Interest in adapting 744x to 7400?

[cobalt60]

Any bets on whether this works?:

Ha awesome. If we can take any required firmware hacks out of the equation, I will bet one G4 7400 module that it works.

 

[alectrona6400]

Someone I knew a while back actually tried that same thing, and I believe it worked. Both chips are pin compatible.

 

[herd]

As predicted, the G3 chip works in the AGP. It's basically the reverse of upgrading a PowerBook G3 Pismo. In fact the PPC750S came out of a Pismo. I can't find much about this chip from the cryptic IBM labeling and secret documentation. But based on where I got it, it must be from around 1999 and before the 750L? The PVR isn't very helpful, as the version is later than the 750CX, which doesn't make sense.

Anyway, this is further evidence that a 7447/7448 should work in a Pismo if the firmware can be fixed. Does anyone want a G3 AGP CPU for testing or novelty, or should I take it apart?

 

[macuserman]

As predicted, the G3 chip works in the AGP. It's basically the reverse of upgrading a PowerBook G3 Pismo. In fact the PPC750S came out of a Pismo. I can't find much about this chip from the cryptic IBM labeling and secret documentation. But based on where I got it, it must be from around 1999 and before the 750L? The PVR isn't very helpful, as the version is later than the 750CX, which doesn't make sense.

Anyway, this is further evidence that a 7447/7448 should work in a Pismo if the firmware can be fixed. Does anyone want a G3 AGP CPU for testing or novelty, or should I take it apart?

View attachment 75088

That's awesome, if there was an adapter to stab a 750gx onto a G4 board it would be awesome to push the boundaries. I'd love to have a high speed G3 that supported AGP. I assume you also get the faster bus benefits from doing this swap as well?

 

[macuserman]

That's awesome, if there was an adapter to stab a 750gx onto a G4 board it would be awesome to push the boundaries. I'd love to have a high speed G3 that supported AGP. I assume you also get the faster bus benefits from doing this swap as well?

I've been searching for the powerlogix pismo bluechip upgrade for years, if I ever find one I"m going to reverse engineer the adapter it would open up so many possibilities for us.

Macintosh powerbook Pismo processor card PowerLogix G3/900

 

[LightBulbFun]

As predicted, the G3 chip works in the AGP. It's basically the reverse of upgrading a PowerBook G3 Pismo. In fact the PPC750S came out of a Pismo. I can't find much about this chip from the cryptic IBM labeling and secret documentation. But based on where I got it, it must be from around 1999 and before the 750L? The PVR isn't very helpful, as the version is later than the 750CX, which doesn't make sense.

Anyway, this is further evidence that a 7447/7448 should work in a Pismo if the firmware can be fixed. Does anyone want a G3 AGP CPU for testing or novelty, or should I take it apart?

View attachment 75088

Oh awesome this is something me and @Daniël have been meaning to try out/have wanted to see done just for the novility of it, it is very awesome to see it works

but I am also curious from a technical presepsctive, do Core Image ATI Radeon cards work? NVIDIA cards require a G4/Altivec to function, you can boot a GeForce Core Image card on a G3 but while the OS will boot any OpenGL applications will crash out, so I am curious if that is true for ATI cards like a Radeon 9800 etc (I actually have a PCI to AGP adapter card on my way from Daniel to try and test this out myself by way of sticking a 9800 Pro into a G3 Blue and white/G3 beige/Power Mac 9600 and seeing what happens there )

also I am curious was it a streight swap, or did you have to adjust anything on the daughter-card like MPX/60x bus straps etc? (if so which resistors are those? this is something I would like to try myself at some point )

as mentioned above I have also often thought about the CBGA360 to 750FX/GX interposer card that the PowerLogix Pismo upgrade used, and if that could be applied to other macs, a 750GX equipped Sawtooth would be a speedy low power thing for where you dont need altivec

as an aside I have also wondered what the PPC750S was about, a few Pismo's I have upgraded for people in the past have also contained them

 

[herd]

The G3 uses a 60x bus mode. The G4 can use either 60x or the improved MPX mode when supported by the other system hardware. For those who say that the G4 is just a G3 with vector instructions tacked on, here is a 450MHz 750S with 100MHz bus and 1MB of external L2 cache, running nbench compiled with -mcpu=G3 -mtune=G3, compared to a 450MHz 7400 in 60x mode running the same G3 build binary, compared to the same 450MHz 7400 in MPX mode running nbench compiled with -mcpu=G4 -mtune=G4. These were back to back tests with otherwise same settings, hardware, OS, etc.:

NUMERIC SORT             2.1      2.13     2.22
STRING SORT              1.9      4.56     4.7
BITFIELD                 2.69     2.69     2.49
FP EMULATION             3.87     3.93     5.29
FOURIER                  2.96     3.38     3.38
ASSIGNMENT               4.9      4.9      6.91
IDEA                     4.05     4.05     4.12
HUFFMAN                  4.46     4.45     4.27
NEURAL NET               3.01     3.91     4.65
LU DECOMPOSITION         5.53     6.6      6.78
==========================ORIGINAL BYTEMARK RESULTS==========================               
INTEGER INDEX           12.945   14.729    16.078
FLOATING-POINT INDEX     6.614    7.996     8.549
Baseline (MSDOS*)         Pentium* 90, 256 KB L2-cache, Watcom* compiler 10.0          
==============================LINUX DATA BELOW===============================
MEMORY INDEX            2.926     3.917     4.322
INTEGER INDEX           3.479     3.504     3.794
FLOATING-POINT INDEX    3.669     4.435     4.741
CPU                 : 450MHz 750S, 7400 in 60x, 7400 in MPX
L2 Cache            : 1MB L2
OS                  : Darwin 8.11.0
C compiler          : gcc version 4.0.1 (Apple Computer, Inc. build 5370)

The next time I have this set up I can try a few video cards.

For 1.8v systems I think the later G4 CPUs make sense. For 3.3v systems I think it would be interesting to compare a 7410 with external cache against a 750G with higher clock and internal cache. That might be a fair fight; I suspect it would come down to the particular software. For the 255-pin format used by the 740 G3, 604, etc. the 750G on an adapter might be the best option because there is no provision for external cache in the first place.

 

[LightBulbFun]

interesting I know the G3 is a 60x only CPU, what im wondering is for those who want to do this swap themselves, was it a straight swap, or does one have to change some resistors to change the bus mode, or does the CPU/PowerMac automatically take care of that?

 

[alectrona6400]

I actually had no idea about this. Very interesting.

 

[LightBulbFun]

Anyway, this is further evidence that a 7447/7448 should work in a Pismo if the firmware can be fixed. Does anyone want a G3 AGP CPU for testing or novelty, or should I take it apart?

BTW on this front, the Pismo's latest firmware is 4.1.9 which is well supported by most of the BootROM CPU upgrade firmware patches out there, I would try the PowerLogix Firmware patcher first, and then the GigaDesigns one

tho of course I would always make a backup of the BootROM first

 

[LightBulbFun]

im just wondering @herd have you had a chance to try one of your 750/7400/7410 to 744X interposers with your Pismo yet? I bet a 7448 Pismo would be a very nice performance portable

 

[herd]

I tried installing a 7447 on a Pismo board using a firmware patch from joevt. It didn't work. It looked like the CPU would run for around 100ms then go idle. Everything I could measure looked ok, and increasing the frequency increased the power draw. I ended up installing a 7410 on the Pismo board (which works fine) and I moved the 7447/adapter to a different board (which works fine). I suspect there is still some firmware issue or perhaps some subtle difference in wiring between the Pismo boards and others.

I don't have any more spare Pismo boards, but the next time I find one I'll try again and spend more time reverse engineering how it's wired.

 

[LightBulbFun]

I tried installing a 7447 on a Pismo board using a firmware patch from joevt. It didn't work. It looked like the CPU would run for around 100ms then go idle. Everything I could measure looked ok, and increasing the frequency increased the power draw. I ended up installing a 7410 on the Pismo board (which works fine) and I moved the 7447/adapter to a different board (which works fine). I suspect there is still some firmware issue or perhaps some subtle difference in wiring between the Pismo boards and others.

I don't have any more spare Pismo boards, but the next time I find one I'll try again and spend more time reverse engineering how it's wired.

interesting what clock frequency/voltage where you running at? I wonder if it was tripping some sort of over current protection somewhere?

I know that back in the day there was a prototype 900Mhz G4 Pismo upgrade https://lowendmac.com/ed/bashur/12db/pismo-unicorns.html

I always wish there was more details of it, but it does strongly suggest it should be doable


one CPU to try just as a sanity check might be a 7441 or 7445 CPU, since I do believe the latest Pismo firmware does natively support that CPU, so you could use one of those to try and rule out the software side of things

and I suppose the 750GX interposer might be worth a shot, since that one we know should work given it was done commercially back in the day

 

[neumaticgrunt]

I'm looking to put a 7445/7447 into a slot loading G3, do you have either an interposer I could buy or perhaps some gerber files so I can get some made.
I have already put a 7400 into one of my slot loaders and am wanting to see if I could go for something a bit more powerful for the next one.

 

[herd]

I had the Pismo out again and I recorded some test results, all with the CPU at 500MHz and the 1MB L2 cache at 200MHz, with 256MB of CL2 memory:

Pismo CacheBasher Sim gcc -march= CPU L2/memory 750/7450 ----------------------------------------- 750S 790/190 MB/s 19.4k 7410 60x 1590/590 MB/s 21.0/21.2k 7410 MPX 1590/730 MB/s 21.8/22.2k

This shows the G4 can run G3 code faster, with everything else unchanged.

 

[adespoton]

I had the Pismo out again and I recorded some test results, all with the CPU at 500MHz and the 1MB L2 cache at 200MHz, with 256MB of CL2 memory:

Pismo CacheBasher Sim gcc -march= CPU L2/memory 750/7450 ----------------------------------------- 750S 790/190 MB/s 19.4k 7410 60x 1590/590 MB/s 21.0/21.2k 7410 MPX 1590/730 MB/s 21.8/22.2k

This shows the G4 can run G3 code faster, with everything else unchanged.

That's a really interesting benchmark. I wonder what's at the root of the improvement? And can the G4 run G1 and 68040 code faster too?

 

[herd]

The difference is mostly explained by the memory speed. Compared with the 750S G3, the 7410 G4 has 2x the L2 cache and 3x (or more) memory bandwidth. So any software that uses the memory a lot (more than the caches), will see a noticeable improvement just by swapping the CPU chip.

68kmla.org/bb/index.php?threads/ppc750gx-vs-ppc750gl.39043/

68kmla.org/bb/index.php?threads/550mhz-g4-powerbook-pismo-vs-titanium.45851/

 

[indibil]

The G3 that came in the last G3 iMacs that reached 700MHz, if I remember correctly, was the 750CXe, which perhaps improved the internal cache, did they incorporate improvements that could match the results with those of a 7410?

 

[herd]

I think the 750G was the last and best G3 version. I saw similar results comparing a 750G with a 7447, both at 1.1GHz and both with full speed on-die L2 cache. The G4 had 2x faster cache and 3x faster RAM:

68kmla.org/bb/index.php?threads/ppc750gx-vs-ppc750gl.39043/post-541461

 

[indibil]

@herd, looking at the 7410 and 7447 datasheets, the 7410 has A[0-31] and the 7447 has A[0-35], so the 7447 has four additional pads. How can it work on a 7410 socket?

The same goes for other pads like AP or DTI.