Bug in IBM's 1Ghz PowerPC 750GX?

[Plutonique9]

I've been following the 750GX ever since it was announced by IBM, hopefully I can pool together some $ to be able to upgrade my Powerbook G3 2000 with one from Powerlogix.
However, I'm a little daunted by this Inquirer article I just came across which says that the 1Ghz PowerPC 750GX doesn't support "all" instructions at the rated speed, and must be downclocked to 933mhz as a work-around. Now, I don't think this is a problem for the Powerlogix Pismo upgrades, as they are now being shipped with 900mhz 750GX's. But other upgrades utiliizing this CPU by them may very well be a problem. Any techies out there that can comment on this?

The Inquirer article;
http://www.theinquirer.net/?article=17680

IBM's tech document which supposedley addresses this problem;
http://www-306.ibm.com/chips/techlib/techlib.nsf/techdocs/FAB66613F3E43FBC87256E47007ADBBE/$file/750gx_errata7-15-04.pdf

 

[wrldwzrd89]

I've been following the 750GX ever since it was announced by IBM, hopefully I can pool together some $ to be able to upgrade my Powerbook G3 2000 with one from Powerlogix.
However, I'm a little daunted by this Inquirer article I just came across which says that the 1Ghz PowerPC 750GX doesn't support "all" instructions at the rated speed, and must be downclocked to 933mhz as a work-around. Now, I don't think this is a problem for the Powerlogix Pismo upgrades, as they are now being shipped with 900mhz 750GX's. But other upgrades utiliizing this CPU by them may very well be a problem. Any techies out there that can comment on this?

The Inquirer article;
http://www.theinquirer.net/?article=17680

IBM's tech document which supposedley addresses this problem;
http://www-306.ibm.com/chips/techlib/techlib.nsf/techdocs/FAB66613F3E43FBC87256E47007ADBBE/$file/750gx_errata7-15-04.pdf

Remember the problems IBM had with the PPC 970fx and that they were eventually solved? I'd think IBM could run some diagnostics at their testing facility and get this bug fixed too. IMO, it's only a matter of time before this gets resolved.

 

[Plutonique9]

But is this something that could be fixed with a firmware flash, or is it embedded in the actual hardware. More specifically, would users who already have upgrade there Mac's with "current" 1gz+ 750GX's be able to resolve this bug, or would they need to have the CPU physically swapped with an updated version of the 750GX?

p9

 

[wrldwzrd89]

But is this something that could be fixed with a firmware flash, or is it embedded in the actual hardware. More specifically, would users who already have upgrade there Mac's with "current" 1gz+ 750GX's be able to resolve this bug, or would they need to have the CPU physically swapped with an updated version of the 750GX?

p9

I have a strong suspicion that the cause is embedded in the chip hardware, and thus would require a CPU swap to correct. My reasoning is twofold: first of all, CPUs don't have any firmware; secondly, there isn't any other logical place for a bug affecting speed of instruction execution (since that's what the CPU handles) besides the hardware.

 

[Plutonique9]

I scrolled down to page 17 & 18 of IBM's errata and it seems that pretty much ALL of the 750GX's are affected by this bug. Which makes me think that I better hold off on upgrading my Pismo until IBM resolves this issue. The problem is that every 750GX must be downclocked 100mhz, so if your talking about a 900mhz 750GX (downclocked to 800mhz), your losing 12% of your Core speed.

