PowerPC overclocking and limitations

[jessenator]

I'm wondering if PowerPC CPUs can be overlocked in the more modern/traditional sense of being clocked beyond the given chip's rated/indicated frequency with a sacrifice of increase voltage/wattage and need for more robust cooling solutions. Case in point, the 603e(v?)* onboard the Starmax 5000 motherboard. Reading the p/n on the CPU itself, XPC603PRX240LD, I'm led to CPU-World's page which indicates the rated frequency of 240 MHz: http://www.cpu-world.com/CPUs/PowerPC-603/Motorola-XPC603PRX240LD - XPC603PRX240LE.html

This is related to a long (mostly)monologue post of mine here (which should've been in Hacks/Development to begin with):





Now, I can reasonably hypothesize that this should work, because another curious individual overclocked their 4400/160 first to 200 MHz, then to 250 MHz with a bus speed bump: https://museo.freaknet.org/gallery/apple/stuff/mac/andreas.kann/44002.html which is 125% of the CPU's rating (at least per my own 4400/200 board: XPC603PRX200LD).

But I'm wondering if anyone has any hard-line, "no, this CPU won't go beyond its indicated amount, it will just crash" kind of statement to this point. I mean, I could just try it and maybe there wouldn't be any adverse effects from doing so, but I'd rather save a motherboard and disassembly time by asking and getting an answer from someone who knows.

In his log, Andreas notes he put a fan on the heatsink, but the heatsink in my "naturally aspirated" 4400/200 has the anemic-looking, single-piece-of-sheet-metal, token heatsink, whereas the Starmax has an omni-fin aluminum design. Was the 4400/160 sold overseas different I wonder? Would I need to give it some additional cooling augmentation if OC'ing by 60 MHz (25%) if that's possible?

*SonnetTech's Metronome displays the CPU as a 603ev 

edit: yes, some machines don't have easy methods for accomplishing an "overclock," but with the Tanzania motherboards, there's a relatively simple method of swapping around resistors on the 'CPU OPTION' portion of the board to alter the multiplier. I'm more focused on whether or not you can push a CPU's clock beyond its rating, not the method used to get there.

 

[johnklos]

PowerPC CPUs are overclockable, yes. Good cooling is always a good idea, overclocked or not. You see a not very fancy aluminum heat sink on many of the 601 and 603 CPUs because they really didn't draw much power and therefore didn't generate much heat.

 

[Byrd]

If the CPU you have is rated for 240Mhz and you can find out more on adjusting the multiplier from 4X to 6X, then yes there isn't any reason to expect it won't work at this speed.  Heat output would be slightly increased.  I am not familiar with the Tanzania case cooling, but you'd assume it not that great, and the faster models may have had a slightly bigger heatsink or fan nearby to accomodate this (as per the crappy heatsink and fan combo in faster 5x00/6x00 603ev machines).  Bus speed overclocking can result in more quirks to your system, so I'd avoid this unless you can't find out the 6X multiplier pads to adjust, and stick to the simpler 200Mhz resistor mod as described.

 

[jessenator]

PowerPC CPUs are overclockable, yes.

Thank you

Bus speed overclocking can result in more quirks to your system, so I'd avoid this unless you can't find out the 6X multiplier pads to adjust,

For sure. The Starmax board is thankfully already at 50MHz vs the 4400's. Thankfully, herr Kann already did the math for the resistor combinations as well, though he mentions using different values than what the SMDs are stock, which is strange.

As far as heatsinks go, the Starmax one is nearly on par with what Sonnet puts on their cards, just no purple

 

[johnklos]

BTW - imgur sucks. First, as we can all see, embedding images from there doesn't work, and second, they want to take hundreds of megabytes and billions of instructions per second to run their Javascript just to load their silly pages.

 

[Daniël]

BTW - imgur sucks. First, as we can all see, embedding images from there doesn't work, and second, they want to take hundreds of megabytes and billions of instructions per second to run their Javascript just to load their silly pages.

