New CUDA compiler enables parallel programmers to take full advantage of industry-standard CPUs from AMD and Intel

PORTLAND, Ore., Dec. 14 -- The Portland Group, a wholly-owned subsidiary of STMicroelectronics and a leading supplier of compilers for high-performance computing (HPC), today announced that a performance-optimized PGI CUDA C/C++ compiler for multicore x86 platforms (CUDA-x86) will ship with its PGI 2012 release due out in January 2012.

CUDA is NVIDIA's parallel computing architecture that enables dramatic increases in computing performance by harnessing the power of NVIDIA GPUs. Originally announced at the 2010 NVIDIA GPU Technology Conference (GTC), CUDA-x86 extends CUDA beyond the GPU into a system-wide programming model. The release of CUDA-x86 is a key step towards making the x86+GPU architecture an integrated parallel platform.

"CUDA-x86 is another key milepost in PGI’s roadmap for comprehensive support of programming of heterogeneous CPU+GPU systems," stated Douglas Miles, director, The Portland Group. "Directive-based GPU programming with the PGI Accelerator compilers makes it extremely easy to get started programming GPUs. CUDA extensions allow programmers to write explicit parallel algorithms that are inherently massively scalable, and to dive as deep as they want to go to optimize for a specific platform. We expect both approaches to be critically important as accelerator-based systems become more and more pervasive."

"CUDA is the most popular GPU parallel-programming model in the world today, and developers need the flexibility to target multiple architectures with the same code,." said Sanford Russell, director of CUDA marketing at NVIDIA. ."The release of CUDA-x86 delivers the benefits of CUDA as a general-purpose programming model for high-performance parallel applications running in heterogeneous computing environments."

The PGI compiler for CUDA-x86 processes CUDA C/C++as a native parallel-programming language for general-purpose multicore x86 microprocessors from AMD and Intel. CUDA-x86 includes full support for NVIDIA's CUDA C/C++ language for GPUs, so programmers can recompile CUDA application source code for execution on an x86 host. Using CUDA-x86 developers can compile and optimize their CUDA applications to run on x86-based workstations, servers and clusters with or without an NVIDIA GPU accelerator. CUDA C/C++ applications compiled for x86 targets use multiple cores and the streaming SIMD (Single Instruction Multiple Data) capabilities of Intel and AMD CPUs for parallel execution.

Highlights of PGI CUDA C/C++ compiler for multicore x86 include:

• Optimization and parallelization of native CUDA C/C++on x86 hosts.
• Low-overhead native parallel execution of CUDA C/C++ on x86 hosts.
• Executes each CUDA thread block using a single host core; automatically eliminates synchronization where possible.
• Support for the latest processors from AMD and Intel including support for the new AVX instructions.
• Automatically in-lines device kernel functions and translates chevron syntax to parallel/vector loops.
• Full support for NVIDIA's CUDA C/C++language for GPUs on x86 hosts.
• Full support for GPU texture memory.
• NVIDIA CUBLAS library support.
• Supports all PGI host optimizations for Intel/AMD.

In addition, PGI CUDA C/C++ for GPU devices is planned for release in mid 2012. At that time, using PGI Unified Binary™ technology, one binary will be able to use NVIDIA GPUs when present or default to using multicore x86 if no GPU is present.

Performance Data

In a performance comparison of popular parallel programming models, PGI compiled the CUDA-x86 version of a LBM benchmark (part of the Parboil benchmark suite) and compared execution time to the same program parallelized using OpenMP. Program execution times for the different programming models are shown below:

System: 4 core Intel i7 920 (2.67GHz), 12GB, Red Hat Enterprise Linux 5.3

Additional performance information is available on the PGI website at http://www.pgroup.com/cuda-x86

Price and Availability

The PGI CUDA C/C++compiler on multicore x86 is part of the PGI 2012 release version 12.1 due out in January. It is available at no charge to PGI Accelerator C/C++ licensees with a current PGI Subscription service. PGI products are supported on the Linux, Apple Mac OS X and Microsoft Windows operating systems. A 15 day evaluation is available from The Portland Group web site at www.pgroup.com. Registration is required.

About The Portland Group

The Portland Group, a wholly-owned subsidiary of STMicroelectronics, is the premier supplier of high-performance parallel Fortran, C, and C++ compilers and tools for workstations, servers, and clusters based on x64 processors from Intel and AMD, and GPUs accelerators from NVIDIA. Further information on The Portland Group products can be found at www.pgroup.com, by calling Sales at (503) 682-2806, or by email to sales@pgroup.com.

About STMicroelectronics

STMicroelectronics is a global leader serving customers across the spectrum of electronics applications with innovative semiconductor solutions. ST aims to be the undisputed leader in multimedia convergence and power applications leveraging its vast array of technologies, design expertise and combination of intellectual property portfolio, strategic partnerships and manufacturing strength. In 2010, the Company’s net revenues were $10.35 billion. Further information on ST can be found at www.st.com.

----

Source: The Portland Group