Argonne National Laboratory
Postdoctoral Appointee – Mixed/Reduced Precision Computing on Modern Hardware
Argonne National Laboratory, Lemont, Illinois, United States, 60439
Postdoctoral Appointee – Mixed/Reduced Precision Computing on Modern Hardware
Because of the drastically increasing demand from AI/ML applications, the computing hardware industry has gravitated towards data formats narrower than the IEEE double format that most computational scientists and engineers are accustomed to. Moreover, the vast majority of the performance associated with these reduced precision formats resides on special hardware units such as tensor cores on NVIDIA GPUs, which specialize for a restricted set of floating point operations only. Many scientific applications, particularly those that are physics-driven and mission‑critical, still struggle to adapt to this new hardware trend. To help bridge the gap, the Argonne Leadership Computing Facility (ALCF) invites applications for a postdoctoral appointment on the subject of mixed/reduced precision computing on modern hardware. The duration of the appointment is one year initially and renewable for up to three years contingent on performance and funding. The successful candidate will be supported by ALCF's Performance Engineering group and is encouraged to engage with the broader Argonne scientific community. Moreover, the successful candidate is encouraged to experiment on the wide variety of systems available at ALCF, including both the large‑scale production machines and the testbed machines featuring novel architectures such as Cerebras and SambaNova. The list below provides examples of the potential tasks for the successful candidate and illustrates the general nature of the work but is not intended to be exhaustive. In addition, the successful candidate is encouraged to bring their inputs in the research direction during the appointment. Diagnosing and analyzing the numerical challenges related to the narrower data width Devising and evaluating novel techniques to exploit the reduced precision hardware Incorporating mixed/reduced precision into existing applications relevant to ALCF’s mission Developing and maintaining tools and libraries that facilitate the adoption of mixed/reduced precision computing in the broader community Collaborating with domain experts to understand how the trend of reduced precision hardware impacts their practices and provide support in return Publishing and presenting at professional venues Required skills and qualifications: A Ph.D. degree completed within the last 0‑5 years (or soon to be completed) in numerical analysis, applied mathematics, computational science, computer science, or another relevant field Proficiency in programming and knowledge on computing hardware Ability and willingness to work collaboratively in a team environment Effective written and oral communication and interpersonal skills Ability to model Argonne’s core values of impact, safety, respect, integrity, and teamwork Preferred skills and qualifications: Knowledge on floating point arithmetic and mixed/reduced precision computing techniques Experience with programming GPUs and/or other accelerators Proficiency in mathematical reasoning and numerical analysis Up‑to‑date awareness about the status and trend of the computing hardware industry Interest in working with domain experts on practical problems Experience with large‑scale distributed systems Knowledge on numerical linear algebra, numerical methods, high‑performance computing, or other related fields Seniority level
Entry level Employment type
Temporary Job function
Research Industries
Research Services
#J-18808-Ljbffr
Because of the drastically increasing demand from AI/ML applications, the computing hardware industry has gravitated towards data formats narrower than the IEEE double format that most computational scientists and engineers are accustomed to. Moreover, the vast majority of the performance associated with these reduced precision formats resides on special hardware units such as tensor cores on NVIDIA GPUs, which specialize for a restricted set of floating point operations only. Many scientific applications, particularly those that are physics-driven and mission‑critical, still struggle to adapt to this new hardware trend. To help bridge the gap, the Argonne Leadership Computing Facility (ALCF) invites applications for a postdoctoral appointment on the subject of mixed/reduced precision computing on modern hardware. The duration of the appointment is one year initially and renewable for up to three years contingent on performance and funding. The successful candidate will be supported by ALCF's Performance Engineering group and is encouraged to engage with the broader Argonne scientific community. Moreover, the successful candidate is encouraged to experiment on the wide variety of systems available at ALCF, including both the large‑scale production machines and the testbed machines featuring novel architectures such as Cerebras and SambaNova. The list below provides examples of the potential tasks for the successful candidate and illustrates the general nature of the work but is not intended to be exhaustive. In addition, the successful candidate is encouraged to bring their inputs in the research direction during the appointment. Diagnosing and analyzing the numerical challenges related to the narrower data width Devising and evaluating novel techniques to exploit the reduced precision hardware Incorporating mixed/reduced precision into existing applications relevant to ALCF’s mission Developing and maintaining tools and libraries that facilitate the adoption of mixed/reduced precision computing in the broader community Collaborating with domain experts to understand how the trend of reduced precision hardware impacts their practices and provide support in return Publishing and presenting at professional venues Required skills and qualifications: A Ph.D. degree completed within the last 0‑5 years (or soon to be completed) in numerical analysis, applied mathematics, computational science, computer science, or another relevant field Proficiency in programming and knowledge on computing hardware Ability and willingness to work collaboratively in a team environment Effective written and oral communication and interpersonal skills Ability to model Argonne’s core values of impact, safety, respect, integrity, and teamwork Preferred skills and qualifications: Knowledge on floating point arithmetic and mixed/reduced precision computing techniques Experience with programming GPUs and/or other accelerators Proficiency in mathematical reasoning and numerical analysis Up‑to‑date awareness about the status and trend of the computing hardware industry Interest in working with domain experts on practical problems Experience with large‑scale distributed systems Knowledge on numerical linear algebra, numerical methods, high‑performance computing, or other related fields Seniority level
Entry level Employment type
Temporary Job function
Research Industries
Research Services
#J-18808-Ljbffr