high performance computing on graphics processing units: hgpu.org

hgpu.org » Performance

Microarchitectural Memory Bandwidth Saturation, KV-Cache Paging Dynamics, and Time-to-First-Token Latency: A Comparative Benchmark of vLLM, TensorRT-LLM, and FlashAttention-3 on NVIDIA Hopper H100 versus AMD Instinct MI300X

Kartik Kothalkar

View

Tags: AMD Radeon Instinct MI300X, ATI, Benchmarking, Computer science, CUDA, LLM, nVidia, nVidia H100, Performance, