Testing 333MHz IBM 604ev 288in PGA module with G3 carrier for a PCI Mac

trag

Well-known member
I have seen a dual 604e card for the ANS but i could not look at the chips. That ANS-card had a separate 1M L2-cache for each CPU on the card and another 1M on the mainboard. There were two additional uncooled white IBM chips on the dual ANS-card. It looks a bit like the Apple Mach5 card but maybe they used the Helmwind (2nd gen 604e) since it is implicated that the chip can handle 100MHz bus-speeds (im struggling to find sources for that, though).

Where did you see the dual CPU ANS card? As far as I know there was only ever one of them, at least only one in the wild and word was it required a revised ANS ROM to work properly. I thought a fellow in Germany had it last, but "last" was back around 2002.
 

trag

Well-known member
I'm curious as to which Carrier card you were testing with. Somewhere around here I posted undocumented speed settings for one of the carrier cards (XLR8?). It's been too long and I don't remember which one. They give some speed settings in the instructions that came with the card, but it turns out the thing is adjustable in .2 MHz increments from something like 40 MHz up to about 70 MHz, IIRC. Anyway, that might get you over 58 MHz for what it's worth.

In my experiments, I could never get a bus speed higher than about 62 MHz to operate. There was an old 604e upgrade card from either XLR8 or PowerLogic which came with its adjustment switches on a little board at the end of a ribbon cable, so you could run them out of the case and adjust the speed without having to open and close the case constantly. PowerForce Pro or something like that. Definitely had the Pro at the end, because there was a non Pro version that was the same card, but without the ribbon cable -- switches were on the card inside the case on the non-pro version. Interesting thing about that card was that the speed adjustments ran through a little PIC on the card which took the settings and translated them into inputs for the clock generator and also for adjustments to the CLKID pins.

I'm going to bet that you were using an XLR8 brand card because they built their upgrades to take advantage of the CLKID pins in the CPU slot. These three pins signal the logic board chips as to what bus speed will be used so that memory timings and such can be adjusted properly. Without adjusting the CLKID pins, PCI Power Macs generally won't run bus speeds above 50 MHz and it's tough to get much over 45 MHz.
 

stynx

Well-known member
Where did you see the dual CPU ANS card? As far as I know there was only ever one of them, at least only one in the wild and word was it required a revised ANS ROM to work properly. I thought a fellow in Germany had it last, but "last" was back around 2002.
If i could find it, i would have posted the pic. Im from Germany as well, so it could have been a german retro forum, in the wild or something? I was into the ANS (having bought one in 2001) for a while and there was a thriving community until about 2006. I had my NetBSD ANS500 with 5 UWSCSI trays using SCSI2IDE converters and 120Gb per drive in raid5 (450Gb usable), CD, 8Gb streamer and 1Gb 50pin boot drive (completely decked out). Very power-hungry and loud despite the fan-mod but very reliable.

I'm curious as to which Carrier card you were testing with. Somewhere around here I posted undocumented speed settings for one of the carrier cards (XLR8?). It's been too long and I don't remember which one. They give some speed settings in the instructions that came with the card, but it turns out the thing is adjustable in .2 MHz increments from something like 40 MHz up to about 70 MHz, IIRC. Anyway, that might get you over 58 MHz for what it's worth.

In my experiments, I could never get a bus speed higher than about 62 MHz to operate. There was an old 604e upgrade card from either XLR8 or PowerLogic which came with its adjustment switches on a little board at the end of a ribbon cable, so you could run them out of the case and adjust the speed without having to open and close the case constantly. PowerForce Pro or something like that. Definitely had the Pro at the end, because there was a non Pro version that was the same card, but without the ribbon cable -- switches were on the card inside the case on the non-pro version. Interesting thing about that card was that the speed adjustments ran through a little PIC on the card which took the settings and translated them into inputs for the clock generator and also for adjustments to the CLKID pins.

I'm going to bet that you were using an XLR8 brand card because they built their upgrades to take advantage of the CLKID pins in the CPU slot. These three pins signal the logic board chips as to what bus speed will be used so that memory timings and such can be adjusted properly. Without adjusting the CLKID pins, PCI Power Macs generally won't run bus speeds above 50 MHz and it's tough to get much over 45 MHz.
I have some Formac G3 Carrier cards. I had a XLR8 card in the past but i must have given it away accidentally with one of the PM7x00 when i reorganized a while back. I can typically get to a stable 57Mhz bus clock with a 7x00 or 8x00 board but all of my 9600 will usually go a bit higher although i never really archived a reliable 60MHz.
 

