Message boards : GPUs : CUDA Toolkit
Message board moderation
Author | Message |
---|---|
Send message Joined: 19 May 15 Posts: 123 |
There are instructions for Linux nVidia CUDA support at Ubuntu, although outdated, linked from install info on the BOINC Wiki. I've seen notes about installing the Toolkit on other sites in regards to CUDA. I came across official CUDA installation page at nVidia. http://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html Do we actually need the TOOL KIT or only the drivers for CUDA? It seems you would only need the Tool Kit if you wanted to develop something with CUDA... however, I know sometime something is needed in a tool kit to run (in general). Irregardless, I have a 1 gb download before I can install anything. That's an offline install. I don't know if an online install (which I didn't see till after I was already downloading) with just drivers would be smaller, although I would think so. TIA ~Yav |
Send message Joined: 5 Oct 06 Posts: 5129 |
To run NVidia applications, you need two things: 1) A suitable hardware driver for your GPU. It's the user's responsibility to obtain and install this. 2) The runtime support files for the programming language used by the application programmer. It's usually the programmer's (or in our case, the BOINC project's) responsibility to supply these. CUDA is NVidia's proprietary programming language, and the toolkit contains vastly more than is needed simply to run compiled programs. Programmers need it, ordinary users don't. The toolkit does, however, contain everything necessary for a programmer to get started. So, there's a driver in there, and all the necessary library files. Downloading the toolkit is a way of getting hold of any missing components that the programmer has failed to pass on to you. I don't know the file names in the Linux context, but for Windows the ones you are most likely to need are: cudartxx_yy.dll cufftxx_yy.dll where xx is the bitness (32 or 64), and yy is the version number. You need the version which exactly matches what your programmer developed for. |
Send message Joined: 19 May 15 Posts: 123 |
That's what I figured, but the page linked from BOINC was for installing the full Tool Kit Package. GPU computing on BOINC needs a good clean-up and round-up. :) It appears it may be a lot more straight-forward than it use to be. I was trying to test it last night but SETI was only serving me OpenGL work so I couldn't tell if CUDA was properly installed and detected. I just tried testing with Asteroids and it gave me a CUDA WU and looks it's running it okay. Yav |
Send message Joined: 29 Aug 05 Posts: 15569 |
...but SETI was only serving me OpenGL workCorrection, that's OpenCL. OpenGL, or Open Graphics Library, is a cross-language, cross-platform application programming interface (API) for rendering 2D and 3D vector graphics. It's a DirectX competitor. OpenCL, or Open Computing Language, is a framework for writing programs that execute across heterogeneous platforms consisting of central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), field-programmable gate arrays (FPGAs) and other processors or hardware accelerators. It's a DirectCompute competitor. While both run on the same hardware, they don't do the same things. |
Send message Joined: 19 May 15 Posts: 123 |
Oops. I didn't see the CL. I just glanced at it and mind went GL. It appears the CUDA unit for Asteroids completed successfully, so that verified things and answered my own question too. I just wished I noticed the online option before I started downloading the toolkit... O.o Do any distros ship using actual nVidia drivers and full support loaded? |
Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License,
Version 1.2 or any later version published by the Free Software Foundation.