Re: UDB speed comparison

Greg R. Roelofs (roelofs@nas.nasa.gov)
Wed, 25 Sep 1996 06:45:21 -0700


I wrote:

>> Here's the C90 result (Cray cc, not gcc; not vectorized/parallelized/etc.):
>>
>> time ./pihex > /dev/null
>> 5.7767u 0.0361s 0:05 97%

David Mosberger wrote:

> $ time ./pi >/dev/null
> 4.40user 0.00system 0:04.41elapsed 100%CPU (0avgtext+0avgdata 0maxresident)k
>
> That's for a 300MHz eb164 running a very old kernel and Linux
> distribution. ;-)

Linus wrote:

> [torvalds@ev5 torvalds]$ time a.out > /dev/null
> Command had non-zero exit status 13
> 3.97user 0.00system 0:04.02elapsed 98%CPU (0avgtext+0avgdata 0maxresident)k
> 0inputs+0outputs (71major+18minor)pagefaults 0swaps
>
> (ev5/333, and I don't expect to win this either, but I enjoy beating cray's
> with my home machine ;)

We should all have such home machines. :-)

But seriously, I'm impressed, if not exactly surprised. It would be
interesting to compare the Linux/Alpha performance on some of the highly
optimized hydrodynamics or stellar dynamics codes that are what people
*really* use the C90s for. I suspect the matrix operations would kill
the Alpha, even against a single C90 processor--if I remember right, the
C90 has a vector length of 1024 or so.

But for sequential code...dang. Is Quake ported yet? ;-)

Regards,

--
Greg Roelofs            newt@pobox.com            http://pobox.com/~newt/
Newtware, Info-ZIP, PNG Group, U Chicago, Philips Research, ...