PDA

View Full Version : 34 Errata in Core Duo Chips


MacBytes
Jan 23, 2006, 03:58 PM
http://www.macbytes.com/images/bytessig.gif (http://www.macbytes.com)

Category: 3rd Party Hardware
Link: 34 Errata in Core Duo Chips (http://www.macbytes.com/link.php?sid=20060123165815)
Description:: Errata left chips are common. The Core Duo has 34, some of which are serious enough to need software workarounds, or problems could theoretically be serious (crashes, failure to wake from sleep).

Posted on MacBytes.com (http://www.macbytes.com)
Approved by arn

SiliconAddict
Jan 23, 2006, 04:42 PM
Hmmm up til now most of Intel's mobile chips have been good. Rush job? :(

I remember when Intel replaced my Pentium 90Mhz chips because it was one of the chips that had a serious bug with FPCs. I never had any problems but hey. It was a new CPU. Who am I to complain.

There was one interesting post on the thread.


E1 - generates a GPF when bad code executes. That ain't a show stopper. It may not the exact behavior Intel intended, but it sure is OK as a fallback.

AE2 - only intentionally bad code can produce the 'overflow/wrap to zero' issue. No real code EVER exploits such wrapping. Further the error isn't an error per se: the code continues to run.

AE3 - is an alert

AE4 - would be a 'run-away' string copy. Bad code again, in any case.

AE5 - only of concern to O/S authoring teams: they need to keep their A and D bits organized. The workarounds are trivial, and again contained in the O/S core.

AE6 - if bad X86 (we're talking pre 286 code) is executed and double-faults, it might not resume. So? All old x86 code should be 6-feet under by now, no?

AE...9 - as you say "any programmer coding this..." and I agree. Intentionally obscure 'near-buggy' coding is hardly expected to have clean cut-aways.

AE...12 - last I heard, no business software uses 'exact computation' flags to begin with. Some specialized math does, and especially some 'general purpose' equation fitters. Yet, the "probem" is a minimum of 16 decimal-places down. Combined with auto-extended 80-bit floating point, I don't think this is more than a "lab curiosity" rather than a "Show stopper".

AE18 - this should be fixed - because the Board Makers depend on it to set fan RPMs.

... anyway ...

I just don't think any of these are going to affect "normal code", even at the very early stage. Intel has bigger fish to fry, just like AMD.

Fiveos22
Jan 23, 2006, 05:47 PM
Hmmm up til now most of Intel's mobile chips have been good. Rush job? :(


Rush job? Maybe. But there are known errors for every chip. I'd like to see the AMD, and even better the IBM chip error sheets for comparison.

nagromme
Jan 23, 2006, 06:18 PM
The raw number of errors is interesting. The real-world EFFECTS of those errors is much more important to me. Which I suspect will be close to zero, as with other chips and their errata.

I certainly wouldn't opt for a slower G4 or Pentium just because of this.

blufire
Jan 23, 2006, 06:23 PM
Rush job? Maybe. But there are known errors for every chip. I'd like to see the AMD, and even better the IBM chip error sheets for comparison.

I checked IBM semiconductor when I read this article, and the PPC970FX has 24 errata in one report: http://www-306.ibm.com/chips/techlib/techlib.nsf/techdocs/79B6E24422AA101287256E93006C957E

bousozoku
Jan 23, 2006, 06:26 PM
Hmmm up til now most of Intel's mobile chips have been good. Rush job? :(

I remember when Intel replaced my Pentium 90Mhz chips because it was one of the chips that had a serious bug with FPCs. I never had any problems but hey. It was a new CPU. Who am I to complain.

There was one interesting post on the thread.

Those weren't the first or the last big errors on Intel x86 processors, though most have been math problems.

They might as well make their slogan "Fast performance first, accuracy, if possible."

nagromme
Jan 23, 2006, 07:01 PM
Note that one assertion of the article is that 34 may be just the tip of the iceberg: supposedly in the past it has taken much longer to find that many errata. Finding them so soon is supposed to be a bad sign.

Now, if that's not just the author's bias, the discovery of errata may be expected to continue to much higher numbers.

If it IS just the author's bias, then maybe the errata have been caught earlier than normal for some other reason, not just because of sheer quantity.

mrzippy
Jan 27, 2006, 10:49 AM
Surely bug AE18 is very serious, as if the temp readout stops so could the fans then the processor frys.

Looking at the comments on geek.com many of these bugs are carry overs from the older pentiums, so they are not all 'new'