Power Mac 7500-9600 Bus Speed Overclocking

[cobalt60]

I have a total of 4 boards that I want to see how far they can go with bus speed. So far, I have only tested one (a 7300/200). I will test the 8600/250 next, and once I do something about a PSU, I can test the 7600 and 9500.

Currently still working with the 7300. All memory slots are filled, and memory is interleaved. All RAM is 60ns EDO RAM. CPU is a ZIF carrier with under-clocked G4 and oversized heatsink. Here are my observations so far:

60 MHz - No boot
58 MHz - Boot not guaranteed
57 MHz - Always boots, but usually get data read errors using Stuffit Expander
55.6 MHz - Seemed fine until I went to install Mac OS 9 (from one ZuluSCSI partition to another). I got an error 'problems were encountered reading the source file "Installation Tome". Installation cannot continue'
54 MHz - Same as above
53 MHz - Made it further into the install, but got a different data read error
52 MHz - Works great

So here are my questions:
1) Are these valid correlations I have made?
2) If so, is there a memory tester for OS 9 that works as well as running the OS installer?
3) Should I worry about individual 60ns EDO modules not being able to handle up to 60MHz bus speeds?
4) Should I worry about PCI cards or SCSI devices restricting bus speed?

I might try un-interleaving the RAM, but I really don't want to go as far as testing individual memory modules unless I am convinced that they are not keeping up with whatever they are spec'd to.

 

[Phipli]

If so, is there a memory tester for OS 9 that works as well as running the OS installer?

I don't expect so. Installing is just a complex and heavy load. Lots of disk activity, decompression etc.

Should I worry about individual 60ns EDO modules not being able to handle up to 60MHz bus speeds?

Possibly yes. Apparently EDO overclocks better than FPM, but faster RAM will help. For your experiments it's good to test with a single simm, the best one you can find. If you find one that works well, use that for tests and only add others (and retest) once you have the machine stable.

Should I worry about PCI cards or SCSI devices restricting bus speed?

These... Shouldn't be affected? The PCI bus is separately clocked isn't it? Video and Video RAM are bigger concerns.

Some macs in that era even struggles to achieve 50MHz. Can I ask why you're looking to overclock the bus, and not the CPU multiplier (I realise there are benefits, just interested what your motivation is).

Once you find a stable speed, you'll want to back off a few percent to give yourself overhead in hot weather if it isn't baking hot now, and ensure reliability under even extreme workloads.

 

[cobalt60]

Can I ask why you're looking to overclock the bus, and not the CPU multiplier

I bought this computer and all these boards for the main purpose of bus speed overclocking testing.

Also, the best CPUs I have at this very moment (that fit this machine) require bus speed overclocking to hit their rated speed. I have a 450MHz 750L with an 8x multiplier, would need over 56MHz bus to hit that. I have a 550MHz 7410 with a 9x multiplier, I was pretty happy with 55.6MHz running it at 500MHz. I am getting a 500MHz 750L in the mail soon which I think has a 10x multiplier. 50MHz would get me to it's rated speed, but I'd have to overclock the bus there in order to overclock the CPU.

The PCI bus is separately clocked isn't it? Video and Video RAM are bigger concerns.

Not sure but I think it is separate. I am using a Rage 128 instead of the on-board video, with the motherboard VRAM removed. The only other variable than the OS install was that I was using the ZuluSCSI for around the first time, and also had a USB card recently installed, so just wanted to verify that didn't have anything to do with it. 55.6 to 52 MHz was quite the drop.

Apparently EDO overclocks better than FPM, but faster RAM will help

I don't know a ton about how RAM interacts with the mainboard, but PowerComputing used 60ns FPM RAM in their 60MHz bus models. I assumed since I am running below 60MHz, and using 60ns EDO RAM, that RAM shouldn't be an issue for me unless a module is "worn out" or faulty. Is that assumption correct, and you're suggesting my RAM may not be functioning as intended? Or is the issue more that I don't know something deeper or more nuanced?

 

[Phipli]

I don't know a ton about how RAM interacts with the mainboard, but PowerComputing used 60ns FPM RAM in their 60MHz bus models. I assumed since I am running below 60MHz, and using 60ns EDO RAM, that RAM shouldn't be an issue for me unless a module is "worn out" or faulty. Is that assumption correct, and you're suggesting my RAM may not be functioning as intended? Or is the issue more that I don't know something deeper or more nuanced?

