CAMB: accuracy boosts

Use of Cobaya. camb, CLASS, cosmomc, compilers, etc.
Post Reply
Hiranya Peiris
Posts: 54
Joined: September 28 2004
Affiliation: University College London

CAMB: accuracy boosts

Post by Hiranya Peiris » November 30 2004

Hello,
To what should I set the accuracy parameters in CAMB in order to get the same level of accuracy as, for example, the CMBFAST high-precision option? [accuracy_boost, l_accuracy_boost, l_sample_boost etc].
Thanks!
Hiranya

Antony Lewis
Posts: 1936
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

CAMB: accuracy boosts

Post by Antony Lewis » November 30 2004

accuracy_boost=2
l_accuracy_boost=2

should be better than CMBFAST with HP. You may also want to try higher l_sample_boost, but I don't think CMBFAST's HP option decreases the l sampling. The lower-l polarization is quite sensitive to the l sampling, which may explain differences you see with CMBFAST at l < 200.
Last edited by Antony Lewis on April 11 2005, edited 5 times in total.

Hiranya Peiris
Posts: 54
Joined: September 28 2004
Affiliation: University College London

CAMB: accuracy boosts

Post by Hiranya Peiris » November 30 2004

Thanks. I have to say I am amazed at the time difference between the codes. I just ran them on a dual 2.5 GHz Apple G5 with the IBM XLC 95 compiler. CAMB with the accuracy parameters you suggested took 5.992s. CMBFAST *without* the HP option took 19.594s. I did not expect such a huge factor in the speed. I wonder if its because the f77 code is not optimizing as well.

Do you have any ideas about how the two codes benchmark? Is what I found reasonable?

Hiranya

Antony Lewis
Posts: 1936
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

CAMB: accuracy boosts

Post by Antony Lewis » November 30 2004

I've not tried systematically benchmarking them, but what you get doesn't sound too unreasonable. I think CMBFAST slowed down a bit quite recently when they improved its accuracy, when at the same time CAMB sped up due to (amongst other things) using the higher order terms in the tight-coupling expansion. CAMB has always been a lot faster for non-flat models.

CAMB will be significantly slower if you run with lensing, massive neutrinos or including the transfer functions, but probably CMBFAST will be too.

Post Reply