Powerbase

Well-known member
If i could find it, i would have posted the pic. Im from Germany as well, so it could have been a german retro forum, in the wild or something? I was into the ANS (having bought one in 2001) for a while and there was a thriving community until about 2006. I had my NetBSD ANS500 with 5 UWSCSI trays using SCSI2IDE converters and 120Gb per drive in raid5 (450Gb usable), CD, 8Gb streamer and 1Gb 50pin boot drive (completely decked out). Very power-hungry and loud despite the fan-mod but very reliable.


I have some Formac G3 Carrier cards. I had a XLR8 card in the past but i must have given it away accidentally with one of the PM7x00 when i reorganized a while back. I can typically get to a stable 57Mhz bus clock with a 7x00 or 8x00 board but all of my 9600 will usually go a bit higher although i never really archived a reliable 60MHz.
I'm surprised Apple didn't really do any dual or quad chips for an actual server platform, that might have been able to take advantage of it.

I know Applefritter used to be hosted off a weird prototype ANS.
 

stynx

Well-known member
I'm surprised Apple didn't really do any dual or quad chips for an actual server platform, that might have been able to take advantage of it.

I know Applefritter used to be hosted off a weird prototype ANS.
Apple seems to have experimented with dual and multi-cpu systems since the 1994 internally. None of the prototypes ever resulted in an actual product on the market. The MacOS was not really MP capable (just soft-controlled multithreading) even thought the microkernel of (at least) MacOS 8.6 was seemingly capable of some kind of SMP. The Daystar/Apple MP-card was a MacOS compatible solution where the 2nd. CPU had to be 'started' by a system library. There are some interesting developer tools the look at the microkernel-threads of which the MacOS is one...
The whole cooperative-multitasking problem of MacOS was the main culprit of the inability to really get any decent performance out of MP-systems until MacOS X came along. It's like throughout the 90s, Apple consisted of multiple companies running in parallel with little oversight and no real vision.
 

Powerbase

Well-known member
Apple seems to have experimented with dual and multi-cpu systems since the 1994 internally. None of the prototypes ever resulted in an actual product on the market. The MacOS was not really MP capable (just soft-controlled multithreading) even thought the microkernel of (at least) MacOS 8.6 was seemingly capable of some kind of SMP. The Daystar/Apple MP-card was a MacOS compatible solution where the 2nd. CPU had to be 'started' by a system library. There are some interesting developer tools the look at the microkernel-threads of which the MacOS is one...
The whole cooperative-multitasking problem of MacOS was the main culprit of the inability to really get any decent performance out of MP-systems until MacOS X came along. It's like throughout the 90s, Apple consisted of multiple companies running in parallel with little oversight and no real vision.
I swear I was watching a video about some of the CPU development Apple was working on in the 90s. It was the reason they bought that Cray too.

Here it be:
 

stynx

Well-known member
I swear I was watching a video about some of the CPU development Apple was working on in the 90s. It was the reason they bought that Cray too.

Here it be:
RetroBytes does a lot of good deep diving. There are some really good YouTubers who cover a lot of obscure tech history.

Apple was a lot more innovative than most people might think based on the bad performance the had in the 90s. They had so many talented people doing groundwork that it is almost impressive that so little came from it. I get the notion that Apple management was seemingly actively sabotaging the motivated engineers and forced last minute changes or cancelled projects outright shortly before they were finished (or even after they were finished). No wonder that some of the most brilliant developers left Apple to have their own companies.

Seeing those documentaries about Apples history and the internal development is like watching someone shooting himself in the foot, repeatedly.

The Quickdraw 3D Accelerator for example is most likely a similar story. The engineers behind it were able to develop a limited but functional device, ready to be mass produced. The hints of what was actually planned and prototyped internally was worlds beyond the released device. The 1994 internal prototype was 8x faster to draw polygons than the release version for example and the 1997 prototype was more than 10x faster overall with even more features and it even did FSAA with only 30% reduction in speed. Now just think about it: The QD3D Accelerator was more capable than the Playstation in terms of 3D and basically ready in late 1994, the same time when the Pippin was finalized for production... Why did no-one think about adding that chip to the Pippin? So much unused potential and barriers within the company.
 

Arbee

Well-known member
Re: the rant, Apple left a lot of potential on the shelf. I just emulated the IIfx's unique SCSI chip and then found out no version of Mac OS (or at least one that runs on an '030) uses the fast DMA transfer mode.
 

Powerbase

Well-known member
Re: the rant, Apple left a lot of potential on the shelf. I just emulated the IIfx's unique SCSI chip and then found out no version of Mac OS (or at least one that runs on an '030) uses the fast DMA transfer mode.
I think the IIfx had a lot of performance left on the table, it used a bunch of ASICs Mac OS didn't apparently do anything with (I've always heard).
 

