Closed Xunius closed 5 years ago
Does it have something to do with mkl? If I create an empty env, install numpy first using conda install numpy
, then install conda install -c conda-forge cdms2
, I got:
The following packages will be UPDATED:
blas pkgs/main::blas-1.0-mkl --> conda-forge::blas-1.1-openblas
ca-certificates pkgs/main::ca-certificates-2019.5.15-0 --> conda-forge::ca-certificates-2019.6.16-hecc5488_0
certifi pkgs/main::certifi-2019.6.16-py37_0 --> conda-forge::certifi-2019.6.16-py36_1
mkl_fft pkgs/main::mkl_fft-1.0.12-py37ha843d7~ --> conda-forge::mkl_fft-1.0.13-py36h516909a_1
mkl_random pkgs/main::mkl_random-1.0.2-py37hd81d~ --> conda-forge::mkl_random-1.0.4-py36hf2d7682_0
The following packages will be SUPERSEDED by a higher-priority channel:
numpy pkgs/main::numpy-1.16.4-py37h7e9f1db_0 --> conda-forge::numpy-1.16.2-py36_blas_openblash1522bff_0
openssl pkgs/main::openssl-1.1.1c-h7b6447c_1 --> conda-forge::openssl-1.1.1c-h516909a_0
pip pkgs/main::pip-19.1.1-py37_0 --> conda-forge::pip-19.1.1-py36_0
setuptools pkgs/main::setuptools-41.0.1-py37_0 --> conda-forge::setuptools-41.0.1-py36_0
wheel pkgs/main::wheel-0.33.4-py37_0 --> conda-forge::wheel-0.33.4-py36_0
The following packages will be DOWNGRADED:
numpy-base 1.16.4-py37hde5b4d6_0 --> 1.14.3-py36h2b20989_0
python 3.7.3-h0371630_0 --> 3.6.8-h0371630_0
@Xunius you can use dask
with cdms (from nightly builds) to use all core. Since you're creating more axes and data with each step I'm not surprised your system gets slower as it goes. @dnadeau4 can you point @Xunius to some dask examples? Thanks.
@xunius the downgraded
package is due from going from the main
channel from anaconda
to conda-forge
channel. conda-forge
is usually not as up-to-date as anaconda
.
I have a test for using Dask
here:
https://github.com/CDAT/cdms/blob/master/tests/test_serialize.py
I have run/demo it in pangeo
in the past.
https://binder.pangeo.io/v2/gh/cdat/dask-cdms/master
the code can be found here: https://github.com/CDAT/dask-cdms
Many thanks for the info.
I think I managed to get it work by pinning numpy
, numpy-base
and blas
in the env, and copying over a few lib*.so files. I'm surprised that cdms2 even imports.
I've never used dask
before, will look into that, thanks for the links @dnadeau4
Hi all, I'm updating from cdat-lite to cdat81, installation is done by:
conda is using v4.6.11
After that I installed nothing else, but jumped into a simple test like this:
The system monitor displays high cpu usage only in 1 thread, at the process gets increasingly slower as iteration goes on.
While in old cdat-lite 6.0rc2, the same script uses all of my cores, and there is no notable slowing down.
Please help. Also, is there a python3 compatible cdat-lite? I'm not sure what's new fancy staff is added into cdat8 but it appears to me only it becoming slower.
Full
conda list
dump: