PPC740L G3 CPU Daughterboard For Blackbird Powerbooks

[Paralel]

From what I've read Newer Technologies was able to essentially drop in a 740 G3 in place of a 603e processor on one of their blackbird CPU daughterboards because they are apparently fully pin-compatible, However, by the time their 740 G3 prototypes were completed, the interfaces for the CPU daughterboards were no longer being produced, so the project was dead in the water.

 

Also, apparently in Japan, a guy was able to take a Newer Technology NuPower 167 Mhz CPU daughtercard, pop off the 603e, drop a 740 G3 in its place, and it worked without any difficulty.

 

From what I can tell, the PPC740L would be a drop-in replacement for the 603e.

 

What do you guys think? Based on this datasheet:

 

http://datasheets.chipdb.org/IBM/PowerPC/7xx/PowerPC-740-750.pdf

 

Would creating a PPC740L G3 CPU Daughterboard For Blackbird Powerbooks using a Newer Technology NuPower 167 CPU daughtercard as a base work? What frequency do you think we could get it to work at? Also, do you think it would be need active cooling? I'm thinking yes. As such, do you think that would be difficult to implement given the space constraints in the blackbird?

[Paralel]

Wow. almost made a big mistake.

 

Checked out the datasheets for the 603e and the 740L. Good thing I did. Turns out the PLL_CFG that is sent for the NuPowr 167 model is 2.5x operation (combined with Core to VCO multiplier) equals out to 167 MHz. However, that same PLL_CFG code was re-purposed by Moto for the 740L for 10x operation! If I had tried that, it would have lead to a multipler of 200X, and tried to drive the processor at 666.66 MHz, when it is rated for, at the very most 533 MHz, which was apparently pushing it. I don't see that turning out very well.

 

Fortunately, the NuPowr 117 model only has a multipler that translates to 3.5 for the 740, which means it would run at 233.31 Mhz, the very minimum of which it is capable. Sadly I can't use an 740L because its minimum base frequency is 250 MHz.

 

Now I have to search for the NuPowr 117 CPU Daugherboard, and secure a 740, along with someone who can do the actual swap For me. The tools I have on hand are nowhere near up to the task. All in the same of SCIENCE!!

Edited December 18, 2018 by Paralel

[Gorgonops]

12 minutes ago, Paralel said:

Sadly I can't use an 740L because its minimum base frequency is 250 MHz

This is sort of an ignorant question, but why would there be a problem slightly underclocking the CPU? I mean, sure, there are old CPU designs that do require at least a minimum clock because they use dynamic refresh for certain elements (IE, they're not fully "static"(*)), I suppose it's possible the 740/750 does too but... even that were the case I don't think a 10% underclock is going to be an issue. The table in the datasheet that claims it has that 250mhz floor even has a footnote that says: "6: Under certain conditions, operation at core frequencies below those stated is possible. Contact IBM for details"

(* Still, though, I'd think a CPU of the G3's vintage would at least support clock throttling for power saving? *shrug*)

[Paralel]

You have a point. The minimum clock speed is met (33.33 Mhz), but the multiplier is wrong, so that might meet this conditions. an interesting idea. I think I will def. try it before anything else since the older 740's are known for their issues. I appreciate your insight.

Edited December 18, 2018 by Paralel

[Paralel]

Okay, turns out I was 100% incorrect about how the NewerTech NuPowr cards are constructed!

 

The 117 MHz uses a QFP, rather than a BGA design like the rest of them, so that's a hard stop.

 

So, that leaves us with the NuPower 167 MHz and the 183 MHz.

 

Now, onto the second problem. The standard 603e is a 3.3v chip, the 740L is a 2v chip, so that's a hard stop.

 

However! It turns out that NewerTech was smart. They used the 603ev in their 167 MHz & 183 MHz models! These run at 2.5v +/- 5%.

 

The original 740 (not the L type 740L) runs at 2.6v +/- 100mv! As such, they should be drop in compatible.

 

The NuPowr 183 MHz model is rather rare and thus, expensive, so I'd rather not use that one for this kind of mod.

 

I'd earlier discounted the NuPowr 167 MHz because of the PLL:CFG multiplier. However, I have discovered it is rather easy to change the PLL:CFG on the CPU daughterboard. The PLL:CFG consists of 4 resistors, PLL:CFG 0, PLL:CFG 1, PLL:CFG 2, & PLL:CFG 3. If a resistor is absent it is has a value of 0, if it is present it has a value of 1. As you can imagine with 4 spots, we have 16 combinations. Value 1100, corresponds to a multiplier of 8x. With a bus of 33 MHz, that gives us 266 MHz! This multiplier can also work with a clock that goes down to 25 MHz. Furthermore, as you can imagine, it has multiple power saving features like "Doze", etc...

