What do you guys know about IBM 604e ZIF modules?
About two weeks ago, I found one on eBay and snagged it while I could. It's a 333MHz part. My guess is that it's out of an IBM RS/6000 43p Model 140, but I know next to nothing about IBM's hardware. Anyway, the module is a 289-pin ZIF module (17x17 - same size and pin count as a G3 ZIF). I did a little Googling, and the only mention of 604e ZIF modules I could find was in reference to IBM's Long Trail CHRP concept motherboard that was supposed to be able to run AIX, Mac OS 7.5 (CHRP edition), and Windows NT (check out the attached PDFs: NOT FOUND: 1, NOT FOUND: 2 - I grabbed these from ftp://ftp.boulder.ibm.com/rs6000/PPC_support/reference_designs/longtrail/pass2/).
So, being somewhat curious, I plugged the module into one of my Rev A beige G3s this evening, and...
I'll be damned! It booted right up!
NOT FOUND: 3
I snagged some screenshots to prove that it boots, and both Apple System Profiler and Gauge Pro recognize it as a 604e @ 333MHz (I did have to set the jumper to that configuration, of course). Gauge Pro says the processor version is 1.2, whatever that means.
Sadly, the module doesn't have any L2 cache. So booting a beige G3 with a 604e is really just an academic pursuit. It's not going to win any speed contests against a similarly clocked G3 that does have a proper L2 cache. Though if one could get an L2 cache working with it, it would be interesting to compare apples-to-apples, so to speak, to see what kind of performance the 604e can put up against a G3 in otherwise identical configurations. I'm guessing the 604e would hold it's own...
Anyway, the beige G3 motherboard has all some legacy CHRP stuff like solder pads for PC floppy drives. There's also a set of solder pads for an L2 cache DIMM. The Long Trail design was supposed to use ZIF-socketed 603e & 604e processors with a COASt (Cache on a Stick, whatever that is) L2 module. The solder pads of this missing L2 cache slot and the installed slots on PCI Macs are indistinguishable in comparison by me. I've speculated before about whether or not the slot would be active if someone actually installed a cache module there. Now I'm even more curious. And what is this COASt L2 module thingy? Is that what the PCI Macs use?
I guess the point of this post (besides rambling) is to share what I learned about the Long Trail design, to speculate on how Long Trail might have influenced the Gossamer motherboard design, and to ask what else you smart people know about IBM ZIF modules, where I could get them, and if any 604e ZIF modules have on-board L2 cache...
(Note to mods: I tried to attach files and use the "inline" tags, but the preview looks messy. Can you fix it for me if I did it wrong?)
Oh, I forgot to mention: I took some crappy pics with my phone and put them in my MobileMe gallery. Eventually, I'll take better pics and post it on www.ppcmla.com, but for now you can find them here.
Picture 1 showing a comparison between the 604e and the G3 ZIF modules.
Picture 2 showing the 604e installed in the ZIF socket.
Picture 3 showing the 604e fully installed under the heat sink.
The 604e module has an extra metal conductor (probably a heat conductor to make contact with a heat sink in the RS/6000) that attaches to the core. It's black and has the IBM logo, PowerPC, and the part number silkscreened on it.
I only ran it for a few minutes. I was worried about heat. The 604e is generally a hot chip for a PowerPC, so I was concerned that the G3's heat sink would be insufficient. I didn't run anything significant - just ASP, Gauge Pro, and Metronome (which crashed on launch). After less than 5 minutes with a basically idle CPU, the heat sink was vaguely warm to the touch, but not hot. So maybe it's enough... I don't know. If I do more experimenting, I'll stick a thermister in there and get some good temperature readings.
Looking at the Long Trail Reference Design in the pdf, it's pretty obvious how they came up with the name Long Trail. I studied it hoping to find a VCI slot like the one on the PEX motherboard:
Nope, not there in the reference design that I can see. The VCI slot would evidently have offered optional GPU daughtercard upgrades for the onboard video. This was the same age of design as the PEX and the PEX would supposedly have come in the option of either a 604e or a G3 processor daughter card.
Of course, many of the Powersurge PCI Macs had L2 cache cards in a separate slot--some offering upgrades to 512k and 1mb (the PM8500, 64xx,65xx, come to mind). The first generation PM9600 came with the L2 soldered to the motherboard and the 604e processor sitting alone on the daughtercard. The second generation put the L2 on the daughtercard with the processor. All these things happening at Apple around the same time as your cacheless ZIF. What to do with the L2? Why not a slot card as the standard? Would there be any problems with upgrade options for the L2? I guess Apple just got tired of the extra complexity. Steve Jobs probably came in, took a look and said, "forget all this complexity, we've got to sell some machines," and thus, the Gossamer was born. Maybe they should have called the Gossamer, the ShorterTrail, instead. The PEX might have been called, the EvenLongerTrail, or the TooLongTrail.
