Thomas,
Thanks for your link to the issue! There were quite a few discussions
about it. It does help me understand it better.
Thanks much,
Fengguang
On Sat, 14 Apr 2007, Thomas De Schampheleire wrote:
Hi, it is normal that there are significant slowdowns when simulating
with gems. I have posted some results of my own in the following
thread:
https://lists.cs.wisc.edu/archive/gems-users/2007-March/msg00075.shtml
In order to simulate a second (75000000 cycles) on a 4-processor
(75MHz) sarek/serengeti machine running OpenSolaris b57, it took me
about 2 hours. If we assume that we can scale the runtime linearly
with the number of processors, extrapolating that to 16 processors
would take 8 hours.
These tests are with a cpu-switch-time of 1, and ran on a Intel Core
Duo (2x1.66 GHz) with 1GB of DDR2 RAM.
You can speedup simulation quite a bit using a higher cpu-switch-time,
for example 1000 or 10000. This will loose some accuracy in the
timing, but maybe you don't need that.
Thomas
|
![[Computer Systems Lab]](http://www.cs.wisc.edu/pics/csl_logo.gif){kind=logo}