Zdenek Kabelac wrote:
> Wrong, make -j spawn as many threads as your computer could handle,
> and if you are compiling something with C++ you will get out of virtual
> memory soon (unless you use hundereds of megs for swap :))
True. Just try a kernel compile `make -j bzImage` and you will probably
run out of memory & error out. I do a `make -j 4` on my BP6.
> But I think the best way to stress cpu is to run two burnP6 programs
> see http://users.ev1.net/~redelm/ for doc and source.
Thank you for the recommendation. I wrote the `cpuburn` suite specifically
to test stability since Intel & AMD won't release their testers (HIPWR30.EXE
and KPOWER.EXE) without NDA.
But it is important to understand what is going on. `burnP6` is designed
to load up the CPU[s], their cooling and power supplies. A stable system
should be able to run one `burnP6` per CPU indefinitely. An unstable
system will normally lockup one or more CPUs, with odd behaviour.
burnP6 will also terminate if it detects an error, but I have no
reports of this happening. Thermal lockup seems to occur first.
But stable CPUs isn't the whole story on stability. The memory subsystem
also has to be stable. `burnP6` really doesn't test this, so I wrote
`burnBX` to stress the SDRAM and its controller (BX chipset, hence the
name, but it runs well on others). `burnBX` is a max-bandwidth memory
tester that tests a user-adjustable size, but lives with whatever pages
Linux/*BSD give it. It also uses particularly error-prone data. I was
very surprised at how eaily I could provoke apparent SDRAM errors. Only
I don't think the SDRAM was at fault. Power supplies (3.3V), CPUs, or even
the Abit design (decoupling caps, and line routing I don't think follows
Intel's recently released FC-PPGA SMP design guide).
In any case, I recommend people first run 2*burnP6 for 1-12 hrs to test
CPU stability, then 4*burnBX for 3-24h to test RAM subsystem stability.
If the system finishes the runs with all burn* running (check the time
counters, burnBX can go zombie), then IMHO it is stable. FWIW, my BP6
is stable at 2 * 5.5 * 97 MHz.
-- Robert author `cpuburn` http://users.ev1.net/~redelm or Metalab.
--
=- To unsubscribe, email [EMAIL PROTECTED] with the -=
=- body of "unsubscribe linux-abit". -=
