GPU

Communication Optimizations on Large-Scale GPU Clusters Using Rail Optimized Networks and NCCL PXN
대규모 언어모델을 효율적으로 수행하기 위해서는 여러 대의 GPU 와 CPU 간의 네트워크 통신 최적화가 필요하다. 본 논문에서는 기존의 네트워크 토폴로지 기반 클러스터 구성의 비효율성을 개선하기 위하여 대규모 GPU 클러스터 환경에서 Rail …
PDF Cite Project Poster
Communication Optimizations on Large-Scale GPU Clusters Using Rail Optimized Networks and NCCL PXN
MOST: Memory Oversubscription-Aware Scheduling for Tensor Migration on GPU Unified Storage
Deep Neural Network (DNN) training demands large memory capacities that exceed the limits of current GPU onboard memory. Expanding GPU …
PDF Cite Project Project DOI
MOST: Memory Oversubscription-Aware Scheduling for Tensor Migration on GPU Unified Storage
SSFFT: Energy-Efficient Selective Scaling for Fast Fourier Transform in Embedded GPUs
Fast Fourier Transform (FFT) is critical in applications such as signal processing, communications, and AI. Embedded GPUs are often …
PDF Cite Project Slides DOI Slide Show
SSFFT: Energy-Efficient Selective Scaling for Fast Fourier Transform in Embedded GPUs
Analysis of GEMV Kernel Computations by Address Mapping Approaches Based on GPU and PIM Architectures
Processing-in-Memory (PIM) is an architectural approach that can overcome bandwidth limitations between host processors and off-chip …
PDF Cite Project DOI
Analysis of GEMV Kernel Computations by Address Mapping Approaches Based on GPU and PIM Architectures
Hierarchical Traversal Stack Design Using Shared Memory for GPU Ray Tracing
Ray tracing is widely used to generate photorealistic images by tracing the paths of light rays through a scene and their interactions …
PDF Cite Project Slides DOI Slide Show
Hierarchical Traversal Stack Design Using Shared Memory for GPU Ray Tracing
Beyond VABlock: Improving Transformer Workloads through Aggressive Prefetching
The memory capacity constraint of GPUs is a major challenge in running large deep learning workloads with their ever increasing memory …
PDF Cite Project Project DOI
Beyond VABlock: Improving Transformer Workloads through Aggressive Prefetching
TLP Balancer: Predictive Thread Allocation for Multitenant Inference in Embedded GPUs
This letter introduces a novel software technique to optimize thread allocation for merged and fused kernels in multitenant inference …
PDF Cite Project DOI
TLP Balancer: Predictive Thread Allocation for Multitenant Inference in Embedded GPUs
VitBit: Enhancing Embedded GPU Performance for AI Workloads through Register Operand Packing
The rapid advancement of Artificial Intelligence (AI) necessitates significant enhancements in the energy efficiency of Graphics …
PDF Cite Project Slides DOI Slide Show
VitBit: Enhancing Embedded GPU Performance for AI Workloads through Register Operand Packing
Performance Analysis of GEMV Kernels by GPU and PIM Memory Address Mapping Approaches
Processing-in-Memory(PIM)은 프로세서와 오프칩(off-chip) …
PDF Cite Project Slides Slide Show
Performance Analysis of GEMV Kernels by GPU and PIM Memory Address Mapping Approaches
Conflict-Aware Compiler for Hierarchical Register File on GPUs
Modern graphics processing units (GPUs) leverage a high degree of thread-level parallelism, necessitating large-sized register files …
PDF Cite Project DOI
Conflict-Aware Compiler for Hierarchical Register File on GPUs