OpenCL 3.0.12 Published With Command Buffers Mutable Dispatch Extension

[phoronix]

Phoronix: OpenCL 3.0.12 Published With Command Buffers Mutable Dispatch Extension

The Khronos Group has published OpenCL 3.0.12 as the newest version of this API for compute across heterogeneous platforms, but mostly known for GPU compute...

https://www.phoronix.com/news/OpenCL-3.0.12-Released

[MorrisS.]

Every article introducing progresses should show in technical way the percentage of advantage for end-user applying the necessary technology able to take benefit of it. How those improvements are defined in practical terms? What technology is necessary to take benefit of it?

[MauganRa]

Shouldn't it be like "Tampere University"?

[coder]

Every article introducing progresses should show in technical way the percentage of advantage for end-user applying the necessary technology able to take benefit of it. How those improvements are defined in practical terms? What technology is necessary to take benefit of it?

That's a nice idea, but a completely unrealistic expectation for a site like this. Michael already did us a service by not only telling us about the new feature, but explaining it in a way that's easily understandable by OpenCL users.

What you're talking about is something you might find on a blog dedicated to OpenCL or GPU computing. And that's simply because the only way someone would be able to make such a projection is by knowing precisely how much time is spent copying the command buffers, in a certain application, rather than modifying them in-place. Moreover, it's going to be very application-specific, where some apps might benefit by a few % (just for the sake of argument--I have no idea if that's accurate) while others wouldn't benefit at all.

Now, what might happen on Phoronix is that once some OpenCL runtimes are updated to support it, and a few apps utilize the new extension, then Micheal could benchmark those with the feature on/off. This is what we saw with Resizable BAR, however it's in no way close to the kind of win that feature can provide, nor is it anywhere near as popular. So, I don't expect to see benchmarks or an article devoted to it.