LLR performance on Core1 / Pentium M

Cruelty · 2007-10-25, 13:56

I have just noticed that running LLR on Core1 / Pentium M hardware is more efficient with CpuSupportsSSE2=0 than with default SSE2 enabled. In my particular example for k*2^n-1, where k<300 and n~600000 time to complete every candidate is 1730s (SSE2) vs. 1580s (no SSE2) - 8,7% improvement.
I wonder if the same rule applies to A64 architecture?

axn · 2007-10-25, 14:28

Perhaps this thread might give you a clue. That testing was done only for the non-base-2 generic FFTs -- not the IBDWT that LLR uses for k*2^n-1 numbers. However, I suspect that the results might follow the same pattern for that also.

Cruelty · 2007-10-25, 15:08

So it works only on Core1 and Pentium M I guess

2007-10-25, 13:56	#1
Cruelty May 2005 2²·11·37 Posts	LLR performance on Core1 / Pentium M I have just noticed that running LLR on Core1 / Pentium M hardware is more efficient with CpuSupportsSSE2=0 than with default SSE2 enabled. In my particular example for k*2^n-1, where k<300 and n~600000 time to complete every candidate is 1730s (SSE2) vs. 1580s (no SSE2) - 8,7% improvement. I wonder if the same rule applies to A64 architecture?

2007-10-25, 15:08	#3
Cruelty May 2005 1628₁₀ Posts	So it works only on Core1 and Pentium M I guess Last fiddled with by Cruelty on 2007-10-25 at 15:08

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Pentium 90 // Pentium ][ 400 years	ValerieVonck	Programming	4	2006-12-12 17:06
Pentium 4 D 2.66Ghz	victor	Hardware	26	2006-05-30 16:30
AMD X2 3800+ vs. Pentium D 920	garo	Hardware	17	2006-05-12 15:56
Pentium M 740 performance?	Turre	Hardware	8	2005-11-13 20:13
Successor to Pentium 4?	jinydu	Hardware	25	2005-08-26 13:42

2007-10-25, 14:28	#2
axn Jun 2003 1010100000000₂ Posts	Perhaps this thread might give you a clue. That testing was done only for the non-base-2 generic FFTs -- not the IBDWT that LLR uses for k*2^n-1 numbers. However, I suspect that the results might follow the same pattern for that also.