Ehm, Imgur doesn't suck, it's a great place to upload photos to link on forums, you don't need to go to their site. The embedding problem is on 68kMLA's side, it won't embed any images regardless of source, nor will it succesfully upload images.  

 

[jessenator]

Yup. It's in the bug tracker, but we're in Q4 of 2019 and our admins have FT jobs outside of 68kmla, so it's click-through for now.

 

[jessenator]

Ok... so after thinking my 5000 board was bricked, I've brought it back to life! I didn't think this would work, but I'm glad it does. Instead of messing with the SMD resistor garbage, I saw this on the board, which I don't think I paid much attention to initially, probably because it's not printed on the Power Macintosh 4400 board at all! it must've been on the boards Apple made for Motorola and that's it:

So hmmmm... 12 pin jumper selector referenced for adjusting the clock multiplier, not unlike what one might find on an Intel SocketX board of the day... what's on the Tanzania I and II boards?

So while Andreas Kann's tinkering was only with the irritating SMD resistors, one could simply follow the chart with jumpers... :facepalm:

I scavenged the pin array from a broken CD-ROM. It took some tinkering to get it to boot, but bear in mind, I was writing this off completely—I had no idea that this would actually work. I was going gung ho and wanted the full speed, but for some reason it doesn't like that. Just a black screen, and probably no chime (I'm just testing the board with no accessories). But the "220 MHz" position worked. Turns out the chart is set up for the 40 MHz clock of the Tanzania I board spec, so I was trying to push it to 300 MHz.

Once I got the jumpers set to the 200 position (initially, which was 250 on this board) I got a blinking disk! Holy $#!@ it worked! I stepped it up to the 5.5 multiplier and viola!

Regarding the 300 MHz non-posting situation. I wonder if there's more going on with this board though. There are two sections with identical resistor and jumper grids. The first is the CPU OPTION section, already referenced. The second is labeled CPU ID. Same resistor arrangement and "jumper" setting as the CPU OPTION. (these photos from the stock StarMax board.

Do these need to be set identically for optimal operation?

On my modified StarMax 5000 board, which is now overclocking to 275 MHz, the jumpers are mismatched now, but it's booted fine. I wonder if the disparity is what's causing the 300 MHz to blank. For reference, the board was originally set for 225 MHz, a multiplier of 4.5x, but the CPU OPTION settings of 5x (250 MHz) and 5.5x (275 MHz) have both booted. I haven't tried a slower one, just for kicks. CPU World indicates the minimum speed for this particular chip is 125 MHz, a 2.5x for reference.

If I removed the resistors on that CPU ID block, soldered in some jumper pins, it should theoretically accomplish the same task, because those resistors are just doing the job of a jumper, yes?

Anyhow, I wonder if matching the jumpers would enable the machine to run at 300 MHz. I'm curious, because the Power Macintosh 6500/300 was technically the first 300 MHz consumer computer, and although it uses the Gazelle architecture, it's still running at the same bus speed as the Tanzania II—50 MHz, so unless there's an undocumented 603ev chip somewhere (CPU World lists that XPC603PRX240LD as the fastest), this one should be able to reach those speeds.

 

[Byrd]

Great news!  Another Mac back from the dead.  You've likely hit the limit of your 240 @ 275Mhz - the usual symptoms of no chime, raster says it's not going any further.  Run some benchmarks and stability tests @ 275 and see if it holds up.  I recall 603ev CPUs you were lucky to get 25 - 50Mhz out of.

 

[jessenator]

I really want to know what the heck was in those x500 machines to get them to 300...

A week ago when a friend sent me all those 30-pin SIMMs, he also bundled a few other things, one of which was a perfectly sized fan to sit atop the much improved clone heatsink (it actually has dimension and fins on it).

Yeah, my plan is to run MacBench tonight and see if I can keep it running at 275.

 

[Byrd]

I really want to know what the heck was in those x500 machines to get them to 300...

 A better binned CPU and later revision.  275Mhz is decent from this age of PPC overclocking - wasn't yours originally downclocked from 160 (but marked as 240)?

