| 研究実績の概要 |
In FY23, we continued with MocCUDA approach to speedup the large scale execution of deep learning frameworks on Fugaku. Additonally, we hosted an intern from ETH Zurich, with whom we published in the high-ranked NSDI'24 conference: "A High-Performance Design, Implementation, Deployment, and Evaluation of The Slim Fly Network", a continuation of our routing work specifically tailored for SimFly networks. Furthermore, we established a collaboration with Univ. of Illinois at Chicago, and our work on co-designing a better Alltoallv for GraphNN and FFT is under review at ICPP. Lastly, together with ETH we designed a holistic toolchain, called EvalNet, which can simulate large-scale networks. The tool performs topology generation, routing, path analysis, and metric evaluation at various scales.
|
