[chedda]

I guessed something was lost in translation

[luis.hijarrubia]

Next Limit office, from last week:

[feynman]

Wow! Looks like you got Steve Jobs' memorial urn! Are there ashes still inside?

[Anders Peter Amsnæs]

[luis.hijarrubia]

Will it fit?

[feynman]

You can take the hair-dryers off and fold the PCB in half no problem.

[Scop-Amac]

Loog good for mac user...
thanks for working on it

[andyjacobs]

This is really great to see!

[dmeyer]

I don't know how the choice of CUDA can come as a surprise. It is the standard among 3D GPU renderers and Apple (as the predominant driver of OpenCL standard) has been extremely neglectful of their OpenCL implementation. So much so that they are pushing their own Metal API instead.

I love the Mac but I'd hate to see NL hold back progress on the GPU renderer by limiting themselves to ancient GPU technology.

[burnin]

Just because majority believes so & have chosen CUDA over OCL, that doesn't makes it a standard (not yet)...
Now observe this: openCL on an 'older' Mac with AMD 7970 GPU (simple test shows engine is a bit faster than redshift, is also spectral & physically accurate as MR is supposed to be (ie. study Moet_Chandon scene), provided along is also it's own shading language, open public beta (not fooling anyone )...
Learn about the profession and expertise, please. Get acquainted with what you don't know & then try to make a meaningful post.

[JDHill]

Learn about the profession and expertise, please. Get acquainted with what you don't know & then try to make a meaningful post.

You chose an ironic target for this statement -- dmeyer has probably forgotten more than most here have ever known, when it comes to such questions.

[burnin]

You chose an ironic target for this statement -- dmeyer has probably forgotten more than most here have ever known, when it comes to such questions.

Respect...
No pun intended.
I have no excuse. Relativity in works: "What matters now are remains of what once was."

[jwiede]

Learn about the profession and expertise, please. Get acquainted with what you don't know & then try to make a meaningful post.

You chose an ironic target for this statement -- dmeyer has probably forgotten more than most here have ever known, when it comes to such questions.

Regardless, his statements about "ancient GPU technology" are inaccurate, and myopic.

First, most of the "technological advantages" of CUDA over OpenCL were erased by improvements to OpenCL standard made in the first couple years of its development. Historically, there's also a strong, established precedent for consumers accepting reasonable, abstraction-associated performance costs in order to support/advance hardware- and platform-generic interface SDKs.

Next, AMD is a major proponent of OpenCL, as are other non-Nvidia GPU makers. Like it or not, CUDA is still Nvidia-only for all practical intents. AMD GPUs alone represent a very substantial portion of the overall GPU market automatically excluded by a CUDA choice, Win or Mac, and there are other GPU makers with substantial market shares of their own (Intel, for example). Nvidia is by no means the only GPU maker with "modern" or "advanced" GPU tech, either. In reality, supporting OpenCL actually guarantees that NextLimit's Maxwell GPU fortunes are not tied to any single GPU manufacturer. The assertion that by supporting OpenCL, NextLimit is somehow "limiting themselves to ancient GPU technology" is irrational and baseless.

Most of the popular GPU render engines are now actively developing (some have even completed) OpenCL ports. ChaosGroup's Vray RT (incl. with Vray 3.x) already supports OpenCL, and Vray is commercially one of the most popular render engines, period. Like it or not, OpenCL adoption rates are rising (and accelerating in their rise). OpenCL ultimately isn't about or dependent on Apple or Mac, it's about offering developers an actual, open, non-proprietary standard SDK for GPGPU usage.

In the end, there's no historical basis for thinking hw-specific/-proprietary GPGPU SDKs will do any better against hw-generic GPGPU SDKs, any more than hw-specific/-proprietary 3D SDKs did against hw-generic 3D SDKs.

Honestly, why NextLimit thought it was a good idea to tie themselves to the proprietary interface owned by a company who themselves are direct competitors of NextLimit in the render engine market is beyond me. Then again, a lot of the decisions being made by NextLimit of late seem rather poorly thought through.

[Mihai]

I have no idea of the differences between OpenCL and CUDA, but are there any render engines that did the reverse, meaning started development on OpenCL and then ported to CUDA? If there aren't any, why?

[JDHill]

Honestly, why NextLimit thought it was a good idea to tie themselves to the proprietary interface owned by a company who themselves are direct competitors of NextLimit in the render engine market is beyond me.

It's tempting to second-guess choices after the fact, but R&D was begun at some discrete point in time, on a project the ultimate feasibility of which was unknown, and with tools being in a given state at that time. Apparently, nvidia was judged the best choice, with the question perhaps then becoming whether it made sense to add/switch technologies midstream. But, unless "made sense" equates to "would have accelerated productivity", then to suggest switching is to suggest delaying release, for the sole purpose of supporting opencl. My own decision would have been: full steam ahead on nvidia, release, and then look at refactoring with the benefit of hindsight, and improvements made in the interim, while porting to opencl.