AMD releases new version of ROCm, becoming a universal open source platform for GPU computing
— Support for new GPU hardware, math libraries, and modern programming languages to further unleash the power of GPU computing —
SALT LAKE CITY - November 14, 2016 - At the 2016 Supercomputing Conference (SC16), AMD (NASDAQ: AMD) today announced a new version of the Radeon Open Compute Platform (ROCm), which includes software support for new Radeon GPU hardware, new math libraries and a strong foundation for modern programming languages, designed to accelerate the development of high-performance, energy-efficient heterogeneous computing systems. AMD also announced plans to support OpenCL™ and a variety of CPUs in the upcoming release of ROCm, including support for AMD's upcoming "Zen" architecture CPUs, Cavium ThunderX CPUs and IBM Power 8 CPUs, solidifying ROCm as the universal open source platform for GPU computing.
“Radeon Open Compute is the platform for a new era of GPU-enabled problem solving, leveraging the power of open source software to enable new solutions for HPC and hyperscale computing,” said Raja Koduri, senior vice president and chief architect of Radeon Technologies Group at AMD. “Today’s announcement of ROCm gives developers unprecedented flexibility in how and how they use GPU computing. We’re excited to showcase these incredible applications at SC16.”
“An open source approach to heterogeneous computing can help bring the benefits of high-performance computing directly to developers, providing them with the flexibility to leverage available compute resources and extract exceptional performance from available hardware,” said Denise Dumas, vice president of platform engineering at Red Hat. “As a leader in cultivating and maintaining open source communities, we are excited to see AMD taking a community-first approach with ROCm.”
“AMD GPU support for ThunderX®, combined with the massive capabilities of ROCm, is a valuable addition to Cavium’s overall solution offering for the HPC and hyperscale server markets,” said Larry Wikelius, vice president of the Software Ecosystem and Solutions Group at Cavium. “ROCm is an excellent complement to the ThunderX® innovations, including 48 ARMv8 cores, I/O integration and dual-socket configuration, that our customers and partners are currently deploying in the market.”
-
Expanded GPU Support - ROCm now supports all Polaris architecture-based graphics products, including Radeon™ RX 460, 470 and 480 graphics cards, and Radeon™ Pro WX 7100, 5100 and 4100 GPUs, enriching the ROCm supported device portfolio. The Polaris architecture is designed for low-level programming, helping developers get more out of the hardware.
-
ROCm virtualizes GPU hardware through OS containers and Linux® Kernel Virtual Machine (KVM) - ROCm now supports Docker containerization, allowing end users to simplify application deployment in ROCm-enabled Linux server environments. ROCm also supports GPU hardware virtualization through KVM pass-through to realize the benefits of GPU hardware-accelerated computing in virtualized solutions.
-
Heterogeneous Compute Compiler (HCC) - HCC is a single-source ISO C++11/14 compiler for CPUs and GPUs with support for the C++17 "Parallel Standard Template Library". It is based on a rich compiler infrastructure including LLVM-based GCN ISA code generation and assembler and disassembler support.
-
Portable Heterogeneous Compute Interface (HIP) - HIP enables developers to port CUDA applications to ROCm using HIPIFY, which automatically converts CUDA applications into the HIP kernel language and runtime APIs, creating portable applications that can run on any GPU using NVIDIA's CUDA compiler or HCC.
-
New math acceleration library – ROCm now supports a new advanced math acceleration library that supports BLAS, FFT, and N-dimensional tensor contraction.
The new version of Radeon Open Compute can be obtained from the following address: https://github.com/RadeonOpenCompute/ROCm
At SC16, booth 1431, AMD will demonstrate:
-
ROCm Technology Clusters running machine learning code on Supermicro® servers
-
Porting CUDA application Caffe through HIP porting tool
-
Demonstrating Ray Tracing and VR Virtualization for HPC Using AMD FirePro™ S9300 X2 and Radeon™ R9 Nano GPUs
-
OpenMP 4.5 interoperability for multiple GPUs and multiple platforms
-
BMPower8 servers and AMD FirePro™ S9170 server GPUs running ROCm
-
ARMv8ThunderX-based Penguin Tundra™ Extreme server uses Radeon™ RX 460 to run ROCm
-
Native rendering with OpenCL using guideless OpenGL/EGL interop on ROCm
-
Learn more about heterogeneous system architecture
-
Learn more about GPUOpen
-
Learn more about ROCm
-
Become a @AMDChina Weibo fan and follow the latest news from AMD
-
Follow AMD China on WeChat for more information
-
Get Radeon on Weibo @AMDChina ™ Pro Latest News
-
Get the latest Radeon™ graphics card news on Weibo @AMDChina
京公网安备 11010802033920号