601 processor replacement experiments

[Trash80toHP_Mini]

HRMMM? Another PDF hunt coming up. Depending on what that one additional pin is up to, 750CX might just work. 6mm more in package size shouldn't matter, assuming the BGA arrangement remains the same. Even if they spread it out, putting an opening in the center it should still fit on a CQFP adapter PCB.

Gotta take a look at the 2300c board to see if the adapter could hang over the edge of the pads a bit for bus multiplier resistors, caps and the like. Hunting down the multiplier components on the logic board would save that space. Gotta do that anyway for my 200MHz 603e (might be 180MHz, don't recall offhand) eventually anyway for that long neglected project.

 

[Franklinstein]

HRMMM? Another PDF hunt coming up. Depending on what that one additional pin is up to, 750CX might just work. 6mm more in package size shouldn't matter, assuming the BGA arrangement remains the same. Even if they spread it out, putting an opening in the center it should still fit on a CQFP adapter PCB.

The 603e/740 use the same basically solid grid of balls as per the diagram in your post upthread. The 750CX does not: it uses a different grid layout with a big void in the center, hence the larger package. This void is where the processor die sits: it's mounted flip-chip style to the underside of the chip carrier and sealed with some black goop. This is the reason the 750CX is such a low-profile chip and what makes it ideal for trying to shoehorn into a PowerBook with an adapter. It's also a very cool-running chip and the fact that there is no naked processor die protruding from the top makes it easier to attach a thermal solution, potentially even the original thermal pads used by the QFP 603e.

 

[Trash80toHP_Mini]

Dang, what was up with IBM's documentation, no package description in the user manuals? But I found the datasheet:

http://datasheets.chipdb.org/IBM/PowerPC/750/PowerPC-750CX.pdf

http://softpixel.com/~cwright/papers/tech/750CX_CXe_UM_prel_V1.1_09May02.pdf

Wikipedia says the 750CX could be harvested from one iBook and iMac revision. Anybody know which ones/what speeds? [}]

Balls/pads on the CPU/adapter are only 24.13mm center to center, well within the 34.6mm footprint of such an adapter. The 27mm outside dimensions of the package only leaves 3.8mm of adapter PCB surrounding the 750CX, just enough a ring of caps resistors and leads to a 3.3V doohickey tucked away somewhere.

So of course it won't be signal compatible with the 603e. :-/

NAP! Still -200 upload error so off to iFrog I go:

 

[Trash80toHP_Mini]

Something strange going on with linking to ImageShack  .  .  .

 

[Trash80toHP_Mini]

Image displays correctly in the reply window and reverts to a link when posting.

 

[trag]

Made an editing mistake in my previous post.    The voltage adjustable I/O that I attributed to the 750 was not a feature until the 750L, if I'm reading the datasheets correctly.

jt, why salvage 750CX when you can buy them new and undamaged and with fresh balls fro about $20 from Quest?  

Though, if you're going to the trouble to build an adapter board, unless the height is an obstacle, I'd use the 750GL instead.    20X  multiplier, instead of 10X, and four times the L2 cache.  True, the chip will cost 2X, ~$40, but compared to the amount of effort that would go into adapting either chip, the extra $20 is not significant.

If you go to this link:   https://www.nxp.com/products/no-longer-manufactured/host-processor:MPC603E?&tab=Documentation_Tab&linkline=Data-Sheet

and look near the upper left corner, you'll see a pull down menu with the label "No Longer Manufactured".   Pull that down and scroll down to the part for which you want documents.   When the page comes up, it will be on "OVERVIEW".    Click on "DOCUMENTATION" in the heading and then enjoy.    Keep an eye out for "more" buttons under the lists of documents.   Also, remember that Motorola's designation for the PPCs is going to be MPC, not PPC.   So you're scrolling down to MPC750, MPC601, etc.

Some documents are missing.   I'm still not seeing a good datasheet for the 68030 and 68040, IIRC.   There's the big fat user's manual, which is usually  harder to find, but not in this case.

For IBM datasheets, Quest Components is pretty good about having datasheets downloadable with their components for sale.   So, e.g., when shopping their list of in-stock 750CX - 750GX, I was able to download IBM datasheets for each of them, by clicking through on the part and selecting Specifications.

