OpenCL 2.2 Driver Support?

The new Macbook Pro with the Radeon Pro 460 (if not other GPUs) has support for OpenCL 2.2, which will make OpenCL much easier to write, compared to earlier versions. Why isn't there support for OpenCL 2.2 in OS X, and when can we expect to get it?


Not having it is a real issue for me, and feels like Apple is trying to force its own proprietary approach onto its customers -- something sure to drive people away. It's already a problem for my productivity; the "Pro" line shouldn't have issues like this.

Replies

I don't think OpenCL 2.2. support is coming any time soon (if ever). OpenCL 2.0 has been available since 2013, with Apple having hardware available to support it since at least the late 2015 27" iMac 27 yet all Apple computers still top out at v1.2. According to what I've seen in some benchmarks, the OpenCL libs provided by Apple have been left to stagnate, with bugs not getting fixed year after year -- performance for some tests dropped by 45%+ between Yosemite and El Cap; of course this just *can't* have anything to do with pushing their own Metal API. Metal simply isn't a replacement for scientific and HPC, where code is often written and demo'd on local machines (in my case, local Apple machines) and then pushed to clusters of linux servers that couldn't support Metal even if they wanted to.


I love the HW Apple provides (and the overall software experience) but pushing their own APIs at the expense of open-source, cross platform APIs is really starting to become a problem for me and others like me who. Wouldn't be at all surprised to see the (non-app) devs start to follow the 'creatives' to windows over the next year or two...I've started giving the move some consideration.

This is really a bad thing.


Currently, the only API that is up to date at Apple is Metal.


Apple hasn't updated neither OpenCL nor OpenGL for years. Vulkan is not present. CUDA is almost abandonned since 2013 which is when the last Mac had a Nvidia GPU. You cannot build or run anything which has been developped with Compute Capability above v3.5 for CUDA.


How could you just imagine making research on machine learning when you have to deal with obselete APIs and video drivers.


Apple is getting more and more ridiculous speaking about development.

Looks like Intel was not allowed to produce own SDK for MacOS. They obviously need no much effort to make OpenCL SDK build for MacOS too but on their forum they only noted that Intel can not comment on Apple OpenCL SDK and can only suggest to address Apple. So it seems to me that Apple can allow Intel to launch OpenCL SDK for MacOS as Apple still can have own generic OpenCL for all Apple architectures. And this issue then would be closed. So we need to address someone who can make such decissions. Can some one brave from Apple support guys help us to reach the right authority? Thank you in advance.

Got the same problem here. we use openCL to support scientific processing on multiple platforms.


The unavailability of openCL 2.x on OSX means that we have to drop OSX as supported platform. Pity since our scientists love macs.


Apple should not see openCL only as something to support video/photo editing....very nearsighted if you ask me.

Apple really doesn't do anything to help its image, does it?!


I like to believe that the company isn't as pompous and egotistic as people often say but when they act like this its hard to see them as anything but that.


Vendor lock-in helps nobody and trying to corner folks in the scientific-compute and HPC market is not going to wash, they will simply take their business elsewhere.


Apple makes great hardware products that make developer's lives easier just to ***** them over with proprietary APIs.