Akosf question: A36 vs C37?

user32094
user32094
Joined: 22 Feb 05
Posts: 22
Credit: 74,648
RAC: 0

On AMD XP 2800 Barton C37 is

On AMD XP 2800 Barton C37 is 3% - 5% faster. And compared to the offifial binary it's 130% faster.

AnRM
AnRM
Joined: 9 Feb 05
Posts: 213
Credit: 4,346,941
RAC: 0

RE: On AMD XP 2800 Barton

Message 25833 in response to message 25832

Quote:
On AMD XP 2800 Barton C37 is 3% - 5% faster. And compared to the offifial binary it's 130% faster.

>Thanks for the confirmation as we are seeing the same 3%-5% gain with C37 on our XP 3000+ over version A36. Doesn't do much for the X2's though....Cheers, Rog.

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 561
Credit: 4,527,270
RAC: 0

RE: RE: On AMD XP 2800

Message 25834 in response to message 25833

Quote:
Quote:
On AMD XP 2800 Barton C37 is 3% - 5% faster. And compared to the offifial binary it's 130% faster.

>Thanks for the confirmation as we are seeing the same 3%-5% gain with C37 on our XP 3000+ over version A36. Doesn't do much for the X2's though....Cheers, Rog.


Small difference, but don't forget that C37 was made for 386+ processors.

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1,358
Credit: 2,817,617,686
RAC: 2,835,167

Will the C37 changes be

Will the C37 changes be rolled into the A36 tree using modern instructions?

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 561
Credit: 4,527,270
RAC: 0

RE: Will the C37 changes be

Message 25836 in response to message 25835

Quote:
Will the C37 changes be rolled into the A36 tree using modern instructions?

Only serious alternative is the usage of SSE2.
Probably it would be faster on 128-bit SSE engine processors.

josep
josep
Joined: 9 Mar 05
Posts: 63
Credit: 1,156,542
RAC: 0

Thank you again, Akosf C37

Thank you again, Akosf

C37 is 3%-5% faster over A36 on my Athlon XP 2600+, too.

And it seems there is a better speed improvement on P4 machines, but i need some time to confirm it.

I think is great to have a highly optimized version using 386 code only. It could be the generic Windows application automatically distributed by the project (in case the Einstein team does not achieve as fast applications due to compiler limitations).

And after that, individual users could manually replace it with a more specific version.

And I suppose that the default use of standard 387 math instrucctions (instead of SSE stuff) could help to reduce numerical precission discrepances, such that we have observed for months between Windows and Linux clients. The validator has now a higher tolerance, but ocasionally some of my linux client results are still rejected...

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 561
Credit: 4,527,270
RAC: 0

C37 is really slower on some

C37 is really slower on some intel processors (Pentium-III <5%, Pentium-M <0.3%), the cost of compatibility.
Can somebody tell me a result of AMD K6?

Crunchers For More Power
Crunchers For M...
Joined: 3 Aug 05
Posts: 69
Credit: 1,071,273
RAC: 0

RE: RE: Will the C37

Message 25839 in response to message 25836

Quote:
Quote:
Will the C37 changes be rolled into the A36 tree using modern instructions?
Only serious alternative is the usage of SSE2.
Probably it would be faster on 128-bit SSE engine processors.

Would you be so good to create such a SSE2 version? :)

CU HiNuN

PS: On my XP 2400+ C37 is also slightly faster as A36.

Crunchers For More Power
Crunchers For M...
Joined: 3 Aug 05
Posts: 69
Credit: 1,071,273
RAC: 0

RE: Can somebody tell me a

Message 25840 in response to message 25838

Quote:
Can somebody tell me a result of AMD K6?

Hmm. Only a result for a K7 650 MHz here:

387: 7500 seconds (http://einsteinathome.org/task/21067840)

C37: 6200 seconds (http://einsteinathome.org/task/21364171)

=> 17% percent faster.

CU HiNuN

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 561
Credit: 4,527,270
RAC: 0

RE: Would you be so good to

Message 25841 in response to message 25839

Quote:
Would you be so good to create such a SSE2 version?

I'm working on it, because of fun.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.