There is a recent paper on using GPUs for remote sensing processing, titled Remote Sensing Processing: From Multicore to GPU. It is using the open-source Orfeo Toolbox (OTB). A GPU implementation of the OTB is also available.
In the CUDA newsletter, there was a small announcement about a pre-build GPU mini-supercomputer. It contains 24 GPUs, a little more than our system with ‘only’ 16 GPUs:
The new GPU Starter Kit from HP is a pre-configured system that provides researchers with a ready-to-use GPU computing cluster, straight out of the box. It combines eight HP ProLiant SL390 G7 servers (containing 24 Tesla M2070 GPUs) with 16 CPUs. It is pre-configured with CUDA 4.0.
More information can be found here.
It includes two interesting presentations related to GPUs:
SnuCL, developed in Seoul National University, has made part of its source code public. It relies on the SNU-Samsung OpenCL compiler, which is delivered only in binary format. It will hold a tutorial in PACT 2011.
Note: It may take 10 minutes to download the package. The download speed is limited to 200KB/sec. The package size is 116MB.
The proceeding of International Symposium on High Performance Computer Architecture (HPCA) 2011 is available. Two interesting papers are related to the GPUs.
The proceeding of International Symposium on Performance Analysis of Systems and Software (ISPASS) 2011 is available. Some papers are related to our research.
Intel has announced ispc, the Intel SPMD Program Compiler. ispc is a new compiler for “single program, multiple data” (SPMD) programs. Under the SPMD model, the programmer writes a program that mostly appears to be a regular serial program, though the execution model is actually that a number of program instances execute in parallel.
The ispc compiler is based on the LLVM compiler and can be downloaded from http://ispc.github.com/
Judging by the example on their website, it looks a bit like CUDA, so it wouldn’t be too hard to get started with if you are familiar to GPU programming.