That's pretty darned cool, Alk. I don't think I've read of anyone else trying this.
As you say, not too useful unless you get a cache dropped in. Hopefully the same 604e cache from the speed-bumped 8600/9600 would work, if you can snag a slot from a scrap PCI Mac.
A feather in your cap, sir.
Wow, I didn't know such a thing existed. I'm suprised that it is simply plug-n-play too.
It's not surprising there was no mention of a VCI slot in the Long Trail spec for two reasons. 1) LT is a cheap, low cost system designed for cloners and to entice PC makers to build PowerPC machines. VCI doesn't fit in that market. 2) PowerExpress is entirely an Apple design based on the PowerSurge/900 motherboard with no CHRP components on board at all. It was entirely Old World. Remember, Gossamer replaced PowerExpress because the cheaper PC-standard components out-performed the custom designed, expensive lagacy components...
Hmm, that's interesting. I'm trying to follow what you're saying. So who was trying to entice PC makers to build PowerPC machines? IBM? Do you mean entice them to build PowerPC Mac clones to run Mac OS? Or entice them to use the PowerPC processor in Windows machines? Would that have been possible? What was going on between Apple and IBM at that time? Why was Apple tolerating the Mac clones? Was IBM nefariously undercutting Apple by supplying the clones with processors, thereby strong-arming Apple to play along with OS licenses? But as we like to say, Apple is not a software company, it's a hardware company, so licensing is BS for Apple. As I understand it, Mac OS 8 was created with the express purpose of wiping out the Mac clone market, which it did successfully. I've not read yet how exactly it was able to do that, but it was evidently one of Steve's great victories, correct?
From what I've read of the history of the PEX, replace might not be the correct term for the Gossamer over the PEX. Both were contenders for production, and Jobs nixed PEX at the last moment because it didn't fit into his vision of the future of Mac, but, at that time, the Gossamer was not satisfactory for many, and the 9600 was brought back into production for a short while because in some ways it was more capable than the earliest Gossamers. It's probably an overly fine semantic line I'm drawing, but the PEX was more abandoned than replaced. It was just too much, but with a G3 processor and a fast SCSI onboard setup, it would have outperformed the Gossamers easily, would it not? Was not SJ abandoning the better performing, but more expensive SCSI, for the less performing, but cheaper IDE, and one of the main reasons was also that IDE was more user friendly as well? Take away the fast SCSI, the VCI, and three of the six PCI slots, and then with the components that are left, the Gossamer is the far better machine with its superior I/O chips. I can just imagine a PEX with those I/O chips and it's six PCI slots and upgradeable video and fast SCSI--wow, that would have been an unreplaceable monster--killable, but not replaceable. Steve abandoned the monster for the slim, trim, mass-marketable Gossamer, probably with the knowledge that onboard USB and Firewire were just a year away. It just seems like with that decision, some good babies might have been thrown out with the bath water, but it was time to make that decision for the financial feasability of Apple's future. There was a certain life and death struggle with a lot of throat slitting and discarding of the corpses just to get underway, and maybe there was some unfortunate "collateral damage." This is the sense of the times that I've picked up in my very limited knowledge of it. I won't be surprised if you can correct me on this.
Hmm. Is it possible to hijack one's own thread?
Clone makers were eating Apple's lunch. Apple licensed Mac OS (and Motorola and IBM licensed PowerPC chips and reference boards) to expand the installed base of Mac OS compatible PowerPC computers by eating away at the Windows/Intel market share. They hoped clone makers would market their boxes to PC users and be able to capture new business. Instead, the clone makers competed directly with Apple for the existing Mac users without capturing much if any new business. Worse, Power Computing was completely annihilating Apple with machines that out-performed Apple's at lower prices. It was a business imperative for Apple to kill off the clones. Clone makers had licenses to versions of System 7, but not OS 8. Apple refused to certify clones as compatible with OS 8 as an artificial way to pull their licenses. Without a license for OS 8, they wouldn't be able to compete against Apple's own hardware when the new OS shipped. Further, Apple refused to certify the G3 as compatible, so cloners couldn't sell G3s installed in their boxes. Some of the cloners got around this by shipping retail copies of the OS 8 CD in the box with System 7 installed. Others shipped 3rd party G3 upgrades in the box with a very low-end processor daughtercard installed to save costs. This was no way to stay in business, and the clone makers shut down one by one (or their license was bought back by Apple as in the case of Power Computing). Apple's was a legal maneuver, not a technical one. All the clone hardware worked fine with Mac OS 8 (and even OS 9 and OS X as many have demonstrated).
