-
Notifications
You must be signed in to change notification settings - Fork 1
/
benchmarks
49 lines (39 loc) · 1.7 KB
/
benchmarks
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
Benchmarks were obtained using the following processor, compilers and
test cases. Timings are given in mm:ss
Processor: Intel(R) Xeon(R) CPU E5-2670 0 @ 2.60GHz
Gnu compiler: g++ version 4.8.1
Intel comiler: icpc version 13.0.1
PGI compiler: pgCC version13.2-0
# Benchmark 1
src/lda -est -alpha 0.5 -beta 0.1 -ntopics 100 -niters 1000 \
-savestep 100 -twords 20 -dfile models/casestudy/trndocs.dat
# Benchmark 2
src/lda -estc -dir models/casestudy/ -model model-01000 -niters 800 \
-savestep 100 -twords 30
# Benchmark 3
src/lda -inf -dir models/casestudy/ -model model-01800 -niters 30 \
-twords 20 -dfile newdocs.dat
------------ GibbsLDA+_ 0.3 ------------
Compiler opt level time_1 time_2 time_3
-------- --------- ------ ------ ------
g++ default 5:17 4:51 0:02
pgCC default 2:19 1:34 0:01
icpc default 3:37 2:50 0:02
g++ -O3 1:50 1:26 0:01
icpc -O3 1:57 1:34 0:01
pgCC -O3 2:03 1:37 0:01
Identical results are obtained for all compiler and optimization
levels. Note that aggressive (-O3) optimization results in
significantly improved run times.
****** Comparison with earlier versions ******
Version 0.3 of GibbsLDA++ is approximately four times faster than the
previously released version (0.2) and yields identical results.
------------ GibbsLDA+_ 0.2 ------------
Compiler opt level time_1 time_2 time_3
-------- --------- ------ ------ ------
g++ default 12:08 10:04 0:03
pgCC default 8:41 6:46 0:02
icpc default 8:48 7:02 0:03
g++ -O3 9:15 7:31 0:02
icpc -O3 8:46 7:02 0:03
pgCC -O3 8:26 6:45 0:02