HP Accelerator Program

High Performance Computing

Catalysts for Innovation Program

HP-CAST – HPC Users’ Group

Related links

Events

Welcome to the HP Accelerator Program where you'll find information on HP accelerator products, solutions, partners, and much more.

The Accelerator Program is a hands-on, on-going effort to investigate various acceleration technologies for high performance computing (HPC) and benchmark them to qualify and productize those most beneficial for our customers. We also develop software for these technologies and collaborate with a number of software tool vendors to help ensure their products work well in scale-out HP systems based on accelerators. We have installed HP accelerated clusters at many customer sites, including the Tokyo Institute of Technology on the TOP500 list, over various industry segments worldwide.

HP has a GPU enabled system for you

While others are just starting to work with accelerator/graphical processing unit (GPU)-enabled servers. HP has been shipping GPU enabled systems for over 3 years and has a robust portfolio of over 14 different GPU enabled systems in multiple configurations. HP is now in its second generation with servers that are designed specifically to integrate GPUs with the HP ProLiant SL390s G7 Server. The SL390s G7 systems offer all the richness of the HP ProLiant heritage. You can choose from 1U HP ProLiant SL390s G7 servers without factory-integrated GPUs, or 2U ½ width HP ProLiant SL390s G7 servers with up to three integrated NVIDIA Tesla 20 Series GPUs or the 4U ½ width with up to 8 GPUs with the ability to put 16GPUs and 2 servers in 4U.

The 1U servers offer:

• Integrated 10GbE/IB
• HP Integrated Lights Out-3 (iLO3) capabilities
• Density – 8 servers in 4U
• Front serviceability
• Rich configuration – full-power CPUs on 1U with large memory capacity

The 2U servers offer all the above plus integrated NVIDIA Tesla GPUs for substantial additional compute capabilities in addition to hot swap drives. The 2U ½ width server HP ProLiant SL390s G7 supports up to 3 NVIDIA Tesla GPUs per server with

• Dedicated PCI e x16 lanes for each of the NVIDIA Tesla GPUs
• 1 additional PCIe for I/O
• 4 servers and 12 NVIDIA Tesla 20 series GPUs in 4U

The 4U ½ width server HP ProLiant SL390s G7 supports up to 8 NVIDIA Tesla GPUs per server with

• 2 servers and 16 GPUs in 4U
• 8 hot swap disks

HP has been shipping GPU enabled systems for over 3 years and has a robust portfolio of GPU enabled systems, in multiple different configurations using the NVIDIA 10xx and 20xx series GPUs. HP has the right GPU enabled system for you.

Learn more about the HP SL390s GPU 2U and 4U enabled systems supporting up to 3 GPUs and 8 GPUs per system http://h18004.www1.hp.com/products/quickspecs/Division/Division.html#13741

Accelerators are co-processing components containing massive numbers of functional units, together with memory and control systems that can be added to computers to speed up applications. Accelerators are like turbochargers in an automobile with the purpose of increasing the speed of an application with high performance per watt, for faster time to business, engineering, or scientific outcomes. Not every application can use accelerators, but for the ones that can, accelerators can really make the application fly. Below is an example of several benchmarks with and without GPUs running on a SL390s G7. The blue bar represents the run with system processors only. The red is with the GPUs added.

For LAMMPS, the benchmark was run both as a single-precision (SP) and double-precision (DP) floating point. All of the others were run using Double Precision.

As you can see from the chart above, GPUs can significantly improve performance.

HP has a rich portfolio of accelerator enabled systems supporting NVIDIA GPUs. Technical details on NVIDIA Tesla GPUs can be found at
http://www.nvidia.com/object/tesla_product_literature.html

The GPU enabled systems are listed below. They are updated on a regular basis – please check back periodically for updates. (If you are interested in having a card added to the qualified list, please click Accelerator-eng-support@hp.com

For information on software and drivers for the SL390 see
http://h18004.www1.hp.com/products/servers/software/index.html

The information from NVIDIA on driver support for GPUs can be found at.
http://www.nvidia.com/Download/index.aspx?lang=en-us

HP provides a set of tools from HP and Partners to provide robust software development, operation and management of GPU enabled systems. HP's GPU enabled products include HP CMU (Cluster Management Utility) for integrated management of the cluster and GPUs as well as HP HPC Linux Value Pack. The Value includes LSF workload-scheduling and HP MPI for message passing. Details on CMU can be found at www.hp.com/go/cmu and HP MPI and LSF in the HP Linux Value Pack at http://h20311.www2.hp.com/hpc/us/en/linux-value-pack.html

Partners make it easier for you to deploy and use GPU-enabled systems. The HP HPC partner ecosystem continues to expand—offering a growing list of development tools and solutions from industry-leading independent software vendors.

GPU development environments and tools – to simplify reliable code development

• PGI: GPU compilers and GPU simulation
• Allinea: GPU debugging
• RogueWave (TotalView): GPU debugging
• CAPS: GPU development environment and training
• Microsoft: GPU development environment
• NVIDIA: CUDA +OpenCL GPU development languages
• Altair PBS Professional: Portable Batch System (PBS) workload scheduling
• Platform LSF: fast, scalable service-oriented architecture grid computing middleware
• ATI: OpenCL GPU development language

• NVIDIA libraries: included in the CUDA Toolkit also includes supporting drivers, a Software Developer's kit with code samples and documentation. Its available to all GPU computing registered developers
• Numerical Algorithms Group (NAG) libraries: offers 1600+ routines that are both flexible and portable. They are used in thousands of programs and applications spanning the globe
• MathWorks – MATLAB: is the language of technical computing used across several industries with native support for GPUs on your desktop or on clusters; MATLAB users can take advantage of CUDA-enabled devices through built-in GPU accelerated functions or by connecting their CUDA kernels to MATLAB.
• Wolfram Research: Mathematica; provides routines and libraries for GPUs for various industries
• Microsoft: Windows HPC Server 2008 R2; native GPU support and GPU workstation integration
• RogueWave: IMSL numerical libraries

Contact HP for a current list of solutions, applications and development environments and tool partners.