Re: On Overclocking - READ THIS!

andrew meggs (insect@antennahead.com)
Tue, 6 May 1997 11:05:49 -0400


While developing the new MacOS client in assembler, I found that one
PowerMac which had been significantly overclocked for ~3 years without any
hint of a problem did in fact have a problem -- after it had been running
for more than a few hours it failed to detect the key when it processed a
block containing it. This was an especially bad problem, because it
appeared to process all nonmatching keys correctly. I rearranged the
instruction sequence that it was messing up, but I see no reason why a
different overclocked PowerPC might not have a problem on a different part
of the code, or why a piece of silicon from Intel or AMD would be immune to
similar problems when run beyond its rated speed.

If such a faulty client were to be run in the general pool instead of a
controlled test situation, and it were to fail to find the correct key and
report to the server that no, the key wasn't in that block, then all the
other clients would merrily continue searching elsewhere and the entire
project would be for nothing. To put it as bluntly as possible, IF YOU HAVE
AN OVERCLOCKED OR OTHERWISE UNSTABLE SYSTEM, THE DESCHALL PROJECT DOES NOT
NEED THE "HELP" YOU MIGHT PROVIDE.

That's not the official line from Rocke, but I imagine he finds the idea of
a client that runs correctly except on the one iteration containing the key
to be quite abhorrent -- isn't that exactly what people are talking about
as the most damaging way to undetectably sabotage the project?

____________________________________________________________________________
Andrew Meggs, content provider Antennahead Industries, Inc.
<mailto:insect@antennahead.com> <http://www.antennahead.com>