[Openmcl-devel] any experience with ccl+macos+egpu for ml?

[Sean Champ]

On 2019-10-15 04:44, Arthur Cater wrote:
> I’m reading that macs don’t go well with nvidia external gpus. I
> wonder if anybody
> on this list has experience with getting ccl on a middle-aged mac to 
> run neural
> network software using an external gpu.

I believe Nvidia hardware has something of a support for NVIDIA CUDA 
programming. The newer hardware probably also has support for the 
vendor-neutral OpenCL APIs and any OpenCL toolkits.

Although perhaps this may be a bit generic, in regards to platform 
support -- furthermore, assuming that an OpenCL system should be able to 
adapt to the PCIe (??) connection for accessing the GPU on the card in 
the external GPU enclosure -- perhaps it could serve as something of 
use, at least insofar as with regards to parallelization in computing 
tasks? Of course, if OpenCL could work out, perhaps any more complex 
linear mathematical programming support would be available on the GPU, 
too?

Insofar as with regards to usage in a Common Lisp environment, 
personally, though I'm not very well familiar with OpenCL in 
applications, I wonder if perhaps it could represent something like a 
microprogramming system in a generic sense?  Not to compare it to AVR 
programming though, LoL, as I'm sure the dedicated GPU can do a lot more 
in terms of maths.

HTH

-- Sean