|
|
Ublas : |
Subject: Re: [ublas] cublas bindings
From: OvermindDL1 (overminddl1_at_[hidden])
Date: 2010-07-26 15:37:00
On Mon, Jul 26, 2010 at 4:21 AM, Thomas Klimpel
<Thomas.Klimpel_at_[hidden]> wrote:
> OvermindDL1 wrote:
>> I just have a quite question of curiosity. Â I was under the impression
>> that cuda was an nvidia-only construct that no one else supported, and
>> it only supports programming the gpu, whereas opencl allows you to do
>> the same things, but on the gpu, cpu, some dedicated vector board,
>> etc... Â Why would someone use cuda over opencl, or am I vastly
>> mistaken in what I had the impression of?
>
> Some points that might help understanding:
> - cuda existed long before opencl
> - even so opencl is the future and well supported by nvidia, nvidia won't drop support for cuda any time soon, and still continues to improve it
> - the porting from cuda to opencl is quite easy (when similar functionality is available in opencl) compared to porting cpu only code to gpu
>
> Of course I agree with Karl Rupp that it makes sense to also look at his OpenCL library
>
> http://viennacl.sourceforge.net/
>
> that might at some time in the future be able to offer sufficient functionality to substitute cublas. (I know that it also offers functionality not available in cublas, but that's not relevant as we are talking about cuda and cublas right now.)
>
> (I'm no expert for GPU computing, but I decided that a half baked answer is better than no answer at all.)
I thought so, and considering that OpenCL is 'the future', why
continue down the Cuda path? My main reasoning for this is that this
project is unusable for me as I have pure AMD hardware, being a 6-core
CPU and a high-end AMD VideoCard, and OpenCL can abstract not only GPU
interfaces, but CPU as well, I have been curious in testing this.