NVIDIA Announces "RAPIDS" Open-Source Data Analytics / Machine Learning Platform

[phoronix]

Phoronix: NVIDIA Announces "RAPIDS" Open-Source Data Analytics / Machine Learning Platform

NVIDIA has announced RAPIDS as their latest open-source project...

http://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-Open-Source-RAPIDS

[Meteorhead]

Does April fools come early this year? Nvidia doing something open-source in relation to CUDA? I thought closed-source binary blobs keep CUDA afloat.

[undu]

I feel like completely depending on CUDA kinda defeats the purpose of being open-source. At least the door to adopting to CUDA-like interfaces so it can interop with other GPU vendors is still open.

[M@yeulC]

NVIDIA [...] Open-Source

Where's the catch?

Edit: just to be clear, I'm just kidding, and willing to leave them the benefit of the doubt.

[cybertraveler]

i dont care bout macheen leerning

pls giv open sauce grphix drivers

k

thx

bye

[L_A_G]

Seems like AMD's "Open Source everything"-strategy is paying dividends when Nvidia, a company known for only open sourcing when absolutely necessary, following suit like this.

Let's hope this is not just a sporadic thing and instead the beginning of a change in overall strategy.

[shmerl]

NVIDIA claims that RAPIDS can allow for machine learning training at up to 50x and is built atop CUDA for GPU acceleration.

Garbage, despite being FOSS. It's like building a FOSS library that requires using DirectX. Unless it's a translation layer that decouples CUDA from Nvidia's hardware, it's something to avoid.

Where's the catch?

Edit: just to be clear, I'm just kidding, and willing to leave them the benefit of the doubt.

Well, see above. CUDA is the catch and perpetuates Nvidia's lock-in. Whether the library is open or not has no effect on it. So Nvidia gets free PR while still being their own nasty self.

[M@yeulC]

Garbage, despite being FOSS. It's like building a FOSS library that requires using DirectX. Unless it's a translation layer that decouples CUDA from Nvidia's hardware, it's something to avoid.


Well, see above. CUDA is the catch and perpetuates Nvidia's lock-in. Whether the library is open or not has no effect on it. So Nvidia gets free PR while still being their own nasty self.

Would they accept OpenCL patches? That would be pretty ironic
Although I have no doubts they would easily find a way to "nerf" an OpenCL backend, or have it fall behind in terms of features (bait-and-switch tactics, basically).

[anarki2]

Does April fools come early this year? Nvidia doing something open-source in relation to CUDA? I thought closed-source binary blobs keep CUDA afloat.

That's the dumbest thing I've read today so far (which is quite an accomplishment because I've already read several dozen articles' comments today). As if the CUDA runtime being open or not had anything to do with anything.

Developers and researchers use CUDA because it's reliable and performant, with a huge ecosystem built around it. Something AMD can't say about OpenCL, and about their GPUs in general.

Yeah, please do insert 9 million "works4me" comments here, the industry couldn't care less about your one man garage projects.

[Grinness]

That's the dumbest thing I've read today so far (which is quite an accomplishment because I've already read several dozen articles' comments today). As if the CUDA runtime being open or not had anything to do with anything.

Developers and researchers use CUDA because it's reliable and performant, with a huge ecosystem built around it. Something AMD can't say about OpenCL, and about their GPUs in general.

Yeah, please do insert 9 million "works4me" comments here, the industry couldn't care less about your one man garage projects.

This is serious garbage.

I have started working with cuda in 2008, it is a nice API and has the clear advantage of kernels being compiled by the nvcc. BUT it ends here.
CUDA is a proprietary interface, brings vendor lock-in -- can be used only with a specific vendor GPU (i.e. I cant use my own code because I do not own a NVIDIA gpu *)

OpenCL is a open standard (not AMD), and indeed works pretty fine -- it has an overhead when it comes to compile kernels -- BUT can be used across GPUs vendors and CPU + FPGA + DSPs .....

Researchers (I worked in research for long years) use it mainly because it was the first API/SDK enabling GPU computing.
It is time to move on, and embrace open (computing) standards!


* anymore