forum@abinit.org

Subject: The ABINIT Users Mailing List ( CLOSED )

List archive

Re: [abinit-forum] v.5.3.x is slower than v.5.2.x ?!

From: Paul Fons <paul-fons@aist.go.jp>
To: forum@abinit.org
Subject: Re: [abinit-forum] v.5.3.x is slower than v.5.2.x ?!
Date: Tue, 17 Apr 2007 13:35:00 +0900

I have posted my results for a 4 cpu mac pro (xeon) machine. I see a slow down in going from 5.2.4 to 5.3.4 but the change is less dramatic. The same (Intel) fortran compiler was used in both cases. Interestingly enough, the speed benchmark shows that 5.3.4 is faster but nonlop is slower. Was something changed in nonlop?

Paul

~/Ports $cat abinit-5.2.4/tests/,summary_speed_070111 abinit/tests/,summary_speed

5.2.4 results

fourwf(pot)

20 : 0.026

30 : 0.029

36 : 0.028

48 : 0.026

64 : 0.026

80 : 0.031

96 : 0.032

nonlop(apply)average= 0.0368571

projbd average= 0.00585714

5.2.4 results

fourwf(pot)

20 : 0.018

30 : 0.024

36 : 0.022

48 : 0.022

64 : 0.024

80 : 0.027

96 : 0.031

nonlop(apply)average= 0.052

projbd average= 0.00571429

Here are the diff_B6 results for

5.2.4

> Total energy(eV)=-1.98692814372417E+02 ; Band energy (Ha)= -2.5177803803E+00

116,123c116,123

< - fourwf(pot) 1.413 48 29.431 1769472 0.017

< - fourwf(den) 0.147 8 18.313 884736 0.021

< - fourdp 1.142 23 49.661 884736 0.056

< - nonlop(apply) 0.163 48 3.392 113238 0.030

< - nonlop(forces) 0.027 8 3.313 113238 0.029

< - nonlop(forstr) 0.022 4 5.625 113238 0.050

< - projbd 0.046 64 0.714 226476 0.003

< - xc:pot/=fourdp 0.936 3 312.033 884736 0.353

---

> - fourwf(pot) 7.262 48 151.286 1769472 0.085

> - fourwf(den) 0.684 8 85.449 884736 0.097

> - fourdp 4.881 23 212.211 884736 0.240

> - nonlop(apply) 0.570 48 11.881 113238 0.105

> - nonlop(forces) 0.105 8 13.184 113238 0.116

> - nonlop(stress) 0.104 4 25.879 113238 0.229

> - projbd 0.223 64 3.479 226476 0.015

> - xc:pot/=fourdp 2.299 3 766.276 884736 0.866

125c125

< +Overall time at end (sec) : cpu= 5.6 wall= 5.6

---

> +Overall time at end (sec) : cpu= 20.7 wall= 20.7

and for 5.3.4

> Total energy(eV)=-1.98692814372417E+02 ; Band energy (Ha)= -2.5177803803E+00

116,123c116,123

< - fourwf(pot) 1.414 48 29.456 1769472 0.017

< - fourwf(den) 0.147 8 18.425 884736 0.021

< - fourdp 1.143 23 49.704 884736 0.056

< - nonlop(apply) 0.160 48 3.325 113238 0.029

< - nonlop(forces) 0.026 8 3.237 113238 0.029

< - nonlop(forstr) 0.022 4 5.525 113238 0.049

< - projbd 0.046 64 0.717 226476 0.003

< - xc:pot/=fourdp 0.940 3 313.500 884736 0.354

---

> - fourwf(pot) 7.262 48 151.286 1769472 0.085

> - fourwf(den) 0.684 8 85.449 884736 0.097

> - fourdp 4.881 23 212.211 884736 0.240

> - nonlop(apply) 0.570 48 11.881 113238 0.105

> - nonlop(forces) 0.105 8 13.184 113238 0.116

> - nonlop(stress) 0.104 4 25.879 113238 0.229

> - projbd 0.223 64 3.479 226476 0.015

> - xc:pot/=fourdp 2.299 3 766.276 884736 0.866

125c125

< +Overall time at end (sec) : cpu= 5.8 wall= 5.8

---

> +Overall time at end (sec) : cpu= 20.7 wall= 20.7

Dr. Paul Fons

Nano-Optics Reseach Team

Team Leader

National Institute for Advanced Industrial Science & Technology

METI

Center for Applied Near-Field Optics Research (CANFOR)

AIST Central 4, Higashi 1-1-1

Tsukuba, Ibaraki JAPAN 305-8568

tel. +81-298-61-5636

fax. +81-298-61-2939

email: paul-fons@aist.go.jp

The following lines are in a Japanese font

〒305-8562 茨城県つくば市つくば中央東 1-1-1

産業技術総合研究所

近接場光応用工学研究センター

近接場光基礎研究チーム　チーム長

ポール・フォンス

On Apr 17, 2007, at 11:00 AM, Masayoshi Mikami wrote:

Dear all,

When I run large jobs with v.5.3.4, I noticed that
it takes much longer time than before (e.g. v.5.2.4).
So, I checked with tests_cpu (on MacOSX/Tiger/gfortran
... keeping the same situation), and noticed something.
(kindly take a look below my signature)
This is just an accident only in my place ?
(if so, sorry for this post...)