Arbee

Well-known member
It wasn't actually *too* wasted outside of the SCSI. The IOP accelerated ADB and floppy are used, and the IOP even does DMA to the floppy. The IOPs appeared again in the Quadra 900 and 950 and do the same jobs there.
 

stynx

Well-known member
I found a supplier that offers XPC604RRX300LD variants in greater volume.

XIA SONG ELECTRONICS CO.,LIMITED
XIASONGDZ@163.COM

The price is at $25 per unit after some haggling.
The transportation fee and additional bank fee is quite high (+$60 and +$50) with $110 on top of the order amount.

I have some problems with payment since they prefer international bank transfer and my bank refuses to transfer to their bank.

QQ图片20240801173941.jpgQQ图片20240801173948.jpgQQ图片20240801150439.jpg
 

stynx

Well-known member
I'm running a test-buy from another source (tech-electr.com) for PPC604E3DB-C400E at $50 per unit.
Shipping fee is $35 to germany.
They have suddenly risen the price from $25 to $50 so they might not be legitimate.

They offer staggered pricing with greater quantities.
Bildschirmfoto 2024-08-01 um 20.37.32.png
As soon as i have an update, i will post it here.


-Jonas
 

stynx

Well-known member
IBM RS6000 Dual 333MHz 604e board. Defective.
Interestingly there are 350MHz capable CPUs installed.
IMG_1516.jpg
The big chip is the 256k cache-controller module.

I will use this as a toner, if i cannot get a good supply of 350/400MHz capable CPUs.
 

stynx

Well-known member
One supplier just offered me $6.26 for one XPC604RRX300LD with minimum 10pcs + shipping $35 and tax + $18 (Germany).
If this is legitimate, it would be the lowest price i have seen so far. It is only a 300MHz part but they would typically overclock to 350MHz.
If the order goes through, i will share the contact info of the supplier.
 

stynx

Well-known member
One supplier just offered me $6.26 for one XPC604RRX300LD with minimum 10pcs + shipping $35 and tax + $18 (Germany).
Update: the price suddenly has risen to about $23 per chips ... i think i see a pattern here ;-)
 

herd

Well-known member
Let me guess, from china?

If you do pull chips off of those ZIF modules, I'd be interested in the left over pin arrays. I'd like to try installing 750G chips on them (btw, these G3 CPUs were less than $4 each delivered).
 

stynx

Well-known member
Let me guess, from china?

If you do pull chips off of those ZIF modules, I'd be interested in the left over pin arrays. I'd like to try installing 750G chips on them (btw, these G3 CPUs were less than $4 each delivered).
I may have found a good source for $3.53/pcs of xpc604RRX300 I will see if/when I get my batch.
The ppc750 will not fit on the BGA255 pads. The ppc740 (603 replacement) might fit but will most likely not work because some pins are different. The difference might not be important, though.
I will not desolder the 604 from the ZIF module.
 

stynx

Well-known member
The difference between 604 and 603
(personal observation, may not be complete)

Signals on the 604 that are not present on the 603:
DrvMod0, DrvMod1
<- drive mode (must be pulled high?)
Halted, L2int, RUN, SHD <- may be related to MP functions
TC2 <- transfer code 2 (603 has only TC0 and TC1)
XATS <- extended adress transfer

Signals on the 603 that are not present on the 604:
QACK, QREQ
<- related to low power mode

All the different signals are on NC pins on the other CPU.
It may be possible that the chips are interchangeable but won't offer some features if the pins are not connected correctly?
 

LightBulbFun

Well-known member
Let me guess, from china?

If you do pull chips off of those ZIF modules, I'd be interested in the left over pin arrays. I'd like to try installing 750G chips on them (btw, these G3 CPUs were less than $4 each delivered).
Ooh a BGA255 interposer for the 750GX? that could be quite fun in its own right, for upgrading Old world PCI Macs :)

I may have found a good source for $3.53/pcs of xpc604RRX300 I will see if/when I get my batch.
The ppc750 will not fit on the BGA255 pads. The ppc740 (603 replacement) might fit but will most likely not work because some pins are different. The difference might not be important, though.
I will not desolder the 604 from the ZIF module.

the PPC740L will work on a PowerMac 6500's motherboard, once ya adjust the vCore suitably


I have always found it curious how OF on the 6500 *does* recognise the 740, but calls it a 608! (if it did not recognise it, it would get called PowerPC 60?)

im pretty sure the 740 is in general designed to be a drop in replacement for the 603/604 CBGA255 CPU's
 
Top