MetaBalls demo updated
Wednesday, October 19, 2005 | Permalink
I just recently got myself a new dual-core Athlon64 3800+, so naturally I had to try taking advantage of this extra processing power. Most of my demos are more GPU limited than CPU though, but MetaBalls is an exception to that rule. So I implemented threading into it to improve performance. Everything couldn't be parallelized though, so the gain is fairly moderate, about 15-20%. Another reason for the moderate increase is that the bottleneck apparently shifted from computations to cache/memory. The gain is larger when running the slower FPU path than the 3DNow path, about 25%.
This CPU also supports SSE3, so naturally I threw in an SSE3 path as well. 3DNow is still a tiny bit faster though.
[ 22 comments
| Last comment by Míša (2006-06-14 16:04:22)