|
| |||||||||||||
AsHES 2027 : 17th Workshop on Accelerators and Hybrid Emerging Systems, with IPDPS 2027 (Seattle) | |||||||||||||
| Link: https://www.ashes-hpc.org/2027 | |||||||||||||
| |||||||||||||
Call For Papers | |||||||||||||
|
CALL FOR PAPERS
The Seventeenth International Workshop on Accelerators and Hybrid Emerging Systems (AsHES 2027) Held in conjunction with IPDPS 2027 June 1, 2027, Seattle, USA (full-day workshop) Website: https://www.ashes-hpc.org/2027 Submission: https://ssl.linklings.net/conferences/ipdps/?page=Submit&id=AsHESWorkshopFullSubmission&site=ipdps2027 IMPORTANT DATES * Paper submission: January 22, 2027 (Friday) * Acceptance notification: February 22, 2027 (Monday) * Camera-ready: March 6, 2027 (Saturday) * Workshop: June 1, 2027 (Tuesday) SCOPE Heterogeneous systems are the new standard. Driven by the demands of energy efficiency, memory capacity, and computational throughput, modern architectures combine diverse processing units and memory technologies within hierarchical systems, and fields ranging from AI and big data to cloud, edge, real-time, and high-performance computing have achieved significant gains from them. ML/AI has become the fastest-moving driver of heterogeneous hardware, and energy has become its binding constraint: leadership-class systems are designed to power envelopes of tens of megawatts, per-accelerator power draw has exceeded a kilowatt, and for large-scale ML/AI what can be deployed is increasingly determined by available power. The efficiency gains now available derive from specialization: domain-specific accelerators, reduced precision, near-memory and dataflow designs, and power-aware scheduling. Emerging hardware such as wafer-scale accelerators, near-/in-memory processing, disaggregated and CXL-based systems, smart NICs and switches, photonic computing, and dataflow accelerators promises both new challenges and opportunities. AsHES examines heterogeneous design implications across the entire computing stack, from hardware and compiler optimizations to application development and programming environments. It brings together researchers and practitioners at the forefront of heterogeneous computing to learn the opportunities and challenges in future heterogeneous system design and to help influence the next trends in this area. For AsHES 2027 we particularly encourage submissions on (1) ML/AI on heterogeneous systems, (2) energy-efficient methods for HPC and ML/AI systems, and (3) the use of emerging hardware features, devices, and accelerators. TOPICS OF INTEREST (not limited to) Applications and Workloads * Applications leveraging GPUs, FPGAs, TPUs, DPUs, on-chip accelerators, or other novel architectures * AI/ML workloads including generative AI training/inference (e.g. LLMs, LDMs), recommendation systems, and AI for science and engineering * Applications and workloads on emerging hardware: wafer-scale accelerators, near-/in-memory processing, disaggregated/CXL systems, smart NICs/switches, photonic computing * Scientific and engineering applications * Big data and cloud computing workloads * Edge computing and real-time applications * Application behavior characterization and performance analysis * Domain-specific applications and benchmarks Systems Architecture and Design * Heterogeneous system architectures (ARM, RISC-V, custom extensions) * Memory hierarchies and novel memory systems, including near-/in-memory processing and disaggregated/CXL memory * Energy efficiency and thermal considerations * Energy-efficient methods for HPC and ML/AI systems: domain-specific accelerators, reduced precision, near-memory and dataflow designs, power capping and power-aware scheduling * Security and reliability in heterogeneous systems * Design space exploration and trade-off analysis * Performance and power modeling and prediction * Emerging accelerator architectures (dataflow, wafer-scale, neuromorphic, photonic, etc.) Programming Models and Development Tools * Programming models and interfaces (CUDA, HIP, OpenMP, OpenACC, SYCL, OneAPI, Kokkos, Raja) * Debugging and profiling tools * Performance and energy optimization and autotuning frameworks * Testing and verification methods for heterogeneous systems * Workload characterization tools * Benchmarking suites and methodologies * Heterogeneous system software and runtime systems * Compiler optimizations for heterogeneous systems Runtime Systems and Resource Management * Operating system support for heterogeneous computing * Virtualization and containerization for heterogeneous systems * Task scheduling and load balancing, including power-aware and energy-aware scheduling * Memory management and data movement optimization * System-level monitoring, telemetry, and energy measurement * Integration of heterogeneous components and interfaces * Communication software/middleware for heterogeneous systems, including smart NICs and switches PAPER TRACKS AsHES 2027 has two paper tracks: 1) Full papers: up to 10 pages, including references. 2) Short papers: up to 4 pages, including references. Intended for early investigations of innovative ideas and position papers on emerging topics. Authors of accepted short papers present their work as a short talk and as a poster in the IPDPS cross-workshop poster session. Papers must follow the IPDPS/IEEE conference format and are submitted through Linklings. Review is single-blind; submissions need not be anonymized. Every submission receives at least three reviews. Accepted papers are published in the IPDPS workshop proceedings. The submission form asks how you learned of AsHES. The answer is used only to improve our publicity and is not visible to reviewers. BEST PAPER AWARD The best paper receives a certificate of recognition, awarded on the basis of originality, impact, and overall quality. POSTER SESSION AsHES 2027 takes part in the IPDPS 2027 cross-workshop poster session on the evening of the workshop day (June 1, 6:00-8:00 pm). Posters are required for short papers, and authors of accepted full papers are strongly encouraged to present a poster as well. Details will be provided with the acceptance notifications. IN-PERSON PRESENTATION At least one author of each accepted paper must register for IPDPS 2027 and present the paper in person in Seattle. All workshop chairs will attend in person. Requests for remote accommodation are referred to the IPDPS Workshop Chairs; the AsHES organizers cannot grant them. The submission form includes a checkbox for authors who anticipate needing a visa letter. The response is not visible to reviewers and plays no part in the decision. ORGANIZERS General Chair * Sarunya Pumma, Meta, USA Program Chairs * Eishi Arima, Technical University of Munich, Germany * Jonathan Lifflander, Sandia National Laboratories, USA CONTACT Questions about the workshop or submissions: sarunya@meta.com |
|