NVIDIA 470 Series Driver Looks Like It Will Bring OpenCL 3.0 Support

[phoronix]

Phoronix: NVIDIA 470 Series Driver Looks Like It Will Bring OpenCL 3.0 Support

We are already quite eager for NVIDIA's 470 series Linux driver due to Wayland / DMA-BUF improvements coming to this next major feature release for their proprietary driver stack. Making it all the more exciting is it looks like the NVIDIA 470 series driver will have OpenCL 3.0 support...

https://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-470-Series-OpenCL-3.0

[RamblingMadMan]

That's cool and all, but can my dGPU turn off when I'm not using it yet?

[d3coder]

That's cool and all, but can my dGPU turn off when I'm not using it yet?

Yes, you can

[You-]

The 470 series is also supposed to support Wayland and XWayland.

[entropy]

Now that CUDA is dominant due to dubious business practices by Nvidia they now finally offer OpenCL 3 when it's no threat anymore to their strategy.

(No, I don't believe there were real showstoppers due to technical issues)

[d3coder]

The 470 series is also supposed to support Wayland and XWayland.

Wayland is supported for many years already. XWayland will be able to use hardware acceleration with nvidia gpus with 470 driver

[RamblingMadMan]

Yes, you can

My GTX 1050 disagrees.

[d3coder]

My GTX 1050 disagrees.

My 1650 Ti agrees

[RamblingMadMan]

My 1650 Ti agrees

Well that's great for you, but for everybody running a Pascal or earlier GPU it's a different story. You would think that an earlier generation of card would have better support than a new card, not worse; but I guess I shouldn't expect any better from NVIDIA's linux offering.

[bug77]

Now that CUDA is dominant due to dubious business practices by Nvidia they now finally offer OpenCL 3 when it's no threat anymore to their strategy.

(No, I don't believe there were real showstoppers due to technical issues)

There's no strategy at work here, just a monumental failure from Khronos: they made SVM mandatory for OpenCL 2.x compliance, but SVM is only useful for running OpenCL on APUs. Nvidia doesn't make APUs, so they didn't implement it (they actually did, but not fully).

However, OpenCL didn't need that fiasco, it is fully capable of failing on its own. Be it 1.x, 2.x or 3.x, people just don't build frameworks on top of that. I don't know specifics, it's just supposedly hard/unpleasant to work with.