 

[jessenator]

I wish I knew what it was.. I guess I could troll digikey or some other supplier and I've might show up...

I think the case for the euro 4400/160 (was there a Pacific market 7220/160?) had a 200 downclocked to 160, but I could be off on that. The StarMax line was far more prolific. Afaik, the 4400/7220 only had the one, maybe two, configurations, but Motorola went all the way to the 300 MHz barrier, just a few weeks after Apple, which was probably a power play if we're honest...

They even implemented a 604e/200, all on the same board arch. I'd love to have one of those boards. Too bad they've probably all gone to eWaste purgatory.

This board I brought back was a 225 originally. If it'd been on a 40 MHz fsb, you could get a nice round 240. In fact if my other StarMax board can manage it, I might bump it up as well.

I am very curious about how much those two jumper sections need to align.

Well, I will say that some basic things are wonky at 275 :lol: I get consistent finder freezes if it boots into the Finder with a CD in the tray. MB 4 was failing to compete anything beyond CPU, FPU, and disk tests. I tried recopying the mb folder from the CD and it crashed both times. Went down to 250 and same story. I might borrow the flashed sata card and use another drive, but I doubt that would matter.

I haven't tried anything else yet, but it's probably safe to say I'll have to go back to 225.

 

[jessenator]

Well, as it turns out, the RAM DIMM I was using wasn't playing nice. I hadn't removed the known-working DIMMs from my 4400 to use in the StarMax, just a spare I had sitting around.

So my tests of each jumper setting were basically invalid :/  but the positive is that with good RAM, each clock speed is booting so far, save the 100 MHz (2x multiplier) which I won't test. CPU World lists 125 MHz as the minimum clock speed for this particular 603ev chip, so I figured, I'd just skip it. I'll test each and get a Main Tests score for each one in MB 4—that's been my "stability" test for now.

The fact that with the bad RAM I had so many false starts and just straight up weird activity gives me hope for the 300 limit :3 

 

[jessenator]

Well, despite things looking up, I can't hit the 300 MHz mark :/  not surprising though, as that's pushing the CPU quite a bit. It sorta wanted to for a few fleeting seconds. My screen would start to get a signal, but then go black. Oh well. At the very least, I've got stability at almost every other frequency on the chart.

Performance seems fairly linear, which is probably not a surprise. I did use a 512KB L2 cache for all the tests, which some of the StarMax models had as standard.

And just for fun, a PowerTower Pro and a 9500 thrown in for fun. Turns out those IX Micro Twin Turbos were packin' in the day  The StarMax board has ATi Rage II + DVD on-board graphics, vs the 4400/7220's 264VT2 if that has any bearing.

 

[Byrd]

Nice info!  Should be a good late 90's Mac.  The Twin Turbo PCI cards were 2D beasts, but useless for anything else notably games - the ATI Rage II is probably a better all rounder.

 

[jessenator]

Thanks. Hopefully this will all prove useful to someone else  also very glad I don't have to solder in another set of jumpers... although part of me is clutching at the invisible straws that it would help me actually get 300  

I've also got a Rage 128 sitting around doing nothing, since I got the DVD-equipped version for the PowerTower, so I could beef it up further. I did notice when comparing, very, very subjectively, that it ran several games much smoother than even my Gossamer G3 did. But that's another thing.

One day I hope to run into other versions of the Tanzania boards, maybe even one equipped with a 604. For now though, I think my next project is designing and printing a cowling for a fan/cooling solution. I do have a 40 mm fan from a 486 heatsink a friend gifted me, but I hate the sound it makes. The same friend also sent me this hilariously-packaged 80mm fan, which is actually quite silent and kept the heatsink cool during testing:



I might also use it for the chassis fan, because the original is pretty loud these days.

Edit:

Also of use for any future Tanzania/clone OC'ers, I managed to track down a (as-near-as-makes-no-difference) replacement heatsink for the super thin 603 piles:
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.

 

[Franklinstein]

Good find on those heat sinks. They could be good replacement candidates for the pathetic postage stamp-sized types on PM 55/6500 boards as well.

