Although the emergence of multi-core CPUs has allowed manufacturers to keep pace with the steadily growing demand for more computing power, energy consumption, space, and cooling have become major inhibitors to computing systems expansion. Hence the success of accelerators such as Intel® Xeon Phi™ coprocessors or GPUs (Graphics Processing Units) offering breakthrough performance plus outstanding space and energy efficiency - though they require an important programming effort to adapt codes to their highly parallel architecture.
Derived from technologies initially developed for game machines, NVIDIA®'s GPUs are highly parallel processors designed to seamlessly boost High Performance Computing applications. Implementing GPU computing is getting easier every month as new tools become available.
The Intel® Xeon Phi™ coprocessors were designed specifically for HPC. Relative to standard Intel Xeon processors, coprocessors have many more smaller cores, many more hardware threads, and wider vector units. The high degree of parallelism compensates for the lower speed of each individual core to deliver higher aggregate performance for highly parallel applications.
To help you innovate with accelerators with maximum payoff, Bull offers complete accelerator-based solutions:
Developed by the Bull R&D teams, the bullx B515 ultra-dense, double-width blades combine
They can be associated with B510 compute blades within the same bullx chassis.
The integration of the accelerators inside the blades maximizes performance. The bullx B515 blades are the only blades on the market designed to provide maximum bandwidth between each accelerator and its host CPU, thanks to a double interconnect on each blade:
Developed by the Bull R&D teams, the bullx DLC B715 accelerator blades are based on the same direct cooling system as the bullx DLC B710 compute blades, but they associate within the same blade:
They can be combined with B710 compute blades within the same bullx Direct Liquid Cooling chassis.
Designed specifically for high performance computing, the bullx R421 E3 accelerated servers combine over 1U:
The bullx R421 E3 is less dense than the bullx B515 or bullx DLC B715 accelerator blades, but it offers more flexibility, with a wider choice of options and connectivity.
Depending on the type of application and its degree of optimization, accelerators can boost processing by a factor of 1 to 100! Expertise is therefore essential to get the most out of accelerators. Bull has a long experience of systems associating accelerators and CPUs, and places this distinctive expertise at the service of Extreme Computing customers. Bull engineers have had hands-on experience of porting and optimizing applications for accelerated computing through many customer projects - AT CEA, GENCI, Petrobras or Barcelona Supercomputing Center. Bull experts deliver end-to-end services, from system design to advanced user training, through application optimization for accelerators.
In the last few years, Bull has designed for its customers accelerator-based systems of all sizes, from test systems with a few GPUs – purchased by several universities for example – to very large hybrid systems with tens of thousands of cores such as those developed for GENCI, CEA, BSC and Petrobras. Bull helps its customers optimize their accelerator utilization, with very varied codes such as BIGDFT (nano-sciences), SPECFEM3D (seismic), NAMD (biochemistry), but also for oil and gas research with the porting of RTM (Reverse Time Migration) codes to accelerators.
The bullx supercomputer suite integrates all the tools needed to operate a hybrid system comprising Intel® Xeon®-based servers and NVIDIA® Tesla™ GPUs or Intel® Xeon Phi™ coprocessors. The task management environment supports the allocation of applications to the relevant computing resource. An application can naturally use both computing resources so as to fully capitalize on the rich potential offered by this type of system.
The best of accelerator technologies to boost your High Performance Computing system
The best of accelerator technologies combined to Direct Liquid Cooling
A flexible server with accelerators
This brochure published by NVIDIA outlines the advantages of GPUs and describes some real-life implementations with GPUs.