Indeed. But what about your input dataset? How big is your scene data?
I find with volume rendering the memory bus starts to saturate,
because each processor has to hit the same volume data (needless to
say, the volume data doesn't fit into cache).
When we got our SGI Power Challenge XL in 1995 (10x 75 MHz r8000), I
saw ~8x speedup (77% parallel efficiency). With 4 CPUs I got 3.83x
speedup (96% efficiency).
A couple of years later I benchmarked a Sun E6000 with 10 CPUs. Again
the same 8x speedup. It looks like the CPU/bandwidth ratios didn't
change much.
This is why I got excited at the launch of the Origin 2000 when I saw
distributed memory and heard mention of automatic page migration, but
still with an SMP model (or made to look that way). Haven't organised
a benchmark yet, though :-(
Regards,
Richard....
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/