Some additions:

The Tanzania boards were used in mostly identical form officially by Apple, Motorola, and Motorola sublicensees, and unofficially by PowerCity after they bought Motorola's unsold inventory for quasi-legal sales in Europe after Apple ended official licensing. These boards had mostly the same features but you'd find occasional units that had onboard Ethernet, VGA instead of DB15, PS/2 ports, and either a 3- or 5-slot PCI riser (or two PCI plus CS II in the 4400). Though there was a provision for an MFM floppy drive, I've never seen the components populated.

These boards could be fitted with either the 603e/ev or 604e in a BGA package. I believe this is what the CPU ID section was for: setting the type and possibly voltage. The fastest 603ev available was 300MHz as sold in the StarMax 5000/300 while the fastest 604e model was a 200MHz StarMax 4000 or 5500 and derivative clones. System bus frequency was 40 or 50MHz, set by a couple of resistors near the PSX chip over by the VRAM connector.

While the 603 has a good bit of headroom for overclocking, I probably won't try to do the same for the 604 since it's already comparatively warmer than the 603. Regardless, a 10% overclock is about as far as I recommend going; anything more is likely to cause instability and shorten your equipment's useful life. For your 240, 275 is roughly a 15% bump, which is a good improvement and still fairly safe, but 300 would be about a 25% increase, which is pushing it a bit.

 

[jessenator]

300 would be about a 25% increase, which is pushing it a bit.

What did you think about my hypothesis that those two "CPU ..." sections need to match for optimal operation? You seem to have more experience with the Tanzania board machines than anyone else active on the forum. I know the board itself had provision for other features, like you were mentioning, as it was supposed to be the board for everything: Mac OS, OS/2, etc. didn't pan out that way. 

Just for kicks I wonder if it's at all possible to touch that plane. I know that's pushing, and don't think I'd ever daily it at 300. I really wish I could track down some files relative to it! schematics or any sort of notes... 

Sort of a Tanzania tangent, but incidentally I have one of the 5-slot PCI risers (without a case to put it in), and wondered, is there a PCI controller on the logic board, and then a second one on the 5-slot riser? the 3-slot riser and the 2PCI+CSII riser don't have a controller. Not that I'll find a case that would fit it. Most LPX and NLX cases have the PCI riser orientation mirrored from what the 4400/StarMax/et al do.

 

[Franklinstein]

I haven't done a ton of independent research on the developer's side, so I have no idea what the CPU ID setting options are or what they do (simple gestalt ID change? Voltage setting?). Maybe we could compare boards and see what's populated on each variant? I'm not enough of an engineer to know what I'm doing with an oscope, at least not without some info as to what I'm looking for; the best I can do is a poke-and-pray approach. These things don't have much enthusiast interest anymore (if they ever did) and a number of old websites that did have info have gone dark. I wish some of the engineers for these things would come drop some knowledge on us (among other models that have spotty support), or at least point to a book that would be helpful. 

I don't know how the LX440-based board used in the Tanzania/II models was licensed: were schematics and pinouts ever available or was it only BTO through Apple or a supplier? Did Motorola build most of them? PowerComputing never used them so no chance they could have leaked any relevant files. Gazelle boards were logically very similar but used buffered 5v RAM instead of unbuffered 3v.

The 5-slot riser uses a DEC 21052 PCI-PCI bridge to get you the extra two slots. The UMAX PowerSurge boards also use this chip to get to six slots, which is unlike the equivalent Power Mac 9500; the Tsunami boards instead use two Bandit chips for two completely independent PCI buses, which is a potentially faster but more expensive solution. The B&W G3 and early G4 Power Macs also use the later Intel version of a DEC PCI-PCI buffer (I think it was originally a 21152 but Intel called it a DC1111 or something) to bridge the native 66MHz PCI bus of the MPC106 and early UniNorth to the 33MHz PCI slots. Later versions of UniNorth integrated the PCI-PCI buffer into the main chip itself.