Europe’s bet on Arm took another step forward today with selection of an Atos BullSequana X1310 system by CEA’s (French Alternative Energies and Atomic Energy Commission) Military Applications Division (CEA/DAM). The new system, which Atos characterizes as the “first Arm-based productized supercomputer,” will use Marvell’s ThunderX2 64-bit processors.
The system was developed by Atos as part of the Mont-Blanc 3 project, funded by the European Commission. Mont-Blanc 3 is assessing the potential of Arm-based clusters to address exascale HPC needs, developing the corresponding software ecosystem, and will issue recommendations for a European exascale ARM-based system. CEA/DAM will use the new machine, at least initially, as an Arm validation platform.
As noted in the official announcement, “CEA has been developing large, multi-physics, mission critical codes for over 40 years. This is particularly the case of CEA’s Military Applications Division (CEA/DAM) since the end of the French nuclear tests 22 years ago. The validation of those codes is an important part of the work of developers and CEA’s teams have observed that running these codes on different platforms from different architectures and using a variety of compilers and libraries from different sources helps find bugs and deviations from standards. Therefore, in order to provide the best validation environment at scale to its code developers, CEA/DAM ordered an Arm-based BullSequana cluster from Atos, using the new ThunderX2 CPU from Marvell.”
Time will tell how broadly Arm is adopted in Europe and elsewhere. The processor market has seldom had so many seemingly viable options. Arm, long a major player in mobile/embedded applications, has been trying to mount a push into datacenter servers and into HPC.
“We see a lot of innovation and experimentation happening in the HPC processor arena. CEA and Bull, now Atos Bull, have had a strong innovation partnership for many years, so it’s not surprising that they would be among the first to exploit the ARM-based development within the Mont Blanc project. Our recent worldwide studies show that Intel x86 processors remain dominant, but there’s substantial interest in AMD x86 CPUs as well as ARM-based and IBM Power processors,” said Steve Conway, SVP of research, Hyperion Research.
Addison Snell, CEO, Intersect360 Research, noted, “The delivery of an ARM-based BullSequana supercomputer to CEA will be another important validation step in the establishment of ARM in HPC. There are already multiple, national-level exascale initiatives based on ARM. One of the biggest concerns for exascale is resiliency, how to maintain an operating environment at that level of scale. Based on its expertise in other supercomputing endeavors, Bull is migrating its operating environment to Mont-Blanc. The industry will be closely watching the success of this project, not only for its own merits, but for what it means to broader HPC community adoption.”
The new system will be delivered by the end of 2018. It includes a BullSequana rack with 92 BullSequana X1310 blades, three compute nodes per blade, dual Marvell ThunderX2 processors of 32 cores (2.2 GHz), based on the Armv8-A instruction set, with 256 GB per node and Infiniband EDR interconnect. The new ThunderX2 processor supports up to four threads per core with 8 memory channels delivering the combination of compute and memory bandwidth required for critical scientific workloads.
Sophie Proust Houssiaux, head of big data and security R&D at Atos, said in the announcement, “This is the culmination of our long-term commitment to the Mont-Blanc European project, which assesses the potential of using Arm-based clusters to address exascale needs. This system is the commercial version of the prototype developed and built, by Atos, as part of the Mont-Blanc project for tests and evaluation. The availability of this new model is fully in-line with our policy of technological openness and our support for the European exascale supercomputer effort.”
CEA/DAM has already received an initial set of blades to begin work on software porting, characterization and optimization. According to the official announcement, “the first results show that this new cluster is a very promising architecture for memory intensive applications.”