 

[Trash80toHP_Mini]

That was a booboo on my part last night, didn't realize the 750CX was one of the CPUs you posted as available. Booboo'd again this morning, it happens. Too dopey tonight to make sense of questcomp.com? Got linkage to the 750 series? Is that even the correct site?

I guess it all boils down to signal compatibility. Peeked at the 603e docs collected in the G3 info folder and found three different versions with different clock multiplier setups and I have no idea which type is in the processor module donor. At this point I'm going to bet I'm stuck with a 4x multiplier which revs that 200MHz part transplanted into the 2300c all the way up to  .  .  . 100MHz. :blink: My brain hurts  .  .  .

edit  .  .  .  bourbon helped.

 

[Trash80toHP_Mini]

jt, why salvage 750CX when you can buy them new and undamaged and with fresh balls fro about $20 from Quest?  

I hate page break discontinuity: see booboo notice above.

Though, if you're going to the trouble to build an adapter board, unless the height is an obstacle, I'd use the 750GL instead.    20X  multiplier, instead of 10X, and four times the L2 cache.  True, the chip will cost 2X, ~$40, but compared to the amount of effort that would go into adapting either chip, the extra $20 is not significant.

You'd said the 750GL etc were 292 pin BGA packages and that the 750CX is a 256 pin BGA. That'a just one contact more than the 603e which came in "240-pin ceramic quad flat pack (CQFP) or 255 ceramic ball grid array." So I was guessing the later versions wouldn't work. A 10x multiplier cranks the 2300c up to 250MHz with 256K of L2 Cache. My 1400c has a CrescendoPB/G3/466MHz/1MB, so getting my precious Duo up to about half of that would be a major achievement  .  .  .  if only I were up to that task. :-/

 

[trag]

I hate page break discontinuity: see booboo notice above.

You'd said the 750GL etc were 292 pin BGA packages and that the 750CX is a 256 pin BGA. That'a just one contact more than the 603e which came in "240-pin ceramic quad flat pack (CQFP) or 255 ceramic ball grid array." So I was guessing the later versions wouldn't work. A 10x multiplier cranks the 2300c up to 250MHz with 256K of L2 Cache. My 1400c has a CrescendoPB/G3/466MHz/1MB, so getting my precious Duo up to about half of that would be a major achievement  .  .  .  if only I were up to that task. :-/

It's one more ball, but the packages for the 750CX and the 603/740 are different sizes, pitches.    So you cannot solder a cx in place of a 603.  

So, you would need to make an interstitial board.   As long as you're building a special adapter board, you may as well build it for the best chip available.

 

[Trash80toHP_Mini]

Not really applicable as I'm not really thinking about BGA to BGA adapters. I'm really only interested in BGA to QFP 603e pads. If I get around to a layout attempt, I may as well try Ball to Ball as you've suggested as well. Do you really think 292 BGA might be compatible with 240 pin QFP 603e? Might there merely be more ground lines due to the tighter grid pitch pattern?

 

[trag]

All these chips do 601X bus, so they are more or less compatible.    

On Quest's website, in the search box at the top, try ibm25ppc750 as a search term.    There's a box near (above?) the search box for "in stock".   I usually click that as well.   It must be clicked before the search.

 

[Trash80toHP_Mini]

Dead slow at work tonight, so I checked out datasheets. 750CX is looking much better than the 750FX.

Power filtering looks easier and the decoupling recommendations are only severe. Decoupling recommendations for the 750FX are downright draconian. No way to meet them on an adapter for a PowerBook that I can see. The most important of caps are likely the ones dead center under the die. Spec is to put the decoupling cap electrodes directly opposite their corresponding BGA pins on the underside of the board where possible or on short paths from thru-vias adjacent to the decoupling caps.

With inverted die surrounded by its box of contacts, it might be possible to achieve adequate decoupling for the 750CX by nibbling away at it from the edges? 10x multiplier of the 750CX cuts the clock down to 250MHz as opposed to 500MHz for the 750FX. But the lower clock means reduced decoupling requirements and the possibility of meeting them. That 250MHz cup would be brimming over, not half empty. The doubled L2 of the 750FX would have been nice, but in a machine with none, I think 256K of L2 would do.

