Without tweaking q3a in any way, and using the above mention DRI+glide driver, with a 1600x1200 desktop, and the standard size q3a window I get 30.2 fps on q3demo1 on a celeron 300A/450 with the slowest voodoo3 (2000).
Can any non-3dfx card do 3d in a window with linux? Faster?
[sorry apparently HTML formatted doesn't inc tables]
>After all, it's speed advantages are rapidly shrinking Shrinking, I've observed just the opposite, take for instance one of our fluid flow production codes:
PII-450 512K cache g77 -O3 75:02
Celeron 450A-128k g77 -O3 74:44
21164-600 4MB cache g77 -O3 29:27
21264-500 4MB cache g77 -O3 17:16
21264-500 4MB cache fort -O3 8:42
21264-500 4MB cache fort -O4 -arch ev6 -tune ev6 8:31
Almost a factor of 10 at 500 Mhz, well over a factor of 5 for the DS-10 starting at $3k or so. Not everything scales, not everything works well on a beowulf. For scalar codes alphas are a huge speedup when doing heavy FP or memory bandwidth intensive work.
>After all, it's speed advantages are rapidly shrinking Shrinking, I've observed just the opposite, take for instance one of our fluid flow production codes: CPU-clock cachecompiler optionsmin:sec PII-450 512K cacheg77 -O375:02 Celeron 450A-128kg77 -O374:44 21164-600 4MB cacheg77 -O329:27 21264-500 4MB cacheg77 -O317:16 21264-500 4MB cachefort -O38:42 21264-500 4MB cachefort -O4 -arch ev6 -tune ev68:31 Almost a factor of 10 over the 450 Mhz intels sounds good to me. Even the 600 Mhz intels aren't that much faster then the 450's on many large codes because the memory system isn't any faster.
Slashdot Mirror
Without tweaking q3a in any way, and using the above mention DRI+glide driver, with a 1600x1200 desktop, and the standard size q3a window I get 30.2 fps on q3demo1 on a celeron 300A/450 with the slowest voodoo3 (2000).
Can any non-3dfx card do 3d in a window with linux? Faster?
>After all, it's speed advantages are rapidly shrinking Shrinking, I've observed just the opposite, take for instance one of our fluid flow production codes:
- PII-450 512K cache g77 -O3 75:02
- Celeron 450A-128k g77 -O3 74:44
- 21164-600 4MB cache g77 -O3 29:27
- 21264-500 4MB cache g77 -O3 17:16
- 21264-500 4MB cache fort -O3 8:42
- 21264-500 4MB cache fort -O4 -arch ev6 -tune ev6 8:31
Almost a factor of 10 at 500 Mhz, well over a factor of 5 for the DS-10 starting at $3k or so. Not everything scales, not everything works well on a beowulf. For scalar codes alphas are a huge speedup when doing heavy FP or memory bandwidth intensive work.>After all, it's speed advantages are rapidly shrinking Shrinking, I've observed just the opposite, take for instance one of our fluid flow production codes: CPU-clock cachecompiler optionsmin:sec PII-450 512K cacheg77 -O375:02 Celeron 450A-128kg77 -O374:44 21164-600 4MB cacheg77 -O329:27 21264-500 4MB cacheg77 -O317:16 21264-500 4MB cachefort -O38:42 21264-500 4MB cachefort -O4 -arch ev6 -tune ev68:31 Almost a factor of 10 over the 450 Mhz intels sounds good to me. Even the 600 Mhz intels aren't that much faster then the 450's on many large codes because the memory system isn't any faster.