just a holdover from the CHRP directives?

Also, I guess I didn't quite catch what you were describing with your description of the CPU ID and CPU OPTION sections among your boards. The OPTION block is most definitely where the PLL/clock multiplier is set. Do the resistor arrangements differ from OPTION to ID on the same board in your collection? Mine match. I'm guessing that the similarities between your 4000 and 5000 boards are because they're set to the same multiplier? I'm assuming that based on the screen-printed chart on the non-4400 boards:



Like you say, I wonder what CPU ID really does... I'm clutching at straws, since I'm only a few years into starting to understand how a logic board and its components all work together. It's probably not memory timing, since that's controlled by the bus speed (or at least the relative sections discovered by Andreas Kann), but I wonder if there's anything else that would be effected by the CPU clock speed and/or the clock multiplier... serial? video? PCI? Again, clutching at straws there.

Which, as tenuous as it is, kinda goes back to my theory as to the reason for my black-screen clocking attempts. I wonder if the disparity/difference of CPU OPTION and CPU ID is what's causing that, and if they were made identical... but that's a stretch, seeing as (with working RAM anyway) I can achieve a stable OC (and even stable operation at any clock, save the one below the min of the chip) without a match on the sections... I'm tempted to really just go all out with one of my boards to test it. I ordered some jumper blocks to solder in with my last mouser order, I just don't know which one to choose. I've already modified the 5000 board, but I have a 3000 board that really isn't doing much of anything. I wish my second 4400 board didn't just halt booting after the chime, otherwise I'd just use it. Just see whether or not instability occurs with a large disparity, whether a max multiplier can be stable when they match, etc.

 

[trag]

Turns out Wakefield-Vette still makes PowerPC specific heatsinks, and one where the clip fits, although instead of the X-pattern, it's two parallel clips.

From Mouser, or from DigiKey.

That's a nice find on the heat sink.  Unless I'm mistaken, that is also the original heat sink used on the Power Computing Power 120;  PPC601 running at 120 MHz.    The CPU is opposite one or two of the NuBus slots and the later, cage type, heat sink sticks up too high and interferes with long NuBus cards such as the JackHammer SCSI card.   But that low waffle griddle heat sink, allows clearance for long NuBus cards.

As far as I can tell, most of the Power 120s shipped with the taller interfering heat sink.   Power Computing had thermal problems with that model, so maybe the "cage" heat sink cools better.    Anyway, the low flat one is hard to find.   Either in rare early models of the Power series (reportedly there was a Power 80 and Power 100  as well, though I've never seen either) or a few appeared on the CPU module for the PPC601/120 equipped original PowerCenter.

 

[trag]

Possibilities for CPU ID, some of which you've already suggested:

1)  CPU power supply voltage

2)  Cache ratio (the ratio of the cache speed to the CPU speed), although with it on a separate module from the CPU (?) it may just run at the bus speed

3)  Clock ID.  It sounds like Kann found this somewhere else, but I think it is worth noting that on the 7500/8500/9500 and similar models, the CPU card has three pins that are used to tell the motherboard what bus speed is being used.

4)  Actual CPU model, however, unless there are several pin compatible CPUs which require different bus handling, this would make little sense.    The logic board must already be laid out for the specific ball pattern of the chosen CPU, so any other configuration can be built into the logic board.  In other words, there's no point in being able to tell  the logic board it has a PPC604e, if there's no way to solder one to the board.

 

[Franklinstein]

As far as I can tell, most of the Power 120s shipped with the taller interfering heat sink.   Power Computing had thermal problems with that model, so maybe the "cage" heat sink cools better. 

Apple's >80MHz 601-based models, except for the 7100/80 and 7500/100, and possibly some of the 7200/90s, used Peltier junctions under their heat sinks. The 9150s and (most) fast 7200s used low-profile waffle-grid heatsinks, though in a different shape to these examples here, with Peltier junctions. The 8100/100 and 110 used Peltiers but just had the normal cage-type heatsink attached. I don't know if PowerComputing used Peltiers or not; I don't have any of their 601-based models.

CPU OPTION is indeed where the processor's clock speed is set on these. These are all different depending on what speed it is (160 or 200 for mine, 240 for yours) with no corresponding difference in CPU ID settings.

All caches used in these computers were board-level so they sat on the 60x bus at 40 or 50MHz. Multi-processing was never an option in these or any other computer with the PSX or slightly modified PSX+ memory/PCI controller (Alchemy, Gazelle, Tanzania, and PCI-based PowerBooks up to the WallStreet).

The BGA 603 and 604 appear to have shared the same grid (as did the 740, though of course no Macs or upgrade cards ever used the 740). It's possible they had different voltages, but since the CPU ID blocks of both 603 and 604-equipped Starmax boards seen here have the same configurations, it's not likely, or at least the CPU ID block doesn't affect it. Reading more on Andreas Kann's website I am reminded that the bus frequency is changed by replacing an oscillator and then changing a set of resistors ("... removing R29 and installing it in the R28 location for 50 MHz operation.") so apparently CPU ID has nothing at all to do with this either. So what does it do? It's still a mystery.

