Re: Athlon SMP blues - kernels 2.4.[9 13 15-pre4]

["Stephen C. Tweedie" <sct@redhat.com>]

Hi,

On Wed, Nov 14, 2001 at 03:08:25PM +0000, Matt Bernstein wrote:

> I hope they do; I've just set up a very similar beast (looks like the same
> mobo and same CPUs). Is the RAM "registered" ECC? Are your CPUs the same
> stepping? One problem we were bitten by was the Radeon DRI, so we disabled
> it (in XF86Config-4) and it now seems to at least boot into X.

There are known problems in AMD760+Radeon setups, and a workaround is
to avoid asserting RADEON_SOFT_RESET_HBP during init.  The latest
kernels have that fix in the radeon drm.  Using that in conjunction
with an X server containing the same fix, I've finally got a stable
761+radeon setup here.

I think the X server fix went in on the 4.1.99 branch, but I know that
at least the Red Hat XFree86-4.1 rpms have got the patch back-ported.

> it's not any faster than a dual PIII (1GHz) at the task it's meant to
> perform :( both CPUs report 75% usage, and vmstat 1 doesn't show the IO
> systems being slugged. Very strange. We're wondering if we've hit memory
> bandwidth as the tasks involve some hard sums with big matrices.

If the CPUs were bottlenecked on memory then they would still be
pegged at 100% according to the OS.  They'd just get less work done in
a given interval.

Cheers,
 Stephen