AMD annuncia la nuova scheda grafica per server AMD FirePro S9150, la più potente GPU per server mai realizzata per l'High Performance Computing (HPC). Basata su architettura Graphics Core Next (GCN), AMD FirePro S9150 è la prima scheda grafica per server a superare la barriera dei 2 TFLOPS in precisione doppia. Con una potenza di 5,07 TLOPS in precisione singola e di 2,53 TFLOPS in precisione doppia (un valore che corrisponde alla metà di quello registrato in precisione singola), mette a disposizione dei supercomputer un'enorme capacità di calcolo, grazie anche ai suoi 16GB di memoria GDDR5 (il 33% in più rispetto alla concorrenza), all'interfaccia di memoria a 512 bit e a un'ampiezza di banda di 320 GB/s. A queste caratteristiche si accompagna un livello massimo di consumo pari a 235 watt, a riconferma dell'impegno di AMD volto al contenimento dei consumi energetici.

AMD FirePro S9150 è supportata da un ricco ecosistema software, che permette così agli sviluppatori di sfruttare la potenza della scheda in misura ancora maggiore. Oltre al supporto a OpenCL (che verrà esteso a OpenCL 2.0 entro la fine dell'anno), AMD annuncia infatti anche l'avvio di una collaborazione con Pathscale per il supporto a OpenMP 4.0 su AMD FirePro S9150 e altre schede selezionate della famiglia AMD FirePro.

AMD presenta inoltre la nuova GPU per server AMD FirePro S9050, con 12GB di memoria GDDR5, supporto alla memoria ECC e prestazioni grafiche e di calcolo all'avanguardia grazie ai suoi 3,2 TFLOPS in precisione singola, agli 806 GFLOPS in precisione doppia, all'interfaccia di memoria a 384 bit, a un'ampiezza di banda pari a 264 GB/s, al supporto a OpenCL 1.2 e a un consumo massimo pari a 225W.

La disponibilità di entrambe le soluzioni è prevista nel corso del terzo trimestre di quest'anno, distribuita a livello globale da SAPPHIRE Technology, mentre server e schede madri compatibili sono disponibili nelle versioni realizzate da ASUS, GIGABYTE, Supermicro e TYAN.

AMD FirePro S9150

[Immagine ad alta risoluzione]

AMD FirePro S9050

[Immagine ad alta risoluzione]

AMD (NYSE: AMD) today announced the AMD FirePro S9150 server card – the most powerful server Graphics Processing Unit (GPU) ever built for High Performance Computing. Based on the AMD Graphics Core Next (GCN) architecture, the first AMD architecture designed specifically with compute workloads in mind, the AMD FirePro S9150 server card is the first to support enhanced double precision and break the 2.0 TFLOPS double precision barrier1,2. With 16GB of GDDR5 memory – 33 percent more than the competition3 – and maximum power consumption of 235 watts, AMD FirePro S9150 server GPUs provide supercomputers with massive compute performance while maximizing available power budgets.

"Today’s supercomputers feature an increasing mix of GPUs, CPUs and co-processors to achieve great performance, and many of them are being implemented in an environmentally responsible manner to help reduce power and water consumption," said David Cummings, senior director and general manager, professional graphics, AMD. "Designed for large scale multi-GPU support and unmatched compute performance, AMD FirePro S9150 ushers in a new era of supercomputing. Its memory configuration, compute capabilities and performance per watt are unmatched in its class, and can help take supercomputers to the next level of performance and energy efficiency."

The AMD FirePro S9150 server card is supported by a powerful software ecosystem, enabling developers to better harness its compute performance. In addition to AMD support for OpenCL, including support for OpenCL 2.0 planned later this year4, AMD today announced a collaboration with Pathscale to support OpenMP 4.0 for AMD FirePro S9150 and selected other AMD FirePro graphics cards.

"Pathscale’s ENZO Compiler with OpenMP 4.0 and support for C, C++, and Fortran is used by customers in the Oil and Gas, Computational Science, Computer Aided Engineering, and other HPC segments," said Christopher Bergstrom, CTO, Pathscale. "Our support for the AMD FirePro S9150 will enable these customers to benefit from the tremendous compute performance of AMD’s new server GPU while leveraging their existing investment in OpenMP software."