FEATURE OPTION is probably where everything to do with features is set, assuming CPU ID doesn't do much other than possibly set a gestalt ID or something. But I can only guess at this until we have a collection of boards with different features installed and pictures of each FEATURE OPTION and CPU ID configuration, or someone decides to solder in a jumper block and start testing. Personally I've never seen an MFM floppy actually used on any PPC machines, though I don't remember exactly what my PowerStack II looks like inside (it looks similar but it's not actually a Tanzania board). I'm not even sure Mac OS supports MFM drives; it must have been for CHRP/PReP machines, though even the beige G3's Gossamer logic board has an MFM spot. In the '90s you could get Windows NT 4.0, OS/2, Solaris, AIX, possibly BeOS, and various Linux flavors to run on your non-Mac PowerPC workstation. One of them may have preferred an MFM floppy drive.

 

[jessenator]

But I can only guess at this until we have a collection of boards with different features installed and pictures of each FEATURE OPTION and CPU ID configuration, or someone decides to solder in a jumper block and start testing

The only thing stopping me is what the defaults on FEATURE OPTION would be if set with pins rather than resistors, really. Because I'd have to remove them, i think... If I don't I'll gladly just go for it, but I have a feeling I'll need to.

The clock multiplier settings are thankfully laid out, so some board deciphering is in order. I'll have a closer look, front and back to see if I can even take an educated guess and then someone who knows can correct me

 

[jessenator]

Random thought: was the 32-MB-single-bank-DIMM (on the first memory slot) a pan-model limitation, or did that only apply to the PM4400 models? (thumbnails)





It looks like that's where traces are going, but that's amateur hour guesswork.

One other thing, it looks like the CPU ID is entrangled with the PCI bus?

 

[jessenator]

The only thing stopping me is what the defaults on FEATURE OPTION would be if set with pins rather than resistors, really. Because I'd have to remove them, i think... If I don't I'll gladly just go for it, but I have a feeling I'll need to.

Judging by the arrangement of the resistors and comparing that to the jumper arrangement for the default (200 MHz) CPU OPTION on the PM4400, the lower portion of this image is my assumption of what the jumpers should be for an 18-pin jumper block on the FEATURE OPTION section. Can anyone confirm I'm on track there?

 

[Franklinstein]

It's super weird how the jumpers are laid out, for sure. Excellent illustrations though. I'll see if I can't hit mine with a multimeter to try to tease out where the resistors go in relation to the jumper block.

CPU ID may affect things built in to either the PSX+ memory/PCI controller, or the O'Hare (I think) multi-function I/O chip, which lives on the PCI bus, so that could explain why traces are wrapped in with PCI lines.

The PSX/PSX+ memory/PCI controllers support both buffered (in Alchemy and Gazelle boards) and unbuffered (in everything else) EDO memory, so CPU ID may tell it what kind of memory is present. In most of the PSX/PSX+ equipped Macs, there are generally two RAM slots (Alchemy, Gazelle), sometimes with onboard RAM (Alchemy, PowerBooks), sometimes not (Gazelle, Tanzania). The exceptions are the PowerBooks (one RAM expansion slot) and the Tanzania (three RAM slots, one with single-bank limitation), and both of these machines have a max of 160MB of RAM. I assume the onboard RAM on machines so equipped lives in the single-bank zone of the controller, leaving the RAM slots available for double-banked DIMMs. So the limiting factor here is the PSX/PSX+ and there's really nothing to do to work around it.

 

[jessenator]

Yeah, some of the positions are wacky... I'm even assuming the jumpers go linearly, just based on what exists elsewhere on the board. Also got to second-guessing my assumption.

 

[jessenator]

Well a little bit of an update, laterally. Since discovering a RAM-specific testing tool, I wanted to run it on this StarMax board overclock.

When I was getting it up and going, I noticed that I had some odd system freezes, which I thought  had been resolved due to me not using the funky IBM 64MB RAM that only registered as 32 MB per DIMM. I tried multiple configurations from what I had (minus the original config which is in the pristine , stock 4400) and even went so far as to buy new RAM for this second working unit. However, even the new RAM would freeze in a different manner than as I was experiencing during this recent bout and during initial testing—very odd.

I swore that the 275 MHz setup would be completely stable full loaded and during operations, but looks like that one is out of reach as well. While doing the Apple Personal Diagnostic (on the RAM short test), I got 93% before a freeze and then only 17% before a freeze on restart. I swapped pins to bump it down to 250 MHz and it's looped almost 10 times now without a crash or freeze.

I didn't note what I was using, so I don't recall if I had the full compliment or if I was running just a single chip, but I'm leaning on the full population, as my config of 7.6.1 uses 11 MB. Well, a bit of a bummer, but at least it's working. I only halfway feel stupid for buying new RAM, but at least it's more or less standard vs those weirdo IBM DIMMs I bought a while back.

 

[jeremywork]

I had a similar experience when testing RAM in my 8100 under 8.5 and 8.6- the RAM test would freeze at various points, and occasionally produce a failure message regardless of which modules I had swapped in. To my surprise (after much frustration) I found that Apple Personal Diagnostics would completely pass the RAM test when I was booted from a retail 7.5.3 CD, even with the same memory installed as had crashed in 8.6 (by this point I had learned how to squeeze my hands in to swap modules without removing the logic board.)

I'd try an older OS just to be sure it's not a failure of the diagnostic itself- I haven't yet tested to see if this is true on other machines.