Still not up to the task now by any means, if ever. But feasibility study is fun all on its ownsome. :approve:

 

[Trash80toHP_Mini]

@trag Now I'm wondering if decoupling caps present on the 100MHz 603e of the 2300c might be sufficient for running a 600MHz 750CX at 250MHz?

Haven't even looked for voltage differential complications or function of that one extra ball. I guess it's back to basics in the kibosh hunt.

 

[Trash80toHP_Mini]

Back on topic!

I wonder if they also redesigned the 601v to be intentionally disruptive? I am guessing no, since the whole PPC project was in a state of turmoil for the first few years as they fine-tuned the platform. I mean, the PPC 740 is both pin and signal compatible with the BGA 603e so at least if they were being vexatious at first they changed their mind in later years.

Not sure about that, the same thing appears to happen in the 740/750 generations.

Looking at the datasheets for the 604 and 601 it looks like it's a hard no. They're both 304 pin packages but the pin assignments don't match up at all. (Like, literally, the same pin is a voltage supply on one but a ground on the other, and the signals are likewise completely scrambled. It almost looks like they intentionally made them as incompatible as humanly possible.) So at the very minimum you'd need some sort of adapter board.

Looking at the decoupling recommendations for the two 750 versions above, scrambling the signals around in order to rearrange power and ground might have been done out of necessity as the rising clocks raised the bar for noise containment.

@FranklinsteinThe BGA to BGA adapters you were talking about would be a lot more practical for desktop/tower boards than QFP to BGA. Doing the adaptation on a smaller board with a larger board balled to that inverted pagoda style offers double the layers from the same multilayer prototype run. Increased surface area on the upper surface for dohickys of all sorts becomes available The possibility of placing SMD caps on the underside of the upper board in between like parts on the logic board cmight work as well. Funky outlines of the two adapter levels and underside components would make for a delicious bar puzzle design exercise.

 

[Trash80toHP_Mini]

Slow again last night , so I worked out 750CX to 603e signal differentiation.  There are about 28 signals on the 603e pinout that ate missing from the simplified 750CX pinout. From the looks of it at this point they can be ignored. They appear to be related to external L2 implementation.

_____________________________________________________________________________________________________

edit: sorry about the formatting, I guess I'll never get the manual quote codes figured out for the new forum software. The irritating situation where you can't edit a manually encoded quote to fix encoding errors seems like it will ensure that. :-/

_____________________________________________________________________________________________________

One 750CX signal is a combination of two lines on the 603e pinout, spec says it needs to be held high:

p.39 

7.9.4  DBWO/L2_TSTCLC

One pin has two functions: DBWO and L2_TSTCLK dependent upon the LSSD_MODE pin. When the LSSD_MODE pin is low, the DBWO/L2_TSTCLK pin is set to L2_TSTCLK function which is used during the manufacturing process for testing.
 
When the LSSD_MODE pin is pulled to the high state, the DBWO/L2_TSTCLK pin is set to DBWO which is identical to those descriptions given in earlier versions of the PowerPC 750CX RISC Microprocessor’s User’s Manuals.


Only one 750CX signal is missing from the 603e pinout:

p. 38 

7.9.3  1.8V and 2.5V I/O Signal Support

Selection between 1.8V and 2.5V I/O is accomplished using the BVSEL pin. If BVSEL is set low then the 1.8V mode is enabled. If BVSEL is set high, then the 2.5V mode is enabled. Due to bus timings restrictions, only uniprocessor implementations are recommended. No multiprocessor support is advised.

So it's back to voltage level basics to determine compatibility. Signals I'm comfortable with, voltages are a brand new concern. Little help?

