high performance computing on graphics processing units: hgpu.org

hgpu.org » LLM

Breaking the Memory Wall: A Study of I/O Patterns and GPU Memory Utilization for Hybrid CPU-GPU Offloaded Optimizers

Avinash Maurya, Jie Ye, M. Mustafa Rafique, Franck Cappello, Bogdan Nicolae

View

Tags: Computer science, CUDA, LLM, Memory, nVidia, nVidia A100, nVidia H100, Performance

June 23, 2024 by hgpu

Kokkidio: Kokkos+Eigen - write fast, readable multi-platform code

Kokkidio: Fast, expressive, portable code, based on Kokkos and Eigen

LLaMA-Inference-Bench

LLM-Inference-Bench: Inference Benchmarking of Large Language Models on AI Accelerators

MATAR

On a Simplified Approach to Achieve Parallel Performance and Portability Across CPU and GPU Architectures

Over-synchronization in GPU Programs

Over-synchronization in GPU Programs

profile_util: simple C++ library to report memory, timing, thread affinity

Profile Util library: A quick and easy way to get MPI, OpenMP and GPU runtime information

MaC: Mamba CPU performance predictor

MambaCPU: Enhanced Correlation Mining with State Space Models for CPU Performance Prediction

Mixed-precision finite element kernels

Mixed-precision finite element kernels and assembly: Rounding error analysis and hardware acceleration

SHMT: The Simultaneous and Heterogenous Multithreading project

General-Purpose Computing on Tensor Processors

Superpipeline: A Universal Approach for Reducing GPU Memory Usage in Large Models

Superpipeline: A Universal Approach for Reducing GPU Memory Usage in Large Models

EnergyUCB-Bandit

Online Energy Optimization in GPUs: A Multi-Armed Bandit Approach

See all packages

* * *

* * *

HGPU group © 2010-2024 hgpu.org

All rights belong to the respective authors

Login | Sitemap | Feedback | Policy

Contact us: