uncomplicate

2017-02-20T20:24:21.000087Z

Barely familiar with powershell and looking for a guide. Is anyone aware of a step by step guide for ATLAS installation on Windows 10?

2017-02-20T20:57:16.000088Z

There is. Please read the getting started guide about how to find (and where) pre-compiled ATLAS for windows. If you already have the libatlas.dll, the ONLY thing to do is to put it somewhere in the PATH, either by putting it in a directory that is already on the path (windows\system or something like this, I forgot the exact name) or to edit the path through environment variables.

2017-02-20T20:58:22.000089Z

The next version won't require that. I switched neanderthal to Intel's MKL, which is free to use and redistribute (but not open source).

2017-02-20T20:59:38.000090Z

MKL has pre-compiled binaries for all major systems that run intel on intel's chips (including amd's chips) so I guess it would be much easier for users who are more comfortable with installers on windows.

2017-02-20T20:59:54.000091Z

@aaelony see my previous messages

2017-02-20T21:03:39.000092Z

Thanks for that. An installer of some kind would be nice. Mostly though I need to get more familiar with win 10

2017-02-20T21:12:43.000093Z

Yesterday I installed MKL on Win 10 to build neanderthal-native for Windows, and it was typical 20 click next-next-next. Typical windows hassle, but works.

2017-02-20T21:13:51.000094Z

A bonus is that MKL is unbelievably fast. Roughly twice as fast as ATLAS on my machine for parallel execution.

2017-02-20T21:23:16.000095Z

will look into it. Also, to run bayadera i had to lein localrepo the commons jar manually just fyi

2017-02-20T21:25:18.000096Z

yes, I know. I will update bayadera to run with newest neanderthal and releasse commons after I get back from Berlin. The problem is that I had to write lots of documentation and tutorials before that release...

2017-02-20T21:25:51.000097Z

So, it runs without issues on your setup?

2017-02-20T21:26:42.000099Z

Well, I need to install atlas first ;) but other than a lot of warnings so far so good

2017-02-20T21:27:35.000100Z

noticed you used quil in your test examples, eager to see that run in a gorilla notebook

2017-02-20T21:27:47.000101Z

these warnings should not be from bayadera, but from the testing libraries.

2017-02-20T21:28:14.000102Z

once i get atlas running i think it will be great

2017-02-20T21:28:17.000103Z

I just used quil for defsketch. the plots are in processing

2017-02-20T21:28:26.000104Z

cool

2017-02-20T21:28:55.000105Z

you have a up-market AMD GPU that supports OpenCL 2.0?

2017-02-20T21:29:25.000106Z

have a titan x and win10

2017-02-20T21:29:57.000107Z

Not supported yet since Nvidia only supports OpenCL 1.2

2017-02-20T21:30:06.000108Z

:( darn

2017-02-20T21:30:20.000110Z

But, don't despair 🙂

2017-02-20T21:30:30.000111Z

I have a new GTX 1080

2017-02-20T21:30:40.000112Z

So, I'll create ClojureCUDA

👍 1
2017-02-20T21:30:44.000113Z

heheh, awesome

2017-02-20T21:30:50.000114Z

and CUDA backend for Neanderthal

2017-02-20T21:31:18.000115Z

and then I'll decide whether I'll support Nvidia through CUDA or OpenCL 1.2 in Bayadera

2017-02-20T21:31:26.000116Z

Probably CUDA, but it remains to be seen

2017-02-20T21:31:40.000117Z

Neanderthal works on Nvidia

2017-02-20T21:32:01.000118Z

And should run really fast on your card I suppose

2017-02-20T21:32:38.000119Z

Because it reaches 5.5 TFLOPS on mine, if I remember well

2017-02-20T21:32:47.000120Z

for sure, it's a beast

2017-02-20T21:33:06.000121Z

So, for Bayadera, you'll have to wait, but Neanderthal already works smoothly

2017-02-20T21:33:40.000122Z

once I have things set, I lmk what I can test

2017-02-20T21:34:34.000123Z

benchmark etc

2017-02-20T21:36:49.000124Z

cool. please share when you do. For the reference, the mm! example from opencl tutorial (see tests) runs in 220ms on GTX 1080 (8192x8192 matrices)

2017-02-20T21:38:34.000125Z

will try it out