The State Of OpenCL Computing Language (2016)

[phoronix]

Phoronix: The State Of OpenCL Computing Language (2016)

Neil Trevett, President of The Khronos Group, presented at the International Workshop on OpenCL (IWOCL) last month in Vienna about the state of the union for OpenCL...

http://www.phoronix.com/scan.php?pag...of-OpenCL-2016

[Marc Driftmeyer]

Can someone explain to me how I should care what Neil Trevett says about OpenCL, when CUDA is the singular focus of Nvidia? Isn't it about time Nvidia steps down as the driving element at Khronos?

[wizard69]

Can someone explain to me how I should care what Neil Trevett says about OpenCL, when CUDA is the singular focus of Nvidia? Isn't it about time Nvidia steps down as the driving element at Khronos?

Who gives a crap about NVidia? Seriously Go NVidia you go proprietary which is stupid right from the start. Then you need to realize that the market for NVidias chips is dying. These days the average user can get along fine with the embedded GPU's in AMD and Intel APUs. Not to mention a huge part of the market these days has zero in the way of an NVIdia install base, mobile isn't NVidia and likely never will be.

The simple reality is that if you hook up with NVidia and its proprietary systems you will loose a massive amount of potential market for your software. The landscape is changing very rapidly and it isn't changing in favor of NVidia. Even vendors in the high end computing markets are starting to wise up and realize being tied to NVidia is like having a ball and chain around ones neck.

[liam]

Who gives a crap about NVidia? Seriously Go NVidia you go proprietary which is stupid right from the start. Then you need to realize that the market for NVidias chips is dying. These days the average user can get along fine with the embedded GPU's in AMD and Intel APUs. Not to mention a huge part of the market these days has zero in the way of an NVIdia install base, mobile isn't NVidia and likely never will be.

The simple reality is that if you hook up with NVidia and its proprietary systems you will loose a massive amount of potential market for your software. The landscape is changing very rapidly and it isn't changing in favor of NVidia. Even vendors in the high end computing markets are starting to wise up and realize being tied to NVidia is like having a ball and chain around ones neck.

Nvidia is, I THINK, making most of their money in the HPC field.

[boffo]

Cuda makes sense in the supercomputer market, because those usually use custom software.
In the consumer market there are many libraries that use Cuda or OpenCL and they don't require you to know how to code Cuda/OpenCL, those libraries are the future for most of the needs.
With OpenCL 2.2, devs will be able to encapsulate OpenCL in c++ classes->better libraries->easier to code.

[coder]

Who gives a crap about NVidia? Seriously Go NVidia you go proprietary which is stupid right from the start. Then you need to realize that the market for NVidias chips is dying.

Yes, it's stupid. But, they do seem to have an edge in PC gaming and they basically own HPC. I can't really understand why they're so dominant in HPC, but I guess a lot of those folks know CUDA and aren't too bothered about being locked into NVidia.

I wonder if AMD's big sell-out to China might turn out to be what saves OpenCL. I'll bet a lot of Chinese HPC installations will be AMD, from now on. Also, I think the Chinese are more wary of getting locked into a particular (especially US-based) vendor. Or, at least one in which they don't have a big stake.

Even vendors in the high end computing markets are starting to wise up and realize being tied to NVidia is like having a ball and chain around ones neck.

Really? What's your evidence? I recently looked at GPU-support among deep learning frameworks, and found no first-class support for OpenCL. They're all CUDA. The only exceptions were one fork by an AMD research lab and one partially-implemented OpenCL backend that it's not clear is still a going concern.