(BTW, the reference data seemingly remain the same ... ?)

Cheers,
Masayoshi

for example, diff_B6 in v.5.2.4 (the last part)
-------------------------------------------
116,123c116,123
< - fourwf(pot) 2.120 48 44.167 1769472 0.025
< - fourwf(den) 0.240 8 30.000 884736 0.034
< - fourdp 1.520 23 66.087 884736 0.075
< - nonlop(apply) 0.290 48 6.042 113238 0.053
< - nonlop(forces) 0.000 8 0.000 113238 0.000
< - nonlop(forstr) 0.000 4 0.000 113238 0.000
< - projbd 0.210 64 3.281 226476 0.014
< - xc:pot/=fourdp 1.060 3 353.333 884736 0.399
---
> - fourwf(pot) 7.262 48 151.286 1769472 0.085
> - fourwf(den) 0.684 8 85.449 884736 0.097
> - fourdp 4.881 23 212.211 884736 0.240
> - nonlop(apply) 0.570 48 11.881 113238 0.105
> - nonlop(forces) 0.105 8 13.184 113238 0.116
> - nonlop(stress) 0.104 4 25.879 113238 0.229
> - projbd 0.223 64 3.479 226476 0.015
> - xc:pot/=fourdp 2.299 3 766.276 884736 0.866
125c125
< +Overall time at end (sec) : cpu= 8.0 wall= 12.6
---
> +Overall time at end (sec) : cpu= 20.7 wall= 20.7
-------------------------------------------

On the other hand, diff_B6 in v.5.3.4
-------------------------------------------
116,123c116,123
< - fourwf(pot) 3.232 48 67.333 1769472 0.038
< - fourwf(den) 0.311 8 38.875 884736 0.044
< - fourdp 2.968 23 129.043 884736 0.146
< - nonlop(apply) 0.354 48 7.375 113238 0.065
< - nonlop(forces) 0.047 8 5.875 113238 0.052
< - nonlop(forstr) 0.038 4 9.500 113238 0.084
< - projbd 0.160 64 2.500 226476 0.011
< - xc:pot/=fourdp 2.364 3 788.000 884736 0.891
---
> - fourwf(pot) 7.262 48 151.286 1769472 0.085
> - fourwf(den) 0.684 8 85.449 884736 0.097
> - fourdp 4.881 23 212.211 884736 0.240
> - nonlop(apply) 0.570 48 11.881 113238 0.105
> - nonlop(forces) 0.105 8 13.184 113238 0.116
> - nonlop(stress) 0.104 4 25.879 113238 0.229
> - projbd 0.223 64 3.479 226476 0.015
> - xc:pot/=fourdp 2.299 3 766.276 884736 0.866
125c125
< +Overall time at end (sec) : cpu= 13.7 wall= 13.7
---
> +Overall time at end (sec) : cpu= 20.7 wall= 20.7
-------------------------------------------

v.5.3.x is slower than v.5.2.x ?!, Masayoshi Mikami, 04/17/2007
- Re: [abinit-forum] v.5.3.x is slower than v.5.2.x ?!, Paul Fons, 04/17/2007
- 5.3.4 on mac osX, Matthieu Verstraete, 04/17/2007
  - Re: [abinit-forum] 5.3.4 on mac osX, Yann Pouillon, 04/17/2007
    - Re: [abinit-forum] 5.3.4 on mac osX, Anglade Pierre-Matthieu, 04/18/2007
- Re: [abinit-forum] v.5.3.x is slower than v.5.2.x ?!, Masayoshi Mikami, 04/26/2007
  - Re: [abinit-forum] v.5.3.x is slower than v.5.2.x ?!, Matthieu Verstraete, 04/26/2007
    - Re: [abinit-forum] v.5.3.x is slower than v.5.2.x ?!, Marc Torrent, 04/26/2007
      - Message not available
        
        Re: [abinit-forum] v.5.3.x is slower than v.5.2.x ?!, Masayoshi Mikami, 04/27/2007