some questions about 64bit mprime

[rudi_m]

Hi,
I got 2 questions.

1. Are you working on a 64bit mprime version for Linux?
2. Is there a comparison benchmark between 32bit and 64bit Windows version somewhere? (just to see weather its worth missing the one for Linux or not ;)

[Peter Nelson]

Yes and Yes.

Yes, you can see comparison for windows between 32 and 64 bit in the "Perpetual Benchmark" thread.

TRIAL FACTORING is MUCH faster under 64 bit. LL testing is a bit faster.

And Yes we would like 64 bit linux mprime and we are trying.

Not available yet.

The problem is porting the 64 bit assembly code in COFF format to EFF format.

This is easily done for 32 bit with Binutils Objcopy.

We don't think it is easily achieved for 64 bit yet. When it can be done eg when objcopy method works for 64 bit, George will release a 64 bit mprime.

[rudi_m]

Originally Posted by Peter Nelson

Yes and Yes.

good answer :)

Originally Posted by Peter Nelson

Yes, you can see comparison for windows between 32 and 64 bit in the "Perpetual Benchmark" thread.

wow, the factoring improvements are really incredible!
So maybe for 64bit machines it would make sense to factorize one bit higher than the default settings do to save some LL test!?

To double the factorizing speed means up to 68bit (under x64) will take the same time as up to 67 bit (under x32) or am I wrong?

[Peter Nelson]

Originally Posted by rudi_m

good answer :)

wow, the factoring improvements are really incredible!
So maybe for 64bit machines it would make sense to factorize one bit higher than the default settings do to save some LL test!?

To double the factorizing speed means up to 68bit (under x64) will take the same time as up to 67 bit (under x32) or am I wrong?

You are right on about twice the work for each additional bit depth.

You can test manually to any bit depth you desire. However, I think there are new features in the forthcoming Version 5 server side software which makes it more flexible in allocating work. Consceivably it could tell the client a particular bit depth based on what client OS is being run. Might take a while to roll out but certainly possible.

[ewmayer]

Originally Posted by Peter Nelson

You can test manually to any bit depth you desire.

...Or any bit depth the software permits, whichever is smaller.

[rudi_m]

Originally Posted by Peter Nelson

You can test manually to any bit depth you desire

I tried one time
FactorOverride=69
on a Celeron which was in factoring as fast as a P4 but very very slow in LL.

But unfortunately it didnt seemed to work with GIMPS, like in undoc.txt mentioned:
"You can limit how far the program tries to factor a number. This feature
should not be used with the Primenet server. To prevent the factoring code
from factoring higher than 2^n"

Originally Posted by Peter Nelson

However, I think there are new features in the forthcoming Version 5 server side software which makes it more flexible in allocating work. Consceivably it could tell the client a particular bit depth based on what client OS is being run. Might take a while to roll out but certainly possible.

Yep, would be good if the client could tell GIMPS how far it wants to factor.

[Peter Nelson]

Originally Posted by ewmayer

...Or any bit depth the software permits, whichever is smaller.

.... [since we're being pedantic].... Or any bit depth the software SUPPORTS without giving errors ...... whichever is smaller.

;-)