Difference between revisions of "Performance"
From Biowerkzeug Wiki
(→Scaling) |
(→Scaling) |
||
| Line 126: | Line 126: | ||
All tests were run with Hippo beta rev35 on empty systems. See remarks on the [[Talk:Performance#Scaling|test methodology]]. | All tests were run with Hippo beta rev35 on empty systems. See remarks on the [[Talk:Performance#Scaling|test methodology]]. | ||
| + | In the table below, '''# cpus''' really means how many independent cores are available. | ||
{| class="wikitable sortable" | {| class="wikitable sortable" | ||
| + | |+ Scaling of the Hippo WALP test case using OpenMP on single-board machines | ||
! vendor | ! vendor | ||
| − | ! # cpus | + | ! # cpus |
! walltime and scaling | ! walltime and scaling | ||
|- | |- | ||
Revision as of 10:23, 8 December 2008
Single processor performance
As a basic performance test we run the simulations in the test directory with these commands (on Linux):
cd testjobs time ./calc_testjobs
- We report the user time.
- CPU properties are found with
cat /proc/cpuinfo
and model names from List of Intelmicroprocessors for Intel (also see the detailed lists of Intel x86 CPUs) and List of AMD microprocessors for AMD cpus (also see the detailed lists of AMD CPUs). - Also report the binary used (hippo or hippo_p3) and the revision.
- Note that these tests only utilize a single core.
- These results only give a rough idea of the relative performance of different architectures.
Please add your own results.
| vendor | model | GHz | cores | time/min | binary | revision |
|---|---|---|---|---|---|---|
| Intel | Core 2 Quad Q9550 | 2.83 | 4 | 1:04 | hippo | rev32 |
| Intel | Core 2 Quad Q9550 | 2.83 | 4 | 1:06 | hippo_p3 | rev32 |
| Intel | Quad Core Xeon E5420 | 2.5 | 4 | 1:16 | hippo | rev35 |
| Intel | Quad Core Xeon E5420 | 2.5 | 4 | 1:16 | hippo | rev32 |
| Intel | Quad Core Xeon L5410 | 2.33 | 4 | 1:27 | hippo | rev32 |
| Intel | Quad Core Xeon L5410 | 2.33 | 4 | 1:31 | hippo_p3 | rev32 |
| AMD | Phenom X4 9850 | 2.5 | 4 | 1:36 | hippo_p3 | rev32 |
| AMD | Opteron 248 | 2.2 | 1 | 2:18 | hippo_p3 | rev32 |
| Intel | Core Duo T2300 (Mac Mini) | 1.66 | 2 | 2:41 | hippo_p3 | rev32 |
| Intel | Xeon 2.4 | 2.4 | 1 | 3:03 | hippo_p3 | rev26 |
| AMD | Athlon XP 1800+ | 1.53 | 1 | 4:30 | hippo_p3 | rev26 |
(You don't have to link to the CPU wikipedia page; just put down whatever you got and even if you're unsure. If in doubt just put the output from
cat /proc/cpuinfo | \
awk 'BEGIN {FS=":"}; \
/vendor_id/ {vendor=$2}; \
/model name/ {model=$2}; \
/cpu MHz/ {GHz=$2/1000}; \
/siblings/ {cores=$2}; \
END {printf("|-\n| %s\n| %s\n| %.1f\n| %d\n| TIME\n| ?\n| ?\n", vendor, model, GHz, cores)}'
into the wiki table and replace TIME by what you measured.
Scaling
All tests were run with Hippo beta rev35 on empty systems. See remarks on the test methodology.
In the table below, # cpus really means how many independent cores are available.
| vendor | # cpus | walltime and scaling |
|---|---|---|
| Intel | 4 | |
| Intel | 8 | |
| AMD | 4 |
