Hi Ivan ...

> Then i found that DeviceInfo.py shows that platform.version is "OpenCL 
> 1.1 CUDA 6.5.20", when the same script on AMD card shows "OpenCL 1.2"

Nvidia refuses (for now) to enable OpenCL 1.2 support. Those features
are mainly related to segmenting a GPU for virtualization purposes. 

They are neither supporting OpenCL 2.0 which contains feature like
dynamic parallelism and unified memory architecture (which they support
on Cuda)

> Does this version difference affect the behaviour of K520 ?

it shouldn't
> According to GRID K520 parameters, it is 2 GPUs in one product. Does 
> OpenCL 1.1 see only 1 GPU on K520?

It should see 2 devices.
 
> How can i load K520 to its full power?

You need 2 contexts, one on each of those devices and balance the work 
"yourself".

Cheers,

-- 
Jérôme Kieffer
tel +33 476 882 445

_______________________________________________
PyOpenCL mailing list
[email protected]
http://lists.tiker.net/listinfo/pyopencl

Reply via email to