Discussion:
[PyCUDA] Need help to get CULA libraries free version
Mayank Jobanputra
2016-09-27 07:22:59 UTC
Permalink
Dear All,
I want to use scikit-cuda and specifically *skcuda.linalg.eig *which
requires CULA Dense library. I tried getting it from the website but there
are some issues with the website itself which is not allowing me to
download the CULA run files for linux.

If anyone of you has that free version of CULA Dense R17 or R18. Please
share the link or what other way I can use
*skcuda.linalg.eig?*

My another concern is Right now I am using CUDA toolkit 7.5. So if I
install any version of CULA will I have to downgrade my CUDA toolkit
version?

Anyhelp will be appriciated.

Thanks and Regards,
Mayank
Lev E Givon
2016-09-27 14:51:47 UTC
Permalink
On Tue, Sep 27, 2016 at 3:22 AM, Mayank Jobanputra
Post by Mayank Jobanputra
Dear All,
I want to use scikit-cuda and specifically skcuda.linalg.eig which requires
CULA Dense library. I tried getting it from the website but there are some
issues with the website itself which is not allowing me to download the CULA
run files for linux.
If anyone of you has that free version of CULA Dense R17 or R18. Please
share the link or what other way I can use skcuda.linalg.eig?
My another concern is Right now I am using CUDA toolkit 7.5. So if I install
any version of CULA will I have to downgrade my CUDA toolkit version?
Anyhelp will be appriciated.
Thanks and Regards,
Mayank
The cusolver library in CUDA 8.0 ostensibly supports eigenvalue
decomposition. If you can download the 8.0 RC through NVIDIA's
accelerated computing developers program, I can modify
skcuda.linalg.eig to use cusolver instead of cula.
--
Lev E. Givon, PhD
http://lebedov.github.io
Hanno Klemm
2016-09-27 16:45:03 UTC
Permalink
Post by Lev E Givon
On Tue, Sep 27, 2016 at 3:22 AM, Mayank Jobanputra
Post by Mayank Jobanputra
Dear All,
I want to use scikit-cuda and specifically skcuda.linalg.eig which requires
CULA Dense library. I tried getting it from the website but there are some
issues with the website itself which is not allowing me to download the CULA
run files for linux.
If anyone of you has that free version of CULA Dense R17 or R18. Please
share the link or what other way I can use skcuda.linalg.eig?
My another concern is Right now I am using CUDA toolkit 7.5. So if I install
any version of CULA will I have to downgrade my CUDA toolkit version?
Anyhelp will be appriciated.
Thanks and Regards,
Mayank
The cusolver library in CUDA 8.0 ostensibly supports eigenvalue
decomposition. If you can download the 8.0 RC through NVIDIA's
accelerated computing developers program, I can modify
skcuda.linalg.eig to use cusolver instead of cula.
Lev,
that would be a great thing to do. Last time, I tried to use CULA, it was almost impossible to get to the homepage, and then they told me that there are no free versions, any longer. It felt a bit like abandon-ware. I gave up trying to use that part to skcuda.

Best regards,
Hanno
Mayank Jobanputra
2016-09-28 20:25:16 UTC
Permalink
Dear Dr. Givon,

I successfully installed CUDA 8.0 as you instructed. Can you please provide
me further guidance and help me on how can I use scikit-cuda and
specifically *skcuda.linalg.eig *using CUDA 8.0?

Thanks and Regards,
Mayank

On Tue, Sep 27, 2016 at 8:49 PM, Mayank Jobanputra <
Thanks Man.
I will download CUDA 8.0 RC and will revert back to you in a
day.
Thank you very much.
Post by Lev E Givon
On Tue, Sep 27, 2016 at 3:22 AM, Mayank Jobanputra
Post by Mayank Jobanputra
Dear All,
I want to use scikit-cuda and specifically skcuda.linalg.eig which
requires
Post by Mayank Jobanputra
CULA Dense library. I tried getting it from the website but there are
some
Post by Mayank Jobanputra
issues with the website itself which is not allowing me to download the
CULA
Post by Mayank Jobanputra
run files for linux.
If anyone of you has that free version of CULA Dense R17 or R18. Please
share the link or what other way I can use skcuda.linalg.eig?
My another concern is Right now I am using CUDA toolkit 7.5. So if I
install
Post by Mayank Jobanputra
any version of CULA will I have to downgrade my CUDA toolkit version?
Anyhelp will be appriciated.
Thanks and Regards,
Mayank
The cusolver library in CUDA 8.0 ostensibly supports eigenvalue
decomposition. If you can download the 8.0 RC through NVIDIA's
accelerated computing developers program, I can modify
skcuda.linalg.eig to use cusolver instead of cula.
--
Lev E. Givon, PhD
http://lebedov.github.io
Loading...