WebCL – Heterogeneous parallel computing in HTML5 web browsers | Hacker News Reader

WebCL – Heterogeneous parallel computing in HTML5 web browsers

khronos.org

72 pointsmatt4228 comments

Relevant Bugzilla bug for Gecko: https://bugzilla.mozilla.org/show_bug.cgi?id=664147#c30

An interesting spec, certainly, but no traction so far. A bit unfortunate. It sounds like on desktops the ARB_compute_shader GL extension provides a lot of the functionality you'd get through OpenCL with less new feature surface (it can piggy-back on WebGL), while on mobile there is currently not common access to OpenCL. Interested to see whether either of those situations change.

yes I'd be interested to see that too. I mean, ultimately, a unified method for cross browser parallel computing should only be a matter of time right?

Close, but it looks like the pipeline for WebGL extensions will get feature-wise, in 5 years, where OpenCL was upon release ~6 years ago.. if even. Basing webgl standards on already obsoleted mobile opengl standards is bad for the web.

(Our startup is deep in this space, and after using both WebCL and WebGL, are going a third way.)

One one hand I'm happy to see the web able to do more and more. On the other hand, I worry that as we turn the browser into a platform with all these client side capabilities, browsers will become big complicated messes.

They already are, to one extent or another. Browsers are rivaling monolithic kernels and entire operating systems with userspace applications in size.

I wish we would see more browser features (cross-browser standards, app deployment style, ability to run random 3rd party code relatively securely, app interoperability) moved into the OS, rather than more OS features moved into the browser.

We already have them, it is just web devs tend to blindly ignore them.

I can't quite yet make a native, cross-OS app I can deploy by clicking a link and that anyone will trust not to eat their computer just by running it. It's also a lot more work to implement things like networking.

JNLP deployment of Java applications.

Click once deployment for .NET applications as another example.

Only nobody likes Java desktop applications. Including me, and I've programmed professionally in the language since 1998.

And they are non starters for demanding multimedia work, which is some of the most interesting stuff you want to do in the desktop as opposed as a web app.

.NET feels better (because MS didn't screw up as much, as Java did with the overengineered uncanny valley mess that is Swing), but it's not cross platform.

So, still, not comparable to deploying in the browser sandbox.

Mono can run .Net apps on lots of platforms. I'm running the same Tomboy (note-taking) binary on Linux and Windows.

> So, still, not comparable to deploying in the browser sandbox.

That much is true, I haven't yet used so brain damaged set of programming tools as the HTML/JavaScript/CSS gimmick required to make the so called web applications in all browser versions required by our customers.

The problem is then you're stuck with OS dependencies. Browsers are cross-platform.

Given how web development is "write once, debug everywhere, implement hacks everywhere" I have another understanding of cross-platform.

At least you have the chance to debug everywhere. Current native tools often are just fundamentally incompatible. Even if we just had hacky cross-os standards that'd be better than nothing.

> At least you have the chance to debug everywhere.

Good luck with mobile and set top box browsers.

> I worry that as we turn the browser into a platform with all these client side capabilities, browsers will become big complicated messes.

They aren't already?

Why was this posted now? It's been around since 2011, and since then hasn't seen any significant browser penetration as far as I know.

Have there been new developments?

Last I see on the Blink front is https://groups.google.com/a/chromium.org/forum/#!topic/blink..., essentially: 'No'.

Clicked the link, then 'Specification 1.0'.

"WebCL Specification Version 1.0, 14 March 2014"

The old specification was a draft. This seems to be a final version.

Embedding computationally hard problems into users' browsers and leeching users' computing hardware to solve these problems for you? I can't think of an actual application's use case for this that's not nefarious.

Reminds me of the MIT Jersey kids who were about to try doing in-browser Bitcoin mining using WebGL.

folding@home? hardly nefarious

Why would you want F@H in a browser instead of as an app?

atwoods law

Well, for one, you get a completely crossplatform method of executing and deploying what would typically be especially painful to distribute to a wide user base spread across various hardware families. Think of everything you can do with openCL then take the subset of that which is applicable to games, or spreadsheets, or novelty,- a non-trivial amount- and make that portable via browsers to create a much better experience for you and target users. Worst case is it provides an easier to use way to learn opencl for beginners in that area.

If I remember correctly, and I may be terribly incorrect, the MIT jersey kids were going to have you mine bitcoins instead of look at ads; I think it got shot down due to ability to abuse it rather than inherent badness.

WebCL is already running on Tizen: https://www.youtube.com/watch?v=TurCVdaUTMY So that is going to be my next cellphone ;)

If you can ever get one.

Imagine Facebook/Google renting out capacity on its users computers for companies and giving money to people for being able to do so.

Reply on news.ycombinator.com