Data pulled from these two PDFs:

PowerPC 750CX Microprocessor Datasheet - 43 pages - June 2001

PowerPC 603e Hardware Spec - 40 pages - November 1996

PowerPC 603e User's Manual - 433 pages - September 1995

 

Po

 

[Trash80toHP_Mini]

Managed to kill the second quote above trying to edit the post. :

Only one 750CX signal is missing from the 603e pinout:

Quotation___________________________________________________
p. 38   7.9.3   1.8V and 2.5V I/O Signal Support
Selection between 1.8V and 2.5V I/O is accomplished using the BVSEL pin. If BVSEL is set low then the 1.8V mode is enabled. If BVSEL is set high, then the 2.5V mode is enabled. Due to bus timings restrictions, only uniprocessor implementations are recommended. No multiprocessor support is advised.

__________________________________________________________

So it's back to voltage level basics to determine compatibility. Signals I'm comfortable with, voltages are a brand new concern. Little help?

edit: IBM PowerPC® 750CX/750 RISC Microprocessor User’s Manual

http://softpixel.com/~cwright/papers/tech/750CX_CXe_UM_prel_V1.1_09May02.pdf

 

[Trash80toHP_Mini]

Back on topic:

@Franklinstein your mention of the dual voltage transplant friendliness of the PowerMac 8100 piqued my curiosity.. I pulled my spare Radius 81/110 board out of its PEx drawer nest to have a quick look. It appears to be a not overly hostile, if not BGA to BGA 750FX adapter friendly design. There are scads of decoupling caps directly under the 601 ball contacts that might be rearranged to take care of much, if not all requirements of the incredibly QFP adapter hostile 750FX package.

Given the external L2 oriented similarities the 601 has with the 603e, there may be more than enough decaps to have some left over! I'll take a WAG that leaving the L2 DIMM slot empty will be enough to remove any conflicts. I wonder if a 1GHz 750GX might be floating around available somewhere for my metal can monster? [}]

edit: Benchmarking a RadiusMonsterBox/750GX/1GHz against 1997's Beige G3 would be interesting. 37MHz vs. 66MHz system bus bottleneck for memory access might be attenuated by the extra 512K of L2, the 4x clock bump over the stock BG3 might give the 81/110 a run for the money? 

http://datasheets.chipdb.org/IBM/PowerPC/750/PowerPC-750GX.pdf

 

[Trash80toHP_Mini]

Well there's another booboo. 601's not BGA, thought it looked odd from the edge but didn't put the multiple diopter readers on until just now. So the decaps and resistors clumped under its footprint probably aren't close enough to work with the 760FX, bummer. But you never know and the 600MHz 750CX might just work.

Partial PPC family chart: https://www.okqubit.net/ppchikaku.html

PowerPC 601 Technical Summary - Advance Information - 32 pages - November 1993

PowerPC 601 User Manual - 777pages - 1995

 

[Trash80toHP_Mini]

Back in here and once again I'm totally confuzzled! I was thinking earlier today that a version of the G4 was compatible with one of the G3 CPUs. IIRC that's the Wegener Media Pismo G4 hack. Looking into the possibility of doing the same type of thing with a Crescendo PB G3 for the 1400 just for the hell of it. I've got another G3 Accelerator from NewerTech(?) if it's a better match.

Still gotta get all these procswappin' thread links together in one topic for easier reference. Crazy stuff!  :wacko:

 

[Franklinstein]

The support chips for the original Power Macs were all the same and rated to at least 40MHz (and were used as such in the 7100/80, 8100/80, and 9150/80 and 120), so simply swap out the clock crystal to change the base speed. I did that on an 8115/110 and it now runs at 120MHz with no problems. Not much of a noticeable performance boost though, but it's well within the margin of safety (by my definition: anything within 10% of original rated speed is "safe", 20% is "maybe", >30% is "gonna melt unless I install these huge, expensive, loud coolers").

I've seen a couple references to people swapping a 7400 onto a 750-based Pismo CPU card but I'm not sure how they do it. I haven't compared the 7400's pin-out to the 750's because I simply assumed they were different (IIRC the 7400 has a wider address and/or data bus than the 750). It's possible that the 7400 and 750 share the same or close enough pin-out to share a similar mount but at this point I can only speculate. If they are interchangeable, I know you won't be able to do this with any chips other than the 750/750L and the 7400 or 7410; the later chips are most definitely incompatible.