Now, I don't understand the technical nature of this bug but IBM states that it's an issue when these instructions are used at maximum speed;

"The specific instructions that are impacted at the rated maximum frequency are: 1) mcrf, 2) crand, 3) crandc, 4) creqv, 5) crnand, 6) crnor, 7) cror, 8) crorc, 9) crxor"

Scrolling further down the document IBM states a future resolve;

"A test screen has been established for DD1.1 to provide improved frequency operation. The screened parts are designated by a new Test Conditions code in the OEM part number. Refer to Figure 1-1 in the IBM PowerPC 750GX RISC Microprocessor Datasheet for the part number legend. Table2 lists the correlation between original and new Test Conditions codes. Please contact your IBM Sales Representative for specific part numbers. A fix is planned for the next revision"

p9

 

[wrldwzrd89]

I scrolled down to page 17 & 18 of IBM's errata and it seems that pretty much ALL of the 750GX's are affected by this bug. Which makes me think that I better hold off on upgrading my Pismo until IBM resolves this issue. The problem is that every 750GX must be downclocked 100mhz, so if your talking about a 900mhz 750GX (downclocked to 800mhz), your losing 12% of your Core speed.

Now, I don't understand the technical nature of this bug but IBM states that it's an issue when these instructions are used at maximum speed;

"The specific instructions that are impacted at the rated maximum frequency are: 1) mcrf, 2) crand, 3) crandc, 4) creqv, 5) crnand, 6) crnor, 7) cror, 8) crorc, 9) crxor"

Don't know what they are, maybe someone else knows?

p9

The fact that all the affected instructions contain the string "cr" is an important clue. To me, it suggests that the bug is in the logic unit (or whatever execution unit in the PPC 750GX handles logic instructions).

 

[iMeowbot]

The fact that all the affected instructions contain the string "cr" is an important clue. To me, it suggests that the bug is in the logic unit (or whatever execution unit in the PPC 750GX handles logic instructions).

Systen Unit in this case, but whatever. Apparently the problem is with the condition register itself.

 

[wrldwzrd89]

Systen Unit in this case, but whatever. Apparently the problem is with the condition register itself.

Aha! cr = condition register - hence the affected instructions, which are handled by the condition register (that's why they contain 'cr'), so they're affected by the bug. Makes perfect sense now, and it's a hardware problem without a doubt now. Fixing it will require a CPU swap.

 

[Plutonique9]

What sort of impact would this have then on system stability? It seems that it is possible to create a work around in code, but it seems 99.9% unlikely that you'd see those change implemented in mac software, seeing as this CPU hasn't been used officially in any Apple computers, only Powerlogix upgrades. I can't imagine how many 750GX upgrades have been shipped so far with this bug, can't be good for there already scuffed reputation.

 

[blackfox]

The G3 upgrade aside...if you are looking to upgrade your Pismo, I believe the G4 550 is still a better bet for long-term usability, as os x and many apps are optimized for the G4. I guess it depends on what one might be doing on the PB...still, you might expect the G3 900/1000 to be 80% faster than the G4 500 (because of clock speed), but test have shown that it isn't...also reviews on xlr8 have been hit-and-miss for the G3 upgrade (partially for reasons noted in thread) while the G4 550 has received glowing reviews.

On the other hand, I believe that G4 upgrade may be the last one available due to differences in chip/pin design in later G4 incarnations (I believe the 550 is actually an "overclocked" 500 (it does run a little hot, I have heard)...It is possible that the G3 may be continued to be ramped up, if their is a market for ant future development.

I have the old G4 500 upgrade from Powerlogix, and coupled with a HD replacement (7200rpm) and max RAM, my PB is a solid performer. If a G3 was offered at around 1.2 Ghz or more, then I might be interested...

FWIW

 

[Chryx]

My reasoning is twofold: first of all, CPUs don't have any firmware

uhh.. yes they do.

Microcode has been alterable aftermarket.. on x86 processors at least, since the Pentium Pro... I'd be AMAZED if PowerPC chips didn't have something remotely comparable by now.

 

[wrldwzrd89]

uhh.. yes they do.

Microcode has been alterable aftermarket.. on x86 processors at least, since the Pentium Pro... I'd be AMAZED if PowerPC chips didn't have something remotely comparable by now.

I think PowerPCs don't. I did a search on Google, and all the results on the first page are about Open Firmware (which isn't on-chip) or emulation (which I'm not interested in).