NVIDIA CUDA Toolkit Offered For SteamOS

[phoronix]

Phoronix: NVIDIA CUDA Toolkit Offered For SteamOS

Today while setting up some CUDA benchmarks, I noticed that CUDA 7.5 is now officially available for SteamOS...

http://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-CUDA-SteamOS

[miah]

I would speculate that since CUDA is required to enable NVENC support, Valve is working on supporting Steam in home streaming with nvenc.

[DarkFoss]

https://www.reddit.com/r/pcmasterrac...visual/cwukpuc

Thanks for the link. Too bad most won't read and even fewer will care.

[samdraz]

//t's interesting they list SteamOS but not upstream Debian.// Debian doesn't have 352.x series in unstable ony in exp.

[samdraz]

//t's interesting they list SteamOS but not upstream Debian.//
Debian doesn't have 352.x series in unstable ony in exp.

[samdraz]

//t's interesting they list SteamOS but not upstream Debian.//
Debian doesn't have 352.x series in unstable ony in exp.

[chepati]

Can someone explain what CUDA is good for? Can I use it to speed up regular gcc compiles of, say vanilla kernel, from source? Or chromium and qt5, which take hours to build. Or is it only to compile apps that will need a CUDA-enabled Nvidia GPU at runtime?

I am building lots of C and C++ programs from source and if I can harness the power of my SLI-ed dual GTX980, that would be a huge improvement. I am now using ccache + distcc, but I suspect that GPUs can greatly improve/reduce my compile times.

[adakite]

Can someone explain what CUDA is good for? [...]

CUDA (as OCL), is good for certain, although very specific, computing tasks. In HPC/Science computing, this is more and more used and useful for accelerating part of existing codes (i.e., loops for instance). That is specifically well designed for image processing, n-body simulations, statistics over large population, seismic inversion and so on. There are many others fields in addition where CUDA/OCL are now used on everyday basis in research labs like mine.

As for CUDA on Debian, yes that is possible. i.e., https://packages.debian.org/jessie/d...a-cuda-toolkit

[nslay]

Can someone explain what CUDA is good for? Can I use it to speed up regular gcc compiles of, say vanilla kernel, from source? Or chromium and qt5, which take hours to build. Or is it only to compile apps that will need a CUDA-enabled Nvidia GPU at runtime?

I am building lots of C and C++ programs from source and if I can harness the power of my SLI-ed dual GTX980, that would be a huge improvement. I am now using ccache + distcc, but I suspect that GPUs can greatly improve/reduce my compile times.

CUDA is good for massively data parallel tasks. These are tasks that involve a lot of repetitive tasks on lots of data with little to no dependency which allows for parallel processing. Think of it more like a generalized version of SSE. Although, instead of very specific kinds of arithmetic operations, you can define your own custom kernels to do arbitrary tasks on chunks of data.

CUDA is so great because it naturally extends the C/C++ language to perform these kinds of data parallel tasks. You can very naturally declare functions that are to be run on host and/or device and launch them with specific kinds of data dimensionality (how you divvy up the data) in mind. It also provides a lot of support libraries for parallel random number generation, linear algebra, STL-like containers and algorithms for GPUs, and convolutional neural networks.

And all relatively easy to pickup if you already know C/C++!