As the world awaits the scientific discoveries needed to bring quantum processors to commercial applicability, Atos’ BullSequana XH3000 allows for ecosystem participation within the compute platform itself and future-proofs any early buyer investments. In its Feb. 16 official announcement of the XH3000 supercomputer, for which TBR was provided pre-briefing access, Atos claims the product will have a six-year life cycle and that it is an open architecture capable of housing up to 38 blades. The blades can accommodate a mix of different XPU processors, with more under consideration and development.
The rapid rise in large data sets and evolving AI/machine learning (ML) algorithms have driven this global appetite for greater compute capacity — an appetite that many data scientists believe will only be sated once quantum computers reach commercial viability. Atos’ early lead in quantum simulators and alliances with various quantum systems vendors imply the company will be capable of pivoting its high-performance computing (HPC) offerings quickly to accommodate the addition of commercial-grade quantum processors when they arrive. Atos’ flexible hybrid supercomputing architecture will sell well in Europe for a variety of reasons and may enable Atos to gain share against notable HPC vendors in North America and Asia.
Data and AI require new compute platforms to address intractable problems
Atos correctly asserts the state of compute trails the size of the data sets that are available to run algorithms. Specifically, the world is running out of computational capacity to address the complex problems that can now be simulated and analyzed through increasing digitization.
Proof points offered in the Atos announcement included:
- Average HPC job durations grow as larger data sets will be applied against systems with as many as 10,000 nodes and 25,000 endpoints.
- Application refactoring and algorithm refinements can provide as much as a 22x speed improvement.
- Data centricity and edge processing grow in use case applicability, requiring greater hierarchical depth and more localized compute near the application.
- Hybrid Sim/AI Workflows for approximate computing are nearing reality. Atos offered the example of Alphafold 2 for protein folding prediction reaching over 90% accuracy, whereas classical methods currently achieve between 30% to 40% accuracy.
- Yet another industry prediction of reaching the physical limits of Moore’s law now that the industry is at 3nm technology.
- Extending the performance gains from classical computing while quantum discovery and commercialization advance will require greater innovation around multiple XPU architectures. These hybrid or heterogenous compute architectures need a new compute system structure, which Atos believes the XH3000 system provides.
The Atos Exascale strategy is a hybrid approach that serves many masters
Atos states the future of supercomputing will be hybrid. According to Atos, the future of supercomputing will involve a hybrid approach, consisting in the near term of a blend of classical CPU configurations and specialized processor architectures to address specific workload requirements. Presently, Atos collaborates with AMD (Nasdaq: AMD), Intel (Nasdaq: INTC), Nvidia (Nasdaq: NVDA), SiPearl and Graphcore, among others. Eurocentric chips based on ARM designs are also in the news and have been discussed by Atos.
Atos has addressed the need for future-proof flexibility in its designs by building the standard chassis of the BullSequana XH3000 to accommodate up to 38 compute/switch blades on one rack to be mixed and matched as workflows require from the different blades currently available and available in the future.
This hybrid architectural design approach serves many masters, such as those addressing:
- Sustainability: Different cooling and processing designs not only generate greater computational capacity but also, when coupled with the hybrid configurations and algorithm innovations, can lead to lower power consumption, and therefore lower carbon footprints.
- Sovereignty: Technonationalism is not going away, and Atos is a flagship European technology vendor. Former Atos CEO Thierry Breton is now the commissioner for internal market affairs within the European Union (EU) and has been tasked with managing many elements pertinent to digitization and “enhancing Europe’s technical sovereignty.” The EU has clearly stated its intentions to ensure there are European-controlled processors in market. Hybrid computing structures enable companies to select different processors to address the computational requirements amid the increased attention nation states place on compute access as a strategic national interest.
- Higher performance: The HPC market increasingly takes on the dynamics of emerging ecosystem business models and requires a physical compute stack that can accommodate the many tech stack variations the ecosystem can create to address the world’s compute and AI challenges. Atos claims it also has built the architecture to be resilient and adaptable for six years without forklift upgrades. This flexibility, Atos asserts, can accommodate new discoveries as the unknowns around deep learning, algorithm development and new processor developments in the classical and quantum computing realms come into view.