Announced in May this year, the SYCLOPS Project brings together 8 leading organizations from across Europe, collaborating to democratize AI acceleration and break the existing monopoly on the acceleration market. Our combined vision is using open standards to enable a healthy, competition-driven ecosystem with widespread adoption of both RISC-V®, an open standard Instruction Set Architecture (ISA) and SYCL™, the open specification parallel to ISO C++ and maintained by the Khronos Group.
Each collaborating partner brings their own expertise and unique insights to the project, and in this blog each member of the SYCLOPS project will introduce their organization, and the work they will be contributing to the project.
ACCELOM is a French SME specializing in the development of bespoke AI-based multi-omics pipelines for performing integrated analysis of genomic data from multiple molecular levels to provide a holistic picture of how genomics drives the progression of a disease in a way that would not be possible using traditional, non-integrated analysis methods. One key challenge faced by ACCELOM is that its pipelines are designed for a scale-up, single-server, CPU-based deployment as it is intended for its customers and collaborators who cannot move their data to the public cloud due to privacy reasons, and who cannot afford to operate large datacenters due to economic constraints. Through project SYCLOPS, ACCELOM will develop SYCL-GAL and use it to develop a new version of the multiomics pipeline. We will demonstrate the effectiveness of the new pipeline via thorough, extensive evaluation using publicly available datasets from European data sources
The European Organization for Nuclear Research (CERN) was established in 1954 near Geneva, Switzerland. At CERN, high energy physics collaborations conduct experiments with the help of particle accelerators. CERN currently operates the world’s most powerful particle accelerator, the Large Hadron Collider (LHC). The ROOT project at CERN provides software libraries and tools for the processing of the ever growing amount of data collected from the LHC. In order to sustain this data deluge the software needs to be ported to computing accelerators. ROOT provides use cases for the SYCLOPS project, porting them to SYCL and measuring their performance.
Codasip is a processor technology company enabling SoC developers to differentiate their products for competitive advantage. Codasip’s offering can be summarized as Custom Compute. The solution is based on the Codasip Studio design automation tools. The tools can be used to customize and optimize the company’s range of RISC-V processor cores by adding domain-specific or application-specific extensions, or to design a processor from scratch. Codasip’s role in the project is to develop domain-specific accelerators based on the RISC-V Vector extension (RVV). This will be achieved using Codasip Studio and the CodAL processor description language.
Codeplay Software, founded in 2002 in Edinburgh, Scotland is a leader in acceleration technologies used in the AI and HPC industry. In 2022, they were acquired by Intel to expand the oneAPI ecosystem based on SYCL, a cross-platform software abstraction for heterogeneous processors. As active members of the Khronos Group and leaders in the open standards ecosystem, Codeplay will use their expertise to lead on the communication activities of SYCLOPS and participate in conferences and events to showcase the ongoing work of the project.
EURECOM is a leading French engineering school and a founding member of the SophiaTech Campus in Sophia Antipolis, the largest Computer Science and Technology campus in Alpes-Maritimes region of France. The Data Science department of EURECOM brings its world-renowned expertise and reputation in the field of statistical machine learning, multimodal data processing and knowledge engineering, which empower artificial intelligence systems. The Data Science department is a founding member of the “Region SUD” Artificial Intelligence cluster, which was recently recognized by the French government as a cluster of excellence (3IA). In SYCLOPS, EURECOM acts as both the technical and administrative coordinator. In addition, EURECOM will also collaborate with various SYCLOPS partners in developing optimized algorithms for scaling the performance of various cross-architecture acceleration libraries.
Heidelberg University – also known as Ruperto Carola – was established in 1386 and is Germany’s oldest university. It is also one of the strongest research universities in all of Europe. The successes of Heidelberg University in the Excellence Competitions and in international rankings are evidence of its leading role on the academic scene. Heidelberg University is a comprehensive university with a broad spectrum of subjects and focus on research-based teaching and well-structured training for doctoral students. Currently, approx. 1.400 Postdocs and 8.900 PhD Students are enrolled and 36 ERC grants are running at Heidelberg University.
The Engineering Mathematics and Computing Lab (EMCL) is a research group at the Interdisciplinary Center for Scientific Computing (IWR) at Heidelberg University. EMCL is particularly noted for its openness to multidisciplinary research in the field of engineering and scientific computing. EMCL currently pursues five major research themes: Computational Mathematics, Uncertainty Quantification, Hardware Aware Numerics and High Performance Computing and IT Security. One of EMCL’s developments, AdaptiveCpp (formerly known as hipSYCL), is a flexible compiler and runtime system that implements the SYCL standard on multiple device architectures. In the course of the project, EMCL will expand AdaptiveCpp to support for RISC-V devices. This requires generalizing and extending the existing compiler and runtime infrastructure. Additionally, EMCL will work on techniques to increase the utilization of devices in the system by leveraging automatic work distribution. As a provider of a developer tool at the platform layer of the SYCLOPS stack, EMCL will work closely with the project partners to ensure that AdaptiveCpp is a reliable platform for the project.
HIRO-MicroDataCenters is a Dutch micro SME specializing in Edge as a Service, specifically MicroDataCenter solutions ( hardware and software) for Big Data processing and AI in data dense Edge environments in various Industries & applications. In SYCLOPS, HIRO will be developing high-performance, energy-efficient, heterogeneous edge micro datacenters (EMDC) that will serve as both development testbed to validate the SYCLOPS stack and as a reference platform for the design of future PCIe 5.0 compatible EMDCs in comExpress form factor.
INESC-ID (Instituto de Engenharia de Sistemas e Computadores: Investigação e Desenvolvimento em Lisboa) is a private, not-for-profit Portuguese R&D and Innovation organization that promotes cooperation between academia and industry and has been establishing excellence in R&D activities since 1999. INESC-ID now stands, and is viewed, as one of the most dynamic research institutes in Portugal in the broad areas of computer science and electrical engineering. The research relevance of the INESC-ID High-Performance Computing Architectures and Systems (HPCAS) group on performance modeling, application optimization, and parallelization on state-of-the-art computing platforms synergies among the objectives of the SYCLOPS project. In this regard, INESC-ID’s role in the project spans multiple layers of the SYCLOPS stack, which include the exploration of cross-architecture insightful performance models, profiling of the three SYCLOPS use cases, as well as enabling acceleration strategies within the SYCL-GAL library with SYCL-based version of pairHMM used for variant calling.