Graphcore, based in Bristol, UK, has developed a processor and cards for machine learning in the data centre. SiPearl is developing a range of high performance processors in France and Germany for supercomputer applications.
Combining machine learning and high-performance computing will accelerate the evolution of simulation and prediction models, say the two companies with systems co-developed Graphcore and SiPearl. These will open up the possibility for developers and users to design new generations of data flows and application tasks that are both more efficient and easier to program and use. GraphcCore for example has its own AI framework called Poplar, and being able to use that across both the Intelligent Processor Uit (IPU) AI chips and SiPearl's processors would be a key advantage for developers.
The two companies also highlight lower power consumption than Nvidia GPUs and Intel processors for AI and HPC applications.
"Graphcore and SiPearl share a vision of enabling innovators across a broad range of industries from medicine to automotive to finance, and beyond. We both understand that new types of computation are essential if we are to realise the full potential of new technologies, like AI. This is a great opportunity for Europe to define the shape of that next-generation compute,” said Fabrice Moizan, General Manager and SVP Sales EMEAI and APAC at Graphcore.
The GraphCore IPU-M2000 is a 1U datacentre blade based around its Colossus GC200 IPU processors, capable of one PetaFlop of AI compute. For many AI training and inference tasks, the IPU systems significantly outperforms the latest GPU-based systems.
The IPU-POD systems offer the ability to run large models at datacentre scale, or to share their compute resource across multiple users and tasks. Since its founding in 2016, Graphcore has raised over $710 million in funding.
"We welcome this partnership with a European leader of Graphcore’s calibre, whose IPU is revolutionizing artificial intelligence. In the microprocessor segment, which represents half of the global semiconductor market, our respective technologies are perfectly complementary. Combining them will allow system manufacturers to develop a turnkey product that brings our end customers more performance, speed and energy efficiency,” said Philippe Notton, CEO and founder of SiPearl.
SiPearl is designing a high-performance, low-power microprocessor using the ARM Neoverse Z1 core for the European exascale supercomputer as part of the European Processor Initiative (EPI) project. This project has 28 partners in the scientific community, supercomputing centres and leading names from the IT, electronics and automotive industries.
SiPearl is also a member of the Mont-Blanc 2020 consortium to equip Europe with a dedicated modular and energy-efficient high performance computing microprocessor, and is a member of the PlayFrance.Digital collective for Europe to lead the field for digital technology.
Other articles on eeNews Europe