updated instructions

Author: marius10p

Docs/readme_win_linux.txt

@@ -2,14 +2,14 @@ Assuming gpu functions have been correctly compiled (see below), a "master_file.
 
 The following are instructions for setting up mex compilation of CUDA files with direct Matlab inputs. Note you need Matlab with the parallel computing toolbox. Instructions below are for Linux and  Windows. Mac instructions are in a separate file (I haven't tested them though). If successful, you should be able to run mexGPUall. 
 
-You should be able to run the code on a CPU without compiling any files, but it will be much, much slower than even on 250$ GPUs. For 32 channel data though, it might be fast enough. 
+You should be able to run the code on a CPU without compiling any files, but it will be much, much slower than even on 250$ GPUs. For up to 32 channels, the CPU code might be fast enough. 
 
 Windows
 
 Install Visual Studio Community (2012 or 2013)
-Install CUDA 7.5 (https://developer.nvidia.com/cuda-75-downloads-archive), not 8.0, which is not yet compatible with Matlab. Comes with compatible Nvidia drivers. If you get a warning of not finding the GPU at the beginning of installation, you should try to install more recent Nvidia graphics drivers, and then try again to install the CUDA drivers without installing the graphics drivers (for GTX10xx series this will be the case). Ignoring the warning appears to install CUDA without proper paths and then compilation does not work.
+Install CUDA 8.0 in Matlab R2017a (and 7.5 in R2016 etc; if in doubt, best to update to latest Matlab and CUDA). If you get a warning of not finding the GPU at the beginning of installation, this might be fine, unless you cannot run mexGPUall, in which case you should try a different combination of Nvidia/CUDA drivers. 
 
-Try to run mexGPUall. If mexcuda gives you an error, try the following: copy mex_CUDA_win64.xml (or nvcc_msvc120.xml, or a similarly named file, compatible with your Visual Studio installation version; 11 for 2012 and 12 for 2013) from here
+Try to run mexGPUall. If mexcuda gives an error, try the following: copy mex_CUDA_win64.xml (or nvcc_msvc120.xml, or a similarly named file, compatible with your Visual Studio installation version; 11 for 2012 and 12 for 2013) from here
 matlabroot/toolbox/distcomp/gpu/extern/src/mex/win64 and into the KiloSort folder (or somewhere in your path). The included file with KiloSort will NOT be compatible with your local environment (unless you start changing paths inside it). 
 
 If your video card is also driving your display, you need to disable the Windows watchdog that kills any process occupying the GPU for too long.