[Wien] Question about accuracy.

[Jorissen Kevin]

Hi Sorin,
I checked the goto p4 for a few small cases and got identical energies (compared to mkl and to lapw lapack).  I can't say about the opteron.
Someone mentioned the same problems on the ML a few weeks ago (or was that you? ;-)), maybe you can contact that person.
 
Kevin. 
 

	-----Oorspronkelijk bericht----- 
	Van: Sorin Lazar [mailto:S.Lazar at tnw.tudelft.nl] 
	Verzonden: vr 1/30/2004 4:45 
	Aan: wien at zeus.theochem.tuwien.ac.at 
	CC: 
	Onderwerp: [Wien] Question about accuracy.
	
	

	Dear all,
	
	Few months ago I have started to compare the performance of Wien2k code
	using various libraries and compilers on various platforms.
	Using the new goto libraries (P4 and OPTERON) I found some differences in
	the results of the calculation in  comparison with an executable compiled
	with normal atlas_blas library.
	For a large case (64 atoms) the difference (see bellow) is around 0.7Ry in
	total energy and decreases at 0.05Ry for a small case (4 atoms) .
	
	My question is if there is any reason to be worried about that?
	
	Thank you,
	Sorin
	
	For a big case: (64 atoms)
	
	classical atlas_blas library on various systems (Athlon, Opteron, Alpha )
	
	:ENE : ********** TOTAL ENERGY IN Ry = -127907.383438
	:FER : F E R M I - ENERGY(TETRAH.M.)= 0.45402
	
	
	new GOTO lib (P4 and Opteron) on Opteron:
	
	:ENE : ********** TOTAL ENERGY IN Ry = -127906.677866
	FER : F E R M I - ENERGY(TETRAH.M.)= 0.45715
	
	
	
	For a small case: (4 atoms)
	
	classical atlas_blas library on various systems (Athlon and Alpha)
	
	:ENE : ********** TOTAL ENERGY IN Ry = -7995.656398
	:FER : F E R M I - ENERGY(TETRAH.M.)= 0.48117
	
	new GOTO lib (P4 and Opteron)
	
	:ENE : ********** TOTAL ENERGY IN Ry = -7995.611899
	:FER : F E R M I - ENERGY(TETRAH.M.)= 0.48294
	
	
	
	************************************************************
	Ph.D. Sorin Lazar
	National Center for HREM
	Department of Materials Science and Technology
	DELFT University of Technology
	
	Rotterdamseweg 137
	2628 AL Delft
	the Netherlands
	
	phone: +31 (0)15 2788397
	e-mail: sorinl at hrem.stm.tudelft.nl
	         S.Lazar at tnw.tudelft.nl
	
	

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/ms-tnef
Size: 5234 bytes
Desc: not available
Url : http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20040202/befded6e/attachment.bin