 

As it stands, there shouldn't be absolutely any problem with converting a NuPower 167 MHz into a G3 740 266 Mhz!

 

Now I just need to gather the parts and find someone who can do the processor swap for me since there is no way I have the tools and finesse necessary to do this without destroying the CPU daughterboard.

[Trash80toHP_Mini]

Sounds like a great project and very promising, good luck with getting it all together. Got pics of one of those PPC boards? I've never seen a 68040 bus ->PPC implementation.

[Bolle]

If they have 2.5V chips working on there then there would have to be some kind of regulator on the CPU card. You should be able to set the desired voltage coming out of that one as well.

[CC_333]

Now I have to figure out how to get one of these!

 

My PPC 540c certainly has a 603 of some kind, but I'm not sure if it's a NuPowr 167 MHz (it's almost certainly 117 MHz). It's inaccessible due to being buried deep in storage, or else I'd look.

 

c

[Paralel]

The G3 processor in question that would be needed is: MPC740ARX266LH

 

Top of the NewerTech NuPowr 167 MHz:

 

 

 

Bottom of the NewerTech NuPowr 167 MHz:

 

 

 

I don't know where I'd be able to adjust the processor core voltage. See it anywhere on these photos?

 

I also have to figure out which of the four clusters of resistors is the PLL:CFG.

 

The frustrating part is that I have a lead on the processor I need, and I actually had a NewerTech NuPowr 167 MHz card, but I traded it for having my Mac Classic II refurbished, (all cleaned up, joints reflowed, recapped with ceramic caps, replacement of the external SCSI port [which was busted]). Now I need to try and get one.

Edited December 19, 2018 by Paralel

[Paralel]

CC_333, if your card is a NewerTech NuPowr 167, if you can find someone who can expertly do the transplant, you are more than welcome to the outcome of the project, including my PPC 740 G3 processor, free of charge. I'm much more interested in knowing if this is possible than actually keeping the hardware.

Edited December 19, 2018 by Paralel

[CC_333]

Oh, wow! Thank you!

 

Alas, I'm pretty sure mine was the 117 MHz model (and thus, probably un-upgradeable). I found it in storage, and I *almost* brought it home, or else I'd look inside and find out.

 

I'll be out of town for a few days, but maybe I can stop by the unit on my way out and get it so I can take a look while I'm away (I'll need something to do )...

 

While I'm a little weary of potential damage, I'm excited because it would be fun to have the *best* 540c available (the only thing better would be a G3-upgraded 550c )

 

c

[Trash80toHP_Mini]

I'm beginning to have doubts about an underlying assumption made in these ProcSwap threads. I'm wondering if "drop in replacement" is necessarily implied by two CPUs in a series having a compatible BGA interface. A newer, faster, bigger cache CPU in a series is part of an upgrade path for existing systems, not necessarily to be implemented without specified changes to a preexisting logic board design for proper function.

 

In looking at implementation specs. for several G3 and G4 CPU generations I see tighter and tighter support circuitry spec'd for the "solder side" of the CPU in terms of caps for power supply and decoupling for specific sections of the CPU to reduce/filter noise within the CPU itself. That's IIRC and as far as I understand it..

 

You might want to take a really close look at what's implemented under your processor card to check if what you find that was required for implementation of the native CPU resembles closely enough the specs for the target CPU support system.

 

It also looks to me that wholesale changes in grid arrays may be following major changes in the arrangement of the support circuitry specified for down under.

 

Color me clueless.

[Bolle]

5 hours ago, Paralel said:

I don't know where I'd be able to adjust the processor core voltage. See it anywhere on these photos?

 

I see the two transistors that are part of the core power supply but I don’t see what controls them.

It might as well be possible they just cut the 5V supply in half with a 50% duty cycle to get the 2.5V. No easy way to change that then.

One would have to trace out where the drive signal for those two is coming from.

[Paralel]

2 hours ago, Trash80toHP_Mini said:

I'm beginning to have doubts about an underlying assumption made in these ProcSwap threads. I'm wondering if "drop in replacement" is necessarily implied by two CPUs in a series having a compatible BGA interface. A newer, faster, bigger cache CPU in a series is part of an upgrade path for existing systems, not necessarily to be implemented without specified changes to a preexisting logic board design for proper function.

 

In looking at implementation specs. for several G3 and G4 CPU generations I see tighter and tighter support circuitry spec'd for the "solder side" of the CPU in terms of caps for power supply and decoupling for specific sections of the CPU to reduce/filter noise within the CPU itself. That's IIRC and as far as I understand it..

 

You might want to take a really close look at what's implemented under your processor card to check if what you find that was required for implementation of the native CPU resembles closely enough the specs for the target CPU support system.

 

It also looks to me that wholesale changes in grid arrays may be following major changes in the arrangement of the support circuitry specified for down under.

 

Color me clueless.

 

It looks like in this instance it may not be too serious of an issue.

 

According to the white papers for both processors, Decoupling Recommendations and Connection Recommendations are identical between the 603ev and the 740.

 

The Pull-up Resistor Requirements show several differences between the 603ev and the 740.

 

The 603ev says:

 

"The 603ev requires high-resistive (weak: 10 KOhms) pull-up resistors on several control signals of the bus interface to maintain the control signals in the negated state after they have been actively negated and released by the 603e or other bus master. These signals are—TS, ABB, DBB, ARTRY.

 

In addition, the 603ev has three open-drain style outputs that require pull-up resistors (weak or stronger: 4.7 KOhms–10 KOhms) if they are used by the system. These signals are—APE, DPE, and CKSTP_OUT.

 

During inactive periods on the bus, the address and transfer attributes on the bus are not driven by any master and may float in the high-impedance state for relatively long periods of time. Since the 603ev must continually monitor these signals for snooping, this float condition may cause excessive power draw by the input receivers on the 603ev. It is recommended that these signals be pulled up through weak (10 KOhms) pull-up resistors or restored in some manner by the system. The snooped address and transfer attribute inputs are—A[0–31], AP[0–3], TT[0–4], TBST, TSIZ[0–2], and GBL.

 

The data bus input receivers are normally turned off when no read operation is in progress and do not require pull-up resistors on the data bus.

 

The 740 says:

 

"The MPC740 requires high-resistive (weak: 10 KΩ) pull-up resistors on several control signals of the bus interface to maintain the control signals in the negated state after they have been actively negated and released by the MPC740 or other bus masters. These signals are TS, ABB, DBB, and ARTRY.

 

In addition, the MPC740 has one open-drain style output that requires a pull-up resistors (weak or stronger: 4.7 KΩ–10 KΩ) if it is used by the system. This signal is CKSTP_OUT. [APE & DPE are also mentioned by the 603ev]

 

During inactive periods on the bus, the address and transfer attributes on the bus are not driven by any master and may float in the high-impedance state for relatively long periods of time. Since the MPC740 must continually monitor these signals for snooping, this float condition may cause excessive power draw by the input receivers on the MPC740 or by other receivers in the system. It is recommended that these signals be pulled up through weak (10 KΩ) pull-up resistors or restored in some manner by the system. The snooped address and transfer attribute inputs are A[0–31], AP[0–3], TT[0–4], TBST, and GBL. [TSIZ[0–2] are also mentioned by the 603ev]

 

The data bus input receivers are normally turned off when no read operation is in progress and do not require pull-up resistors on the data bus. Other data bus receivers in the system, however, may require pullups, or that those signals be otherwise driven by the system during inactive periods. The data bus signals are DH[0–31], DL[0–31], DP[0–7].

 

If address or data parity is not used by the system, and the respective parity checking is disabled through HID0, the input receivers for those pins are disabled, and those pins do not require pull-up resistors and should be left unconnected by the system. If all parity generation is disabled through HID0, then all parity checking should also be disabled through HID0, and all parity pins may be left unconnected by the system." [This bolded section in the 740 whitepaper is not mentioned at all in the 603ev whitepaper]

Edited December 20, 2018 by Paralel

[Paralel]

Based on the differences in the pull-up resistor information, does this look like a potential issue?

[Trash80toHP_Mini]

Not a clue, sorry, that's the kinda stuff I'll need help with if the proposed 1400 processor card can get past the signal compatibility hurdle. Glad to hear the decoupling and supply specs mesh so perfectly for your project. I've got board space to address such differences between 603e and 750FX where you wouldn't have had any. Hopefully pull-ups will work out just as nicely. The resistors must be on there in place for the 603ev and may be accessible to swap in proper values for the 740?

 

I've gotta find more info on my CPUs, just wanted to give you quick a heads up on my suspicions before you went to the trouble of getting all your parts lined up.

 

I like that you're into this for the satisfaction of seeing that it can be done.

 

 

[Paralel]

It's the only way to work on this type of stuff, for the love of it, and my love is testing a hypothesis.

Edited December 23, 2018 by Paralel

[techknight]

I only have the 117 version as well. Seems as thats the only ones I see, and I only ever saw that one. 

 

The newer faster ones are alot more rare, and honestly due to that, your less likely to find willing participants. 

Edited December 23, 2018 by techknight

[Paralel]

1 hour ago, techknight said:

I only have the 117 version as well. Seems as thats the only ones I see, and I only ever saw that one. 

 

The newer faster ones are alot more rare, and honestly due to that, your less likely to find willing participants. 

