Gwangsun Kim

Employment

• Assistant Professor, POSTECH, Nov. 2018 - Present

• Senior Research Engineer, Arm Inc., Mar. 2018 - Oct. 2018

• Senior Performance Engineer, Arm Inc., Sep. 2016 - Mar. 2018

• Research Intern, NVIDIA, Jun. 2015 - Sep. 2015

• Research Intern, Samsung Electronics, Jul. 2014 - Sep. 2014

Research Areas

I am interested in various topics in computer architecture and its interaction with different layers of computer system including  algorithm, operating system, and programming models. Below are some of the topics I have been working on:

• Domain-specific Accelerators for Machine Learning

• Near-Data Processing / Processing-In-Memory 

• HW/SW Co-design

• Memory system with Storage-Class Memory

• Massively parallel architectures (e.g., GPU)

• Large-scale systems (Data centers and supercomputers)

• Interconnection networks

On-going Projects

• General-purpose Near-Data Processing (NDP) in CXL memory expander for datacenters

  • While directly accessing CXL memory from the host CPU results in high latency and low bandwidth, NDP in a CXL memory expander can overcome this challenge. However, application-specific NDP in CXL memory is not suitable for datacenter servers due to the large number of applications that need to be supported. Additionally, using conventional CPU and GPU cores for NDP provides limited efficiency. In this work, we propose a low-overhead, general-purpose NDP in CXL memory to address these challenges.

• Architecture for efficient inference of LLM (Large Language Model) and RAG (Retrieval Augmented Generation)

  • Efficiently serving LLMs on servers and edge devices poses enormous challenges. In particular, very long contexts of several millions of tokens require huge memory capacity and high bandwidth. Efficiently supporting RAG similarly requires memory system support for large vector databases and knowledge graphs. We build on our general-purpose NDP architecture for CXL memory to efficiently support these applications and overcome the limitations of current systems based on GPUs and NPUs.

• Scalable Neural Processing Unit (NPU) system architecture

  • Datacenters that serve a massive amount of machine learning service requests require that the NPUs have a good scalability at the chip-level (with many NPU cores), package-level (with multiple dies in a package), node-level (with multiple NPU cards), and rack-level (with multiple NPU nodes in a rack). This project explores software/hardware co-designing approaches to enable a very scalable NPU system for large-scale deep learning systems.

• Lossless tensor compression for high-performance DNN inference/training

  • DNN inference/training requires very high memory capacity and bandwidth. Meanwhile, the inherent redundancy and sparsity in DNN's tensors pose an opportunity for significantly reducing the size of tensors to increase the effective memory capacity and bandwidth. In this project, we are working on developing an effective hardware-based tensor compression algorithm to improve the overall system performance and energy-efficiency for DNN inference and training. 

Students

Ph.D./Integrated MS-PhD Students

• Hyungkyu Ham

• Jeongmin Hong

M.S. Student

• Wonhyuk Yang

• Geonwoo Park

• Yunseon Shin

• Jinhoon Bae

• Okkyun Woo

Researcher

• Heeeon Lee

Alumni

• Junkyung Choi (M.S., 2021, first employment: Nota AI)

• Junho Lee (M.S., 2022, first employment: OPENEDGES Technology, Inc.)

Publications

2024

• Low-overhead General-purpose Near-Data Processing in CXL Memory Expanders

  • Hyungkyu Ham*, Jeongmin Hong*, Geonwoo Park, Yunseon Shin, Okkyun Woo, Wonhyuk Yang, Jinhoon Bae, Eunhyeok Park, Hyojin Sung, Euicheol Lim, Gwangsun Kim

  • Preprint [*: Equal contribution]

  • [ arXiv ]

• NeuPIMs: NPU-PIM Heterogeneous Acceleration for Batched LLM Inferencing

  • Guseul Heo, Sangyeop Lee, Jaehong Cho, Hyunmin Choi, Sanghyeon Lee, Hyungkyu Ham, Gwangsun Kim, Divya Mahajan, Jongse Park

  • ACM International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS)

  • [ ACM DL ] [ arXiv ]

