New LLR

[paulunderwood]

I have checked out the new LLR35 (12/27/04). I ran the dynamic linux version on a P4 and on an Athlon, both passing about 700 double-checks against "the original LLR" each at about n=260000 -- no errors

[Jean Penné]

Originally Posted by paulunderwood

I have checked out the new LLR35 (12/27/04). I ran the dynamic linux version on a P4 and on an Athlon, both passing about 700 double-checks against "the original LLR" each at about n=260000 -- no errors

Thank you, Paul, for this QA work ! I think the "Beta" would be soon removed.
Jean

[Thomas11]

I've tested the new LLR35 on a 2GHz Opteron.
To get an idea about how the non-SSE2 functions compare in speed to the SSE2 functions, I played a bit with the "CpuSupportsSSE2" switch.

Here are the timings (while testing 3*2^1572660-1):
SSE2 enabled: 4.186 ms/iteration
SSE2 disabled: 4.275 ms/iteration

You see that the time/iteration using the non-SSE2 library is very close to that of the SSE2 library, only about 2% more. This is a great break-through for the Athlon users and the Athlons do not need to fear comparison with the P4s any longer!

Thank you very much, George and Jean!

-- Thomas

[paulunderwood]

During my tests I saw:

0.4 milli-seconds per iteration on the P4 @ 2.47 GHz

0.7 milli-seconds per iteration on the XP2200 @ 1.8 GHz

making Athlons competative at LLR.

Once we have finshed sieving at the end of 2005, I will be putting my Athlons on LLR.

Athlon users: