Project A:
Scaling test of maximally twisted Wilson fermions with tree-level Symanzik improved gauge action (tlSym)


Runs on 24348 lattices

β

3.9
3.9
3.9
3.9
κ

0.1609
0.1609
0.1609
0.1609
μ

0.0040
0.0052
0.0064
0.0075
Run by

Urs Wenger
Urs Wenger
Urs Wenger
Andrea Shindler
 Machine

IBM Regatta in Jülich
IBM Regatta in Jülich
IBM Regatta in Jülich
IBM Regatta ZiB Berlin
<UP> MC history

postscript
postscript
postscript
postscript

Runs on 16332 lattices

β

3.9
3.9
3.9
3.9
κ

0.1609
0.1609
0.1609
0.1609
μ

0.0040
0.0052
0.0064
0.0075
Run by

Mauro Papinutto
N. N
N. N.
Frezzotti/Papinutto
Machine

APEmille Zeuthen
not running yet
not running yet
apeNEXT Rome
<UP> MC history

plaq , correlator
no data yet
no data yet
postscript


Possible setup for runs on 32364 lattices
these are presently suggestions which should be discussed/agreed on

β

3.9
3.9
3.9
3.9
κ

0.1609
0.1609
0.1609
0.1609
μ

0.0040
0.0052
0.0064
0.0075
Run by

Vicent Gimenez Gomez
Roberto Frezzotti
Carsten Urbach
Federico Farchioni
Machine

Mare Nostrum
apeNEXT
DEISA/DECI
Jump Jülich
<UP> MC history
plaq

postscript
postscript


Testing β=4.05; a=0.075(?); tuning to full twist
(Carsten Urbach, Urs Wenger)
Status: just begun


Optimization of apeNEXT tao code
(Philippe Boucaud, Nils Christian, Karl Jansen)
Status: almost optimal static performance
Dirac-Rotunes are optimized, verified and runtime performance checked
Routine

dirac_qphi8
dirac_qphi9
dirac_qphi8_even
dirac_mee_tm
Factor

5.5
6.3
5.4
4.1

Total performance gain, depending on operating system and code
The new linear Algebra routines give another 30% gain
The gauge update part is also substantially (factor 2.4) accelerated
Operating System and Code

os7 + new_code & on crate
os7 + old_code & on crate
os3 + old_code & on crate
Factor

1.0
3.2
4.3


Storing configurations in ILDG format
(Carsten Urbach +N.N.)
Status:
Carsten's page with explanations

Porting of C-code to apeNEXT
(Nils Christian, Andrea Shindler,Carsten Urbach)
Status: work in progress, aim at apeNEXT inversion programme