NVIDIA and its ecosystem partners will deliver, over the next few
months, the industry’s broadest set of software releases to developers using GPU
Computing in their work.
These updates feature major releases across a broad spectrum of GPU Computing
development languages, tools and libraries. Included are updates from NVIDIA for
its CUDA C compiler, with additional support for C++ and its upcoming GPU
codenamed “Fermi.” NVIDIA is also releasing its R195 driver that includes new
extensions to its OpenCL 1.0 conformant driver and toolkit, and a beta release
of the NVIDIA code name Nexus, the industry’s first development environment for
massively parallel computing, which is integrated into Microsoft Visual
Alongside NVIDIA’s own updates, several partner releases from industry
leaders in software tools are available now, including The Portland Group’s CUDA
Fortran solution, Allinea’s Distributed Debugging Tool (DDT) and the TotalView
“The only effective way to scale performance in demanding applications is to
move to a parallel computing model,” said Sanford Russell, general manager, GPU
Computing software at NVIDIA. “The NVIDIA CUDA architecture facilitates this
critical transition with its broad industry support and network of software
consultants and training resources for massively parallel computing.”
Updates to NVIDIA and its partners’ parallel computing development tools
include the following:
CUDA Toolkit 3.0
Beta: With the CUDA Toolkit 3.0 Beta, developers can start
developing applications today for the NVIDIA Fermi architecture. This beta
release includes features such as ECC reporting, Dual DMA Engine, Concurrent
Kernel Execution and NVIDIA Fermi HW debugging support in cuda-gdb. Performance
profiling is included for both CUDA Visual Profiler and the OpenCL Visual
Profiler. Also included is support for a new unified interoperability API for
Direct3D and OpenGL including Direct3D 11.
OpenCL 1.0 Extensions: NVIDIA
is the only vendor supporting OpenCL features beyond the minimum conformance
level. New extensions released by NVIDIA include support for double precision,
OpenGL interoperability and the new OpenCL Installable Client Device (ICD).
These new features supplement existing NVIDIA-only support for 2D image, 32-bit
atomics and byte addressable stores.
NVIDIA “Nexus”, the codename for the
industry’s first development environment for massively parallel GPU
applications, integrated into Microsoft Visual Studio IDE:
Comprised of a Debugger, Performance Analyzer and Graphics Inspector,
this beta release gives GPU Computing developers an immediate boost in
productivity through common and easy to use tools.
The Portland Group (PGI) – CUDA Fortran: Production release of the world’s first
Fortran compiler compatible with the NVIDIA CUDA-enabled GPUs. CUDA Fortran will
accelerate the adoption of GPU Computing in areas where applications are written
in Fortran, such as ocean modeling, weather forecasting, environmental modeling,
seismic analysis, bioinformatics and other areas.
Professional HPC Debugging Solutions from Allinea and TotalView were also launched this
week. These tools provide CUDA GPU features that complement existing
capabilities for parallel debugging using MPI, OpenMP and pthreads on the Linux
platform. It enables developers to debug applications that are running on hybrid
clusters of x86-64 CPUs and Tesla GPU-based servers.
Numerical Analysis Packages: Significant advances in the use
of CUDA-enabled GPUs have also been made in prominent numerical analysis and
mathematical modeling packages such as MATLAB from Mathworks, Mathematica from Wolfram Research and LabVIEW from
CUDA Libraries: In addition, developers can take advantage
of a rich set of CUDA-accelerated libraries available from NVIDIA and its
partners including BLAS, FFT, LAPACK (EM Photonics CULA), MAGMA (ICL at the
UTK), NVIDIA Performance Primitives (NPP), CUDA Vision Workbench (CVWB) and
video and image processing libraries.
To accelerate the momentum of CUDA optimized tools and applications,
customers can take advantage of worldwide training and consultancy services
offered by a growing number of CUDA
Consultants, such as Acceleware Corp., ANEO, CAPS, Elegant Mathematics, EM
Photonics, Fixstars, GASS Ltd., HPC Project, Infosys, SagivTech, Stone Ridge
Technology and Tech-X Corp.
The breadth of languages, APIs, libraries and other tools that are now
supported by NVIDIA graphics processing units (GPUs) based on the CUDA parallel
processing architecture represent the industry’s most flexible and pervasive set
of tools available for parallel computing today.