Sorry, I mean "fast RAM is good". I didn't mean to imply that yours wasn't fast enough. Off the top of my head I can't remember what speed RAM is needed by a 50MHz PCI Mac, but 50ns RAM isn't very common (I'm lucky enough to have some 72pin 50ns sticks that I use for Quadra/Centris overclock experiments (with onboard RAM desoldered)).

"Worn out" shouldn't be a thing. It should either be good or broken RAM.

require bus speed overclocking to hit their rated speed

That's a fair reason.

55.6 to 52 MHz was quite the drop.

Old upgrade card manuals seem to imply not all of these machines can do 50MHz even, so I don't find the results surprising. An OS installer can be quite a load on the machine.

To get the most out of your machine,
I'd suggest switching to using a SATA PCI card and a modern SSD, and dropping the bus to 50MHz for stability. The SATA will give a substantial speed boost, especially with booting and loading applications. If you're able to flash one, a cheap black card off eBay should be sufficient. They play up in G4S, but generally behave in beige machines. We can walk you through any changes needed - worst case is replacing the ROM chip and a voltage regulator. Best case is running a program on the desktop to flash it. They cost about $15 I believe.

 

[Coloruser]

Right, PCI Bus (Bandit IC) and processor bus run independently and asynchronous. However, for higher bus speeds, you sometimes have to adjust bus delay. Not all CPU upgrades support that feature. In addition, lots of ram puts lots of stress or load on the memory bus. It may not be ram that fails but simply the memory controller (Hammerhead) used on (Tsunami/Nitro/TNT). That IC wasn‘t designed for 50+ mhz operation. It has some margin and with bus delay can reach 58 mhz stable. The platinum (non interleaved) controller of the catalyst series (Apple 7200 and Power Computing) was able to go higher - 60 mhz with power computing catalyst based units. Even power computing never made a hammerhead version faster than 50 mhz

 

[cobalt60]

I'd suggest... dropping the bus to 50MHz for stability.

I mean really the point of this computer is for testing bus speed, as well as testing peripherals and components for a future 9600 build. After that, I might sell it. I already have another 7300, 2x B&W G3s, a 1.42 MDD, a Mini G4, a 7200, and a 9600.

I'd suggest switching to using a SATA PCI card and a modern SSD

I've got that Rabbit Hole Computing SATA Hard Card, which is another peripheral I plan to test in this machine before it goes into the 9600

We can walk you through any changes needed - ... Best case is running a program on the desktop to flash it.

I was planning on starting another thread for this if I couldn't figure it out, but yes I have no idea how to flash it at the moment. I don't run Windows; only Linux, Mac OS 9, and soon OS X. I would be willing to run FreeDOS, but I am not sure if I have a PC with a PCI slot. Flashing from Linux would be ideal.

lots of ram puts lots of stress or load on the memory bus. It may not be ram that fails but simply the memory controller (Hammerhead) used on (Tsunami/Nitro/TNT).

That's what I figured and just wanted to make sure with the questions I asked, and is really the point of this testing.

with bus delay can reach 58 mhz stable

Never heard of this, any more info on this? How to set it for example? I will look into this more.

 

[Phipli]

I mean really the point of this computer is for testing bus speed, as well as testing peripherals and components for a future 9600 build. After that, I might sell it. I already have another 7300, 2x B&W G3s, a 1.42 MDD, a Mini G4, a 7200, and a 9600.


I've got that Rabbit Hole Computing SATA Hard Card, which is another peripheral I plan to test in this machine before it goes into the 9600


I was planning on starting another thread for this if I couldn't figure it out, but yes I have no idea how to flash it at the moment. I don't run Windows; only Linux, Mac OS 9, and soon OS X. I would be willing to run FreeDOS, but I am not sure if I have a PC with a PCI slot. Flashing from Linux would be ideal.


That's what I figured and just wanted to make sure with the questions I asked, and is really the point of this testing.


Never heard of this, any more info on this? How to set it for example? I will look into this more.