• Bandwidth-Effective DRAM Cache for GPUs with Storage-Class Memory

  • Jeongmin Hong, Sungjun Cho, Geonwoo Park, Wonhyuk Yang, Young-Ho Gong, Gwangsun Kim

  • The 30th International Symposium on High-Performance Computer Architecture (HPCA) (Accept. rate: 18.3%)

  • [ IEEE Xplore ] [ arXiv ] [ Slides ] [ Github ]

2022

• Overcoming Memory Capacity Wall of GPUs with Heterogeneous Memory Stack

  • Jeongmin Hong*, Sungjun Cho*, Gwangsun Kim

  • IEEE Computer Architecture Letters [*: Equal contribution] 

  • [ Paper ]

• Dynamic Global Adaptive Routing in High-radix Networks

  • Hans Kasan, Gwangsun Kim, Yung Yi, John Kim

  • The 49th International Symposium on Computer Architecture (ISCA) (Accept. rate: 16.8%)

  • [ Paper ]

2021

• Near-Data Processing in Memory Expander for DNN Acceleration on GPUs

  • Hyungkyu Ham*, Hyunuk Cho*, Minjae Kim, Jueon Park, Jeongmin Hong, Hyojin Sung, Eunhyeok Park, Euicheol Lim, Gwangsun Kim

  • IEEE Computer Architecture Letters [*: Equal contribution] 

  • [ Paper ]

2018

• TCEP: Traffic Consolidation for Energy-Proportional High-Radix Networks

  • Gwangsun Kim, Hayoung Choi, and John Kim

  • The 45th International Symposium on Computer Architecture (ISCA) (Accept. rate: 16.9%)

  • [ Paper ] [ Slides ]

2017

• Toward Standardized Near-Data Processing with Unrestricted Data Placement for GPUs

  • Gwangsun Kim, Niladrish Chatterjee, Mike O'Connor, and Kevin Hsieh

  • The International Conference for High Performance Computing, Networking, Storage and Analysis (SC) (Accept. rate: 18.7%)

  • [ Paper ] [ Slides ]

• History-Based Arbitration for Fairness in Processor-Interconnect of NUMA Servers

  • Wonjun Song, Gwangsun Kim, Hyungjoon Jung, Jongwook Chung, Jung Ho Ahn, Jae W Lee, and John Kim

  • The 22nd ACM International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS) (Accept. rate: 17.4%)

  • [ Paper ]

2016

• Contention-based Congestion Management in Large-Scale Networks

  • Gwangsun Kim, Changhyun Kim, Jiyun Jeong, Mike Parker, and John Kim

  • The 49th IEEE/ACM International Symposium on Microarchitecture (MICRO) (Accept. rate: 21.6%)

  • [ Paper ] [ Slides ]

• Automatically Exploiting Implicit Pipeline Parallelism from Multiple Dependent Kernels for GPUs

  • Gwangsun Kim, Jiyun Jeong, John Kim, and Mark Stephenson

  • The 25th International Conference on Parallel Architectures and Compilation Techniques (PACT) (Accept. rate: 26.1%)

  • [ Paper ] [ Slides ]

• Accelerating Linked-list Traversal through Near-Data Processing

  • Byungchul Hong, Gwangsun Kim, Jung Ho Ahn, Yongkee Kwon, Hongsik Kim, and John Kim

  • The 25th International Conference on Parallel Architectures and Compilation Techniques (PACT) (Accept. rate: 26.1%)

  • Nominated for the Best Paper Award

  • [ Paper ] [ Slides ]

• High-Throughput System Design with Memory Networks

  • Gwangsun Kim

  • Ph.D. Thesis, School of Computing, KAIST

  • [ Paper ]

• Transparent Offloading and Mapping (TOM): Enabling Programmer-Transparent Near-Data Processing in GPU Systems

  • Kevin Hsieh, Eiman Ebrahimi, Gwangsun Kim, Niladrish Chatterjee, Mike O’Connor, Nandita Vijaykumar, Onur Mutlu, and Stephen W. Keckler