In the meantime, there was a partnership with IBM, Motorola, and Apple to build motherboards using industry standard parts to lower the cost of building a generic PC using a PowerPC processor (Common Hardware Reference Platform (CHRP) which was derived from the PowerPC Refence Platform (PReP) and which later became the PowerPC Platform (PPCP)). This resulted in a few motherboard lines. The Tanzania board (used in the 4400, Starmax clones, and Apple's LPX-40 logic board reference design), the Long Trail board, and ultimately real CHRP models like the Starmax 6000 which never shipped. Motorola and IBM supplied the processors, 3rd parties supplied the various ICs and controllers, and as a carrot, Apple supplied the Mac OS ROM to license holders.
All this has little to do with PowerExpress or Gossamer, but it does relate to Long Trail and why CHRP/PReP/PPCP died on the vine. CHRP was supposed to be able to run Windows NT, Mac OS, or AIX. Sun announced a port of SunOS/Solaris. But with the clone market drying up, Microsoft stopped work on Windows NT for PowerPC, and Apple pulled all the Mac OS licenses for clones. So the only available OS for PowerPC based computers would have been AIX (or very early versions of Linux). Not exactly a mass-market appeal. So Motorola left the market and stopped working on motherboards (actually, I'm not sure they ever designed their own or if they just licensed Tanzania & Long Trail), but IBM didn't. Long Trail eventually became a real reference design, and you can see it today in machines like the Pegasos Linux boxes. IBM never made it big because Intel hardware was too entrenched, too good, and too cheap to beat. It was messy, and IBM had a vested interest in selling PowerPC processors to cheap clone makers, but I don't think it would be appropriate to say that IBM was forcing Apple to do anything.
As for Gossamer, Apple took the lessons they learned from CHRP and the clone disaster and rolled it into a new, cheap to manufacture motherboard design. They were surprised when it SMOKED the PowerExpress, but happy because it was so much cheaper to build. It doesn't just beat PowerExpress because of the G3, though. It has a 66MHz system bus with 66MHz SDRAM versus the 50MHz bus and 50MHz EDO DRAM of the PowerExpress. The IDE hard drives were a hair slower than U2 SCSI, sure, but they were also orders of magnitude cheaper. Better still, they didn't have to design and manufacturer a bunch of custom ICs like memory controllers - Motorola did that already with the MPC106.
U2 SCSI could happen on the Gossamer with a PCI card (and did when Apple shipped server models with an ATTO ExpressPCI PSC PCI Ultra Wide SCSI card). Video I/O could happen with PCI cards and Apple's Wings AV card (which has a DAV slot, incidentally). The ONLY area in which PowerExpress beat the Gossamer was in having 6 PCI slots. That is the reason Apple kept the 9600 around. Everything else could be done by Gossamer better, faster, and cheaper.
So, yeah, I think you may have a misunderstanding of that period in the Mac OS market.
Wow, thanks for all that. Saved me reading the book. I don't think this sidetrack throws the thread too far afield as it places your ZIF in the context of where it came from and is a great history lesson for everyone. You know your stuff! Much appreciated.
So, are you thinking your ZIF is a pre-G3 construction? Or was produced concurrently as an option to the G3?
I'm virtually certain my 604e ZIF came out of an RS/6000 43p Model 140. This model used 604e ZIF processors...
I'm making an educated guess that IBM's and Motorola's work on CHRP boards directly lead to the ZIF design used in the Gossamer board and some RS/6000 systems. After spending all that money on designing dead end hardware, it would have been silly to throw it all away without using the designs for something in the future. So my guess (which turned out to be correct) was that the 604e ZIF would have enough in common with a 750 ZIF and the beige G3 design to work in the Gossamer board.
If anyone else has any interesting stories about CHRP hardware or the origins of the Gossamer board, I'm all ears. I'm similarly interested in any other oddball PowerPC ZIF processors that might be out there (especially any Mach 5 modules with L2 cache on board). The only other ZIF modules I know of were designed by Umax for the C500/C600, carried 603e processors (though there was one G3 upgrade for this slot I heard about once), and that is electronically and physically incompatible with Gossamer boards.
I've always wondered if you could drop a G3 into a 43p.
I've got a couple of spare 43p-150 at the office. They are going to be scrapped, so it won't hurt anything to try.
I just might have to scrounge up a G3 and give it a shot. You what would be really cool is if one of the upgrade processors would work. A 1GHz 43p would be pretty damn slick.
i think that has got to be one of the coolest things i've seen in regards to hardware that was never meant to work together and never really talked about. Im still curious about IBM and their new Power series CPU's. would they not make really good machines for both mac and PC?
These might touch on your question, especially the second:
wow i actually wrote something in the 2nd thread way back then!
But. I have some information that may help with this thread. Through another thread here at applefritter I have found the URL for the PowerPC Processor and Cache module information PDF. (http://cache.freescale.com/files/32bit/doc/data_sheet/MPCPCMEC.pdf?fsrch=1&sr=41) Its the original spec sheet from 1997 too. Could this help get an L2 cache card working on a 604e ZIF CPU on a Rev A. PowerMac G3? Ideas?