PCRM AlphaSC-667 Serial Performance

Performance Studies using

PCRM

Compaq AlphaServer SC Serial Performance

Index Layout Sensitivity

Date/Person:	April 27, 2000 / P. Worley
Platform:	Compaq AlphaServer SC at Oar Ridge National Laboratory (colt.ccs.ornl.gov):
	16 ES40 4-way SMP nodes (667 MHz Alpha 21264a with 8MB L2 cache)
Environment:	Digital UNIX V5.0; RMS 2.36
Code Version:	2.1.1 of CRM
Common Compiler Flags:	f90 -align dcommons -align records -arch host -tune host -extend_source
Compilation Options:	-O4 -assume accuracy_sensitive
or	-O4 -assume noaccuracy_sensitive
or	-O5 -assume accuracy_sensitive
or	-O5 -assume noaccuracy_sensitive
Library Options:	-mathlibrary accurate
or	-mathlibrary fast
Number of columns:	512

no warm-up steps before timing

Longitudes Maximum Rate Best 2nd Best 3rd Best 4th Best

1 255.417000252589 O4.fast O4.accurate O5.accurate O5.fast

2 265.559090288355 O5.accurate O5.fast O5-accurate.fast O4.accurate

4 283.534069089278 O4.fast O4-accurate.fast O4-accurate.accurate O4.accurate

8 289.21058917744 O5.fast O5.accurate O4.fast O4.accurate

16 294.706197248776 O5.accurate O5.fast O5-accurate.fast O4.accurate

32 294.568837100909 O5.fast O5.accurate O5-accurate.accurate O5-accurate.fast

64 294.414459907995 O5.fast O5.accurate O4.fast O4.accurate

128 279.258741784038 O4.fast O5.accurate O4.accurate O5.fast

256 268.906473779385 O5.accurate O5.fast O5-accurate.fast O4.fast

512 246.212784027271 O5.fast O5.accurate O4.fast O4.accurate

one column computed before timing

Longitudes Maximum Rate Best 2nd Best 3rd Best 4th Best

1 253.102699239087 O5.fast O5.accurate O4.accurate O4.fast

2 268.520873121249 O5.accurate O5.fast O4.accurate O4.fast

4 286.13353254103 O4.accurate O4.fast O4-accurate.fast O4-accurate.accurate

8 291.949389074951 O5.accurate O5.fast O5-accurate.fast O4.accurate

16 297.078531053528 O5.fast O5.accurate O5-accurate.accurate O5-accurate.fast

32 298.024139109932 O5.fast O5.accurate O4.fast O4.accurate

64 298.006573146293 O5.accurate O5.fast O5-accurate.accurate O4.fast

128 287.238922849676 O5.fast O5.accurate O4.fast O4.accurate

256 278.275057515548 O5.fast O5.accurate O4.fast O4.accurate

512 262.716524811639 O5.accurate O5.fast O4.accurate O5-accurate.accurate

full experiment run before timing

Longitudes Maximum Rate Best 2nd Best 3rd Best 4th Best

1 252.735792051987 O5.fast O5.accurate O4.fast O4.accurate

2 265.978195591562 O5.accurate O5.fast O4.fast O4.accurate

4 286.182120337352 O5.accurate O5.fast O4.fast O4.accurate

8 290.490061476587 O5.fast O5.accurate O4.fast O4.accurate

16 296.78208030054 O5.fast O5-accurate.fast O5-accurate.accurate O4.fast

32 297.078531053528 O5.fast O5-accurate.accurate O5-accurate.fast O4.fast

64 299.011149092199 O5.fast O5.accurate O5-accurate.fast O4.accurate

128 285.149146692234 O5.fast O5-accurate.fast O4.accurate O5-accurate.accurate

256 279.042967051162 O5.fast O5.accurate O4.fast O4.accurate

DISCUSSION

As with most compilers, the Digital Fortran compiler has a very large number of optimization options. While we have not experimented with all options, the initial indications are that performance is more sensitive to the longitude dimension declaration than the compiler options examined here. In general,

-O5 -assume accuracy-sensitive -mathlibrary fast
is the best overall performancer, but even

-O4 -assume accuracy-sensitive -mathlibrary accurate
performs well if at least 4 longitudes are available (and no more than 256). Timing with or without "warm-up" calculations has little effect on these results.

Performance Studies using

PCRM

Compaq AlphaServer SC Serial Performance

Index Layout Sensitivity

DISCUSSION

Patrick H. Worley / ( worleyph@ornl.gov) Last Modified Monday, 15-Jul-2002 10:05:58 EDT. 3377 accesses since 1/2/96.

Patrick H. Worley / ( worleyph@ornl.gov)
Last Modified Monday, 15-Jul-2002 10:05:58 EDT.
3377 accesses since 1/2/96.