Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: which hardware is better

Author: Robert Hyatt

Date: 13:47:05 11/23/98

Go up one level in this thread


On November 23, 1998 at 15:13:14, Howard Exner wrote:

>On November 23, 1998 at 14:05:34, Robert Hyatt wrote:
>
>>On November 23, 1998 at 10:39:54, Tom Kerrigan wrote:
>>
>>>On November 23, 1998 at 06:24:34, Alessandro Damiani wrote:
>>>
>>>>There was a post from Bob Hyatt some days ago. He said the AMD k6 has an error.
>>>>If I remember it right it was wrong branching prediction. I don't know if the
>>>>error has been eliminated in the k6-2.
>>>
>>>If the "error" is branch prediction, that just makes the chip slower, but it'll
>>>stll do what you want it to...
>>>
>>>-Tom
>>
>>
>>No.. this error is a killer.  If it happens, the chip locks up and not even
>>a soft reset will clear it... requires a total power-down.  It is similar to
>>the "foof" bug for intel, but intel found a quick work-around to fix this.
>>AMD hasn't so far.  Been a big topic on the linux-SMP mailing list.
>
>Is this bug you are referring the one from over a year ago?
>
>This is an article from last year:
>
>AMD K6 Sprouts Rare Bug
>
>by Brian McWilliams, PC World NewsRadio
>
>
> September 11, 1997
>Advanced Micro Devices today confirmed that early shipments of its K6 processor
>contained a bug that may cause unreliable system behavior in rare situations.
>
>According to AMD spokesman Dean Whitehair, the bug or erratum only affects K6
>users who are running the Linux operating system on PCs with more than 32MB of
>RAM.
>
>Whitehair says a small number of users have reported encountering system errors
>when trying to compile or update Linux on their K6 machines. The erratum does
>not impact K6 users on Windows 95 or NT, and Whitehair says it's been fixed in
>versions of processor that are shipping now.
>
>"There're always going to be errata in any processor. Go visit Intel's errata
>list, you'll see. And the other thing is we have IBM and Digital using us right
>now; they have to go through some very extensive tests that are a lot more
>rigorous than many review labs. So we're pretty confident that we have a robust
>part."
>
>Linley Gwennap, editor of the Microprocessor Report, says the K6 Linux bug is in
>the same league as Intel's flag erratum that was discovered last May.
>
>"I would tend to say this was on a similar level; potentially maybe slightly
>higher in that nobody really found anyone who was affected by the flag erratum,
>whereas in this case clearly some people are affected. But the number is pretty
>small and I think that the K6 is certainly safe for the vast majority of
>computer users."
>
>This article sums up my recollection of the AMD bug, as I followed it
>on the computer-site newsgroups over a year ago. It was detetected in the early
>B series of AMD K6 chips.
>
>Are Linux users still having problems with the newer K6's (by newer I
>mean anything from last summer to the present. Not the B series)?


Hard to say whether that is "the bug" or not... because it doesn't say a thing
about what it is doing.  The bug I saw, reported again by Alan Cox (the person
that keeps the 2.0 linux kernel distribution process going) has to do with a
branch to a very high address that is a multiple of something like 20000000,
that fails.  The code he posted was 3 lines of assembly long, and crashed every
AMD that it has been run on, to date...

At least that was the story on the linux-SMP mailing list.  Several reported
their machines failed this test, including very recent ones.  I don't have an
AMD here having run into problems last year with one, so I can't test this
myself...

But the sources are certainly reliable if you know Alan and the other "kernel
hackers"...



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.