  • The 43rd International Symposium on Computer Architecture (ISCA) (Accept. rate: 19.6%)

  • [ Paper ] [ Slides ]

• iPAWS : Instruction-Issue Pattern-based Adaptive Warp Scheduling for GPGPUs

  • Minseok Lee, Gwangsun Kim, John Kim, Woong Seo, Yeongon Cho, and Soojung Ryu

  • The 22nd IEEE International Symposium on High Performance Computer Architecture (HPCA) (Accept. rate: 22.1%)

  • [ Paper ]

• Design and Analysis of Hybrid Flow Control for Hierarchical Ring Network-on-Chip

  • Hanjoon Kim, Gwangsun Kim, Hwasoo Yeo, John Kim, and Seungryoul Maeng

  • IEEE Transactions on Computers, vol. 65, no. 2, pp. 480-494, 1 Feb. 2016

  • [ Paper ]

2015

• Overcoming Far-end Congestion in Large-Scale Networks

  • Jongmin Won, Gwangsun Kim, John Kim, Ted Jiang, Mike Parker, and Steve Scott

  • The 21st IEEE International Symposium on High Performance Computer Architecture (HPCA) (Accept. rate: 22.1%)

  • [ Paper ]

2014

• Multi-GPU System Design with Memory Networks

  • Gwangsun Kim, Minseok Lee, Jiyun Jeong, and John Kim

  • The 47th Annual IEEE/ACM International Symposium on Microarchitecture (MICRO) (Accept. rate: 19.4%)

  • [ Paper ] [ Slides ]

• Memory Network: Enabling Technology for Scalable Near-Data Computing

  • Gwangsun Kim, John Kim, Jung Ho Ahn, and Yongkee Kwon

  • The 2nd Workshop on Near-Data Processing (WoNDP) (in conjunction with MICRO-47)

  • [ Paper ] [ Slides ]

• Transportation-Network Inspired Network-on-Chip

  • Hanjoon Kim, Gwangsun Kim, Hwasoo Yeo, Seungryoul Maeng, and John Kim

  • The 20th International Symposium on High Performance Computer Architecture (HPCA) (Accept. rate: 25.6%)

  • [ Paper ]

• Low-overhead Network-on-Chip Support for Location-oblivious Task Placement

  • Gwangsun Kim, Michael M. Lee, John Kim, Jae W. Lee, Dennis Abts, and Michael Marty

  • IEEE Transactions on Computers, vol. 63, no. 6, pp. 1487-1500, June 2014

  • [ Paper ]

2013

• Memory-centric System Interconnect Design with Hybrid Memory Cubes

  • Gwangsun Kim, John Kim, Jung Ho Ahn, and Jaeha Kim

  • The 22nd International Conference on Parallel Architectures and Compilation Techniques (PACT) (Accept. rate: 17.3%)

  • Best Paper Award

  • [ Paper ] [ Slides ]

2012

• Scalable On-chip Network in Power Constrained Manycore Processors

  • Hanjoon Kim, Gwangsun Kim, and John Kim

  • The 3rd International Green Computing Conference (IGCC)

  • [ Paper ]

2011

• FlexiBuffer: Reducing Leakage Power in On-Chip Network Routers

  • Gwangsun Kim, John Kim, and Sungjoo Yoo

  • The 48th ACM/EDAC/IEEE Design Automation Conference (DAC) (Accept. rate: 22%)

  • [ Paper ] [ Slides ]

Teaching

• CSED503: Advanced Computer Architecture, POSTECH, Fall 2020, 2021, 2023

• CSED311: Computer Architecture, POSTECH, Spring 2019, 2020, 2021, 2022, 2023

• CSED490V: Parallel Architecture and Programming, POSTECH, Fall 2019, 2022

• CSED499: Research Project, POSTECH, Spring 2019

• CSED199: Freshman Research Participation, Fall 2021

Contact

• Email: g.kim at postech dot ac dot kr

• Phone: +82-54-279-2260

• Office: POSTECH R4 (연구4동) #4411, 67 Cheongam-ro, Nam-gu, Pohang, Gyungbuk, Korea 37673