Learn our solutions and experiences for scaling up and increasing utilization of GPU resources on DGX and cloud platforms. Our solution, Backend.AI, is the GPU-first framework to host machine learning workloads and fully integrates with NGC. We will describe how it supports diverse-sized computations using both aggregation of multiple GPUs and fractional scaling of GPUs, and how it reduces modelling time for data scientists through a public competition using the DGX platform.
ABOUT THE SPEAKER: 현재 Lablup ("래블업") CTO를 맡아 Backend.AI를 개발하고 있습니다. CUDA, OpenCL, Xeon Phi 및 Intel DPDK를 활용하는 80 Gbps급 패킷 처리 가속 프레임워크로 KAIST 전산학과에서 박사학위를 받았으며, 래블업에서도 GPU 및 분산처리 기술을 중점적으로 다루고 있습니다.
Joongi is the main author of Backend.AI and CTO in Lablup. He received Ph.D in Computer Science from KAIST by developing a GPU-accelerated packet processing framework offering world-first 80 Gbps performance. He has diverse experiences on analysis and design of scalable backend systems via a research internship in Microsoft Research Cambridge and an engineering internship in NexR (acquired by KT). He is also an open-source enthusiast. He has contributed to a number of open source projects, such as Textcube, iPuTTY, Python, aiodocker, aiohttp, pyzmq, DPDK, and more. Currently enrolled as a global open-frontier program member in KossLab, sponsored by NIPA.