Around 2000 year, computer scientists in various fields discovered the usage of GPUs to speed up scientific computation, which led to the appearance of General-Purpose computation on GPU(GPGPU). But at that time, because GPU was designed for graphics displaying only, such programs could only be written by using graphics programming APIs, such as OpenGL and Cg, but it's rather hard to code.
Later, GPU companies develop and publish interfaces for GPU programming, which results in the fast growth of GPU computing. Today, as NVIDIA's web site says, 600 universities around the world have courses on parallel computing; hundreds of thousands of developers are actively coding on GPUs.
A student has used this technique in 15-213 Datalab to compute two's complement arithmetic.
The resulting speedup made the student's code faster than optimal two's complement arithmetic on the hardware, surprising the course staff.
This comment was marked helpful 0 times.
lazyplus
Something worth noting about @Avesh's post: Based on the introduction of Autolab, all the students' code is run in virtual machines.
So, we have been demonstrated that GPU acceleration is provided even in virtual machines. Recently, Nvidia and VMware announced their first fully virtualized 3D graphics adapter in Aug 2012.[link].
So important the GPU acceleration is that people even want it in virtual machines!
This comment was marked helpful 0 times.
tnebel
This may be pretty obvious, but when looking back over this slide I had forgotten the point of setting the two triangles. So if anyone else has this confusion, it is just that setting two triangles is an easy way of creating fragments (correct terminology?) that cover the whole screen, since the renderer gets run over pixels contained in the fragments.
This comment was marked helpful 0 times.
kayvonf
@tnebel: Correct. Rendering two triangles that exactly ocver the screen with no overlap was simply a hack to emulate a "forall output pixels" loop.
Around 2000 year, computer scientists in various fields discovered the usage of GPUs to speed up scientific computation, which led to the appearance of General-Purpose computation on GPU(GPGPU). But at that time, because GPU was designed for graphics displaying only, such programs could only be written by using graphics programming APIs, such as OpenGL and Cg, but it's rather hard to code.
Later, GPU companies develop and publish interfaces for GPU programming, which results in the fast growth of GPU computing. Today, as NVIDIA's web site says, 600 universities around the world have courses on parallel computing; hundreds of thousands of developers are actively coding on GPUs.
http://www.nvidia.com/object/what-is-gpu-computing.html
This comment was marked helpful 0 times.
A student has used this technique in 15-213 Datalab to compute two's complement arithmetic.
The resulting speedup made the student's code faster than optimal two's complement arithmetic on the hardware, surprising the course staff.
This comment was marked helpful 0 times.
Something worth noting about @Avesh's post: Based on the introduction of Autolab, all the students' code is run in virtual machines.
So, we have been demonstrated that GPU acceleration is provided even in virtual machines. Recently, Nvidia and VMware announced their first fully virtualized 3D graphics adapter in Aug 2012.[link].
So important the GPU acceleration is that people even want it in virtual machines!
This comment was marked helpful 0 times.
This may be pretty obvious, but when looking back over this slide I had forgotten the point of setting the two triangles. So if anyone else has this confusion, it is just that setting two triangles is an easy way of creating fragments (correct terminology?) that cover the whole screen, since the renderer gets run over pixels contained in the fragments.
This comment was marked helpful 0 times.
@tnebel: Correct. Rendering two triangles that exactly ocver the screen with no overlap was simply a hack to emulate a "forall output pixels" loop.
This comment was marked helpful 0 times.