"Having developed a supercomputer that ranked number two on The Green500 list, it takes exceptional compute performance and efficiency to achieve that status -- we couldn’t have done it without AMD FirePro S-Series server GPUs," said Professor Dr. Volker Lindenstruth, Frankfurt Institute for Advanced Studies, Goethe University, Frankfurt. "We’ve been putting the AMD FirePro S9150 through its paces and we’re blown away by the performance-per-watt. Any supercomputer that leverages massive compute power with the AMD FirePro S9150 and makes a run for the top of the Green500 list stands a very good chance of succeeding."

"We’re currently testing the AMD FirePro S9150 server card and hands down it’s the fastest HPC GPU accelerator we’ve ever tested. We’re even more excited by the prospects of combining its raw compute performance with OpenCL 2.0 later this year," said Simon McIntosh-Smith, head of the Microelectronics Research Group at the University of Bristol. "Our research benefits from an open and royalty-free API, and we’re committed to OpenCL to accelerate our computations. We applaud AMD’s planned support for OpenCL 2.0. New features such as Shared Virtual Memory and Nested Parallelism will allow us to make even better use of GPUs to increase performance and efficiency."

"Supermicro offers the industry’s widest range of GPU server solutions optimized for HPC and supercomputing in 1U, 2U, 3U, 4U SuperServer and 7U SuperBlade platforms," said Don Clegg, Vice President of Marketing and Business Development at Supermicro. "Supermicro Green Computing solutions, featuring AMD’s latest AMD FirePro S9150 and AMD FirePro S9050, enable extreme compute density with support for up to 6x GPUs per node in 4U dual node, dual processor configurations, maximizing performance per watt, per dollar, per square foot."

"ASUS is pleased to feature AMD FirePro S9150 accelerators to enable server workstations with dedicated and reliable functionality matched with double precision floating point performance," said Tom Lin, general manager, ASUS. "We’re dedicated to offering our customers solutions that deliver the most performance possible, and the energy efficiency and performance-per-watt of the AMD FirePro S9150 are an added bonus our customers will greatly appreciate."

"AMD FirePro S9150 delivers exceptional double- and single-precision floating point performance. Unlike other GPU accelerators, AMD FirePro S9150 features enhanced double precision, eliminating the need to buy a more expensive accelerator to get similar performance," said Etay Lee, general manager at GIGABYTE’s Network & Communications Division. "Together with AMD, our product delivers the right solution for the market to enable power-efficient performance for supercomputer-class results."

"AMD FirePro S9150 GPU accelerators will be an outstanding addition to our product line, allowing us to offer our customers solutions that deliver maximum performance," said Albert Mu, vice president of MiTAC International Corp.’s TYAN Business Unit. "During the testing and evaluation of the AMD FirePro S9150 with TYAN servers, our team was impressed with the sheer horsepower. It far exceeded our expectations and our customers will be anxious to take advantage of that horsepower to accelerate their compute-intensive workloads."

The AMD FirePro S9150 server GPU features:

• 5.07 TFLOPS of peak single-precision floating point performance – up to 18 percent more than the competition1
• Maximum double-precision floating point performance resulting in 2.53 TFLOPS peak double-precision compute performance – half of its peak single precision performance.
• Industry leading memory configuration3 – 16GB GDDR5 memory, 512-bit memory interface and up to 320 GB/s memory bandwidth
• 2,816 stream processors (44 GCN compute units)
• Error Correcting Code Memory support (external)
• Ready to support OpenCL 2.04
• 235W maximum power consumption
• AMD STREAM technology5

• 12GB GDDR5 memory, 384-bit memory interface and up to 264 GB/s memory bandwidth
• 1,792 stream processors (28 GCN compute units)
• OpenCL 1.2 support
• Error Correcting Code Memory support (internal and external)
• 225W maximum power consumption

• Comments
• Commenti