Just a heads-up - you're quoting in a slightly unusual way (are you clicking the quote button an pasting in the text?), And so it doesn't alert the people you're responding to.
If you instead select the text you're replying to, a little floating "Reply" button appears. Click that and it adds a quote of the selected text to your current draft message. You can do it as many times as you like.
Doing it that way (or just clicking the main reply button at the bottom of the post you're replying to) alerts the other user that you've said something. Another option is using "@ username" (with no space)

 

[cobalt60]

If you instead select the text you're replying to, a little floating "Reply" button appears.

Didn't know that, I tried the "+Quote" before and didn't like how it worked, so I've just been copy/pasting. Will use the method you suggested now.

 

[trag]

I've had 9500 systems adn clones up to 62 MHz using a PowerBoost Pro. I think that was a PowerLogix upgrade, but maybe XLR8? It was a PPC604e upgrade though, not a G3 or G4.

There are three pins in the CPU socket called ChipID which signal Hammerhead how to adjust the bus and memory timings. They are a method for the CPU card to inform the motherboard what bus speed will be used.

Early CPU upgrades didn't know about these pins and so none (few) of them could get a stable bus speed above 40 or 45 MHz. This was true even of early G3 upgrades.

The PowerBoost Pro has a PIC on board with links to the speed setting dial and to the CLKID pins and adjusts the reported timing as one changes the bus speed.

I don't know if the Carrier cards do that or not.

Also, the 7500 and 8500 family machines tend to have lower top bus speeds because of the video circuitry. While Bandit doesn't slow down the CPU bus speed, the CHAOS/CONTROL chip set seems to do so.

Apparently Hammerhead has internal addressable registers which can be used to adjust the memory timings. NewerTech had a tool they used for this internally. It would be nice if there was a way to get our hands on that tool. I exchanged some emails back in the oughts with an employee who mentioned some of the things he could do with it.

 

[macuserman]

I mean really the point of this computer is for testing bus speed, as well as testing peripherals and components for a future 9600 build. After that, I might sell it. I already have another 7300, 2x B&W G3s, a 1.42 MDD, a Mini G4, a 7200, and a 9600.


I've got that Rabbit Hole Computing SATA Hard Card, which is another peripheral I plan to test in this machine before it goes into the 9600


I was planning on starting another thread for this if I couldn't figure it out, but yes I have no idea how to flash it at the moment. I don't run Windows; only Linux, Mac OS 9, and soon OS X. I would be willing to run FreeDOS, but I am not sure if I have a PC with a PCI slot. Flashing from Linux would be ideal.


That's what I figured and just wanted to make sure with the questions I asked, and is really the point of this testing.


Never heard of this, any more info on this? How to set it for example? I will look into this more.

Just use dosdudes utility in OS 9 and your done. This works for your Rabbit Hole computing card and is very simple. http://dosdude1.com/files/macstuff/PCISataROMs/flashers/SeriTek1S2Flasher_5.1.3.sit.hqx

 

[Coloruser]

I mean really the point of this computer is for testing bus speed, as well as testing peripherals and components for a future 9600 build. After that, I might sell it. I already have another 7300, 2x B&W G3s, a 1.42 MDD, a Mini G4, a 7200, and a 9600.


I've got that Rabbit Hole Computing SATA Hard Card, which is another peripheral I plan to test in this machine before it goes into the 9600


I was planning on starting another thread for this if I couldn't figure it out, but yes I have no idea how to flash it at the moment. I don't run Windows; only Linux, Mac OS 9, and soon OS X. I would be willing to run FreeDOS, but I am not sure if I have a PC with a PCI slot. Flashing from Linux would be ideal.


That's what I figured and just wanted to make sure with the questions I asked, and is really the point of this testing.


Never heard of this, any more info on this? How to set it for example? I will look into this more.

As @trag mentioned, Hammerhead could be adjusted for bus timings and memory timings. The CPU card (JoeCARD) that I co-developed used a microcontroller to adjust multiplier, bus speed and delay. The latter means it adjusted Hammerhead‘s registers to match the requested bus speed. Other manufacturers also made use of this opportunity. Some directly via dials on the card (like JoeCard and PowerBoost) others only for internal testing and some even in software. But i don’t know whether any of the ZIF carrier cards have this feature. If they just set the bus speed without Hammerhead adjustments, chances are, the bus is not ging to be stable at higher speeds and bus loads.

