"Despite the last few years of dismal news on the employment front, software engineers with backgrounds in high performance computing are in high demand. This is mainly due to the fact that HPC systems require engineers trained in the intricacies of parallel programming -- OpenCL, MPI, OpenMP, CUDA, and such. While these software frameworks are well known in the HPC realm, most computer science programs do not offer classes in them at the undergrad level. And there are only a handful of specialized HPC curriculums in the country, most of which are associated with DOE or NSF supercomputing centers." The Khronos Group is hoping that their Khronos Institute of Training and Education (KITE) initiative will help to ease the void in this field.
AMD is using the Taiwan Semiconductor Manufacturing Companies new 28nm manufacturing process to build the new high-end Radeon HD 7970 GPU. The HD 7970 sports 4.3 billion transistors in a 365mm2 die. The 7970 supports OpenCL and OpenGL 4.2. It is assembled from 32 GCN compute units, which translates to 2,048 stream processors, each based on AMD’s new SIMD-plus-scalar architecture. As well, the 7970 includes 768KB of L2 cache and eight render back-ends, features a 384-bit interface to 3GB GDDR5 memory and a PCIe 3.0 interface. The GPU is capable of peak throughput of 264GB/s.
HT4U staff have benchmarked 55 GPUs using three tests with three applications compatible with OpenCL. Especially interesting are the results with the 3D rendering application LuxRender. The article is written in Portuguese, but the results are easy to understand for all.
AMD gDEBugger is a real-time OpenCL and OpenGL debugger and memory analyzer integrated into Microsoft Visual Studio. The new v6.1 adds concurrent debugging of multiple kernels, greatly improved overall kernel debugging performance, especially in scenarios where large sections of code are being skipped, and overall application stability improvements both for the client (Visual Studio) and server (debugged application) sides.
AMD APP SDK v2.6 is officially released. Notable features are enhanced performance, multi-GPU support on Linux, inclusion of the Khronos C++ wrapper, use of OpenCL kernels in data-driven applications and a preview of OpenCL 1.2 and the static C++ kernel language.
In this talk form the 2011 LLVM Developers Meeting, Intel presents their OpenCL SDK 1.5 and its core technology the vectorizer compiler.
The video presents an overview of the implicit vectorization module and discusses experience with the LLVM compiler toolkit. The presenter also presents some of the design decisions and and plans for future features (future instruction sets, vector select, predicated instructions, etc).
A paper recently published by the Astronomical Society of Australia on how GPGPUs are dramatically changing the landscape of high performance computing in astronomy.This paper identifies and investigates several key decision areas, with a goal of simplifying the early adoption of GPGPU in astronomy. The merits of OpenCL are considered as an open standard in order to reduce risks associated with coding in a native, vendor-specific programming environment, and present a GPU programming philosophy based on using brute force solutions.
In a quiet post on the LLVMdev mailing list, it was announced that AMD is going Open Source with the AMD IL Code Generator which is compatible with LLVM 2.9. Also, video and slides from the recent 2011 LLVM Developers' Meeting are starting to appear online. Including SKIR: Just-in-Time Compilation for Parallelism with LLVM, Univeristy of Colorado; Intel OpenCL SDK Vectorizer, Intel; Polly - First successful optimizations - How to proceed?, ENS/INRIA and finally Symbolic Testing of OpenCL Code from Imperial College London.
AMD has released an interesting white paper on OpenCL and the Future of Accelerated Application Performance, as seen through the eyes of their FirePro product lineup. Whether you are a beginner OpenCL developer or an advanced developer, AMD has done a good job of making this a worthwhile read.