Thanks for this excellent introduction to GPU computing. In particular, the pros and cons provided for the technologies comparison are very interesting (and the analogy with guitars is fun ^^ ). Just a small remark about boost.compute: you can actually perform a copy from host to device asynchronously with copy_async, which returns a future. However, there is currently no continuation, which would be a nice evolution.
This talk is mostly designed around GPU based C++ libraries and frameworks for GPU CUDA or OpenCL programming and less about C++ as a language on GPUs.
Excellent introduction. A small comment: when Peter talks about vector-sum example, the formula a=a*scale+b is not actually the type of operation that is called xAXPY (saxpy, daxpy) in BLAS community. The actual xAXPY op is a = b * scale + a (or in more traditional notation, y = ax + y). But generally, AXPY stands for "A times X plus Y" so probably Peter's notation also fits the definition.
1:13 'please dont break it' -did it still work afterwards? Personally I think the chance of destructive electrostatic discharge is actually quite high.
Thanks for this excellent introduction to GPU computing. In particular, the pros and cons provided for the technologies comparison are very interesting (and the analogy with guitars is fun ^^ ).
Just a small remark about boost.compute: you can actually perform a copy from host to device asynchronously with copy_async, which returns a future. However, there is currently no continuation, which would be a nice evolution.
This talk is mostly designed around GPU based C++ libraries and frameworks for GPU CUDA or OpenCL programming and less about C++ as a language on GPUs.
Excellent introduction. A small comment: when Peter talks about vector-sum example, the formula a=a*scale+b is not actually the type of operation that is called xAXPY (saxpy, daxpy) in BLAS community. The actual xAXPY op is a = b * scale + a (or in more traditional notation, y = ax + y). But generally, AXPY stands for "A times X plus Y" so probably Peter's notation also fits the definition.
Amazing
1:13 'please dont break it' -did it still work afterwards? Personally I think the chance of destructive electrostatic discharge is actually quite high.
Oh yes, but the casing should be grounded.
usually with these cards the board is directly exposed, and not cased.
Comparing libraries - nitpicking. :D