It's funny, the 167 MHz is supposedly the one of which they sold the most, with the 183 being quite rare, and the 117 being the second most common (and of course, their G3 only made it into the prototype phase)

Edited December 23, 2018 by Paralel

[techknight]

I only ever found the 117, and it was in a parts machine. Its got issues though, it will randomly freeze or bomb until it warms up. Cache issue possibly. 

 

I ended up finding the apple official one as well. 

Edited December 23, 2018 by techknight

[Paralel]

14 minutes ago, techknight said:

I only ever found the 117, and it was in a parts machine. Its got issues though, it will randomly freeze or bomb until it warms up. Cache issue possibly. 

 

I ended up finding the apple official one as well. 

The frustrating part is I actually had a 167 years ago, traded it to someone for some work on my Macs that were having serious hardware issues.

 

Thought I'd never use it for anything...

[techknight]

haha. I feel the same now with all my blackbird stuff. Everything I own is parts units because of the brittle plastic. Thats why I wish the CAD/3D printing stuff went somewhere. 

[Franklinstein]

It has been done. It'll work with no problems. The 740 and 745 are both limited to 10x multipliers, so max of 333MHz unless you also want to try to overclock the memory bus to 40MHz. I don't know if an OC like that will affect anything else; only the memory and the first bridge chip operate at the full 33MHz while the rest of the system is on a separate buffered '030 bus that runs at 25MHz or less. If you did OC it, I'd probably recommend pulling the modem and its daughter card and devising a more elaborate heatsink or itty bitty fan to go in there to cool things down. Also probably a SSD, maybe drill a few holes in the frame stiffener to aid airflow.

 

Like the PB 2300, 5300, and 1400/117, none of these had any sort of L2 cache, just the 16k+16k L1 caches that were built-in to the 603e. The 740's 32k+32k L1 caches would likely provide a noticeable boost, in addition to the higher clocks.

 

That said, I'd be willing to do it (I posted in the Trading Post that I was looking for a damaged upgrade card so as not to ruin a good one), but my current rework tools are limited (just a computer-controlled handheld hot air nozzle) and I haven't yet either found a good rework station for under $7k or the justification to pay full price for one. Supposedly I would also need some sort of oven to bake the new chips before installing them to the PCB (to prevent chip popcorning), and I don't have that either so my success would be a maybe at best. Modern microelectronics is an expensive hobby and I sadly can't justify getting that into it. Maybe if it's still in demand when I retire in 10 years I can do it as a side job but I just don't have the time or space for much more than basic repairs or mods now.

[Paralel]

The 740 H revision removed the 10x multiplier, the multiplier is now limited to 8x and maximum bus is now limited to 83.3. Apparently stability problems were discovered with the 10x multiplier and the 100 MHz bus on previous revisions. The 745 in this case is not pin compatible with the 603ev because of a one pin difference. I think a CSE input was switched to a BVSelect output. Plus, the core voltage of the 745 is not drop-in compatible with the 603ev since it's 2.0v, whereas the 603ev is 2.6v (of course, that's not insurmountable).

 

I definitely don't want to try and overclock the memory bus to 40 MHz, the heat issue is likely to be bad enough already.

 

I definitely plan on a heatsink and a fan of some kind since the heat dissipation, worst-case scenario, doubles from 4 W to 8 W. I doubt the design, without these additions can handle that extra heat dissipation.

 

Hey, if you're willing to try, I'm game. I can actually get more than a few of the exact kind of processor needed, so it won't matter if one or two gets killed.

 

A damaged card is a great idea, assuming it can be made operational again. Much better than potentially murdering a good card for an experiment.

 

I see you're in Tokyo. That works out great, I'm used to sending packages to Japan since my brother lives over there.

 

One thing I'm interested in, from the pics (which I admit are crap) of the NuPower 167, can you figure out which of the resistor sets is the PLL:CFG? I have looked at it again and again and just can't seem to identify them by sight. The current settings for the PLL:CFG needs to be changed so the card can run at 266 MHz. If one tried to run the 266 MHz G3 with the current PLL:CFG, they will be disappointed to find it only runs at 167 MHz.

Edited December 24, 2018 by Paralel

[techknight]

I might be getting my hands back onto the old ACHI rework station I used to use back in the X360 days. Honestly though, I have no use for it as I dont do BGA rework anymore. It was a royal PITA when I did. Every step had to be exactly perfect or things would go wrong in a hurry. 

 

So if I can get my hands back on it again, It would be available to who needs it. It requires a 220V circuit though. 

 

The last time I used it was doing repairs on a failed 500Mhz Sonnet L2 card. 

 

The chip was ceramic so it wont popcorn. the PCB however.... 

Edited December 24, 2018 by techknight