Facebook AI guru alt-tabs out, Nvidia EULA audits, Baidu changes, chip tricks, and more

Welcome to El Reg's January roundup of AI-related news beyond all the wonderful and terrible things we've covered separately. Drop us a line if you have any machine-learning news or gossip to share. Facebook AI chief LeCun steps aside – Yann LeCun, considered to be a pioneer of neural networks for computer vision, has stepped …

[Teiwaz]

The thrill risk of a surprise software cavity search license audit is a really attractive inducement to using their toolkit, hardware tied or not for small projects.

Corporate or even Academic use, it's just another compliance issue.

[ibmalone]

Watch out for universities dropping CUDA then... people used to be quite keen on getting a GPU card to try stuff out and this would lead to people writing code that would eventually make it to Teslas and create demand for computing clusters, but academic institutions are not particularly keen on surprise audits and have less oversight over what individuals are doing than businesses.

[Anonymous Coward]

OpenCL it is then

I for one have only recently started to port code to CUDA, and was going to commit to that

and unified virtual memory & Pascal and later. However, it makes me think again, and I

need to cross platform target AMD and even integrated Intel graphics too. Complex memory

model be damned.

A chip vendor and compiler vendor applying constraints on the downstream

usage model (say what you can and cannot compile and run) is outrageous.

I buy the part, and the software to use as I see fit.

They are simply trying to gouge scientists and AI farms to for them to use the £10,000 product instead

of the £500 product. This will not end nicely. The pressure is there, and the invisible hand will find a way to drive them off this track.

If it has to be OpenCL for my code instead, well fine. OpenCL it is.

By charging so much for a Tesla, the whole GPGPU goal of democratising supercomputing is being placed at peril, and now they are trying to use the language lock in that made us all fear a proprietary language extension like CUDA in the first place to hold a gun to the heads of those who invested in developing code.

Sod them. Cuda is dropped.

[Anonymous Coward]

EULA on Titan drivers + EULA on CUDA

These two moves are incredibly adversarial against a user base who

exploits GPU's with tolerance of the quirky programming model & prioprietary

software environment.

Large elements of the user base will remember this.