 

[Phipli]

As @trag mentioned, Hammerhead could be adjusted for bus timings and memory timings. The CPU card (JoeCARD) that I co-developed used a microcontroller to adjust multiplier, bus speed and delay. The latter means it adjusted Hammerhead‘s registers to match the requested bus speed. Other manufacturers also made use of this opportunity. Some directly via dials on the card (like JoeCard and PowerBoost) others only for internal testing and some even in software. But i don’t know whether any of the ZIF carrier cards have this feature. If they just set the bus speed without Hammerhead adjustments, chances are, the bus is not ging to be stable at higher speeds and bus loads.

Do you happen to know if any of the software solutions for adjusting Hammerhead's registers were released publicly? It is there any documentation on what is needed to allow modifications to cards?

The most common card out there is the XLR8 Mach ZIF Carrier. I'll see what documentation I can find.

 

[Bolle]

I don't know if the Carrier cards do that or not.

I had a Carrier card running 62MHz bus speed in a 9500 board, so very likely they do so as well.

For some reason the high bus clock would only work with one specific ZIF module and would fail with all others I have tried.
If I remember right the ZIF that would work had a copper CPU from IBM, the others all were Motorola G3s.

 

[Phipli]

If I remember right the ZIF that would work had a copper CPU from IBM, the others all were Motorola G3s.

Was the Copper IBM ZIF from a B&W? They more often are. That would mean spec'd for a 100MHz bus and so perhaps tighter timings, better slew rates, lower proportion delays?

 

[trag]

Do you happen to know if any of the software solutions for adjusting Hammerhead's registers were released publicly? It is there any documentation on what is needed to allow modifications to cards?

The most common card out there is the XLR8 Mach ZIF Carrier. I'll see what documentation I can find.

I've never found any public documentation of it and the NewerTech employee who told me about asked me to keep it secret -- NDAs or some such. I think it's been long enough... I wish I had asked him for those internal apps. He probably would have had to say no, but I never asked.

"modifications to cards" Some CPU cards just have the CLKID pins tied to resistor pads which are populated or not. Pretty sure the other end of the pads go to GND, but I haven't applied a DMM to this question in 20 years....

AFAICT, the CLKID pins are tied to GND or not. Probably Vss vs. Vcc. So one of my ambitions is to probe the pins on the PowerBoost Pro at different bus speed settings. Should also be possible to do this on the ZIF carrier.

Three pins gives 8 possible settings, and for some reason I've forgotten, I believe they go in 5MHz increments. That could be completely wrong.

CPU slot pinout is here: http://sphinxgroup.org/Apple_pinouts/

CLKID pins are pins 15 - 17

'P' and 'S' are primary and Secondary. So PINT is Primary Interrupt. SBG is Secondary Bus Grant. Apple had two CPUs in mind from the beginning of the slot design.

IIRC correctly, DRTRY is needed by the Catalyst chips set (7200 and clones) but not by the PowerSurge (X500/X600) chipset. Apple CPU cards don't connect it. This is why most non-PCC cards don't work in the PowerCurve, PowerTower, PowerCenter and PowerCenter Pro, all of which are Catalyst based clones.

Anyway, if someone else wants to lead the way on probing out CLKID's operation, that would be great.

I wonder if any of the Unix-flavor ports to X500/X600 worked out the register addresses/uses of Hammerhead....

 

[Coloruser]

Do you happen to know if any of the software solutions for adjusting Hammerhead's registers were released publicly? It is there any documentation on what is needed to allow modifications to cards?

The most common card out there is the XLR8 Mach ZIF Carrier. I'll see what documentation I can find.

There was no public software I know of. I will check whether I still find the documentation for the CPLD that we used to program clock, bus and delay. But it may take some time. It‘s been 25 years since I worked on it.

 

[Phipli]

There was no public software I know of. I will check whether I still find the documentation for the CPLD that we used to program clock, bus and delay. But it may take some time. It‘s been 25 years since I worked on it.

25 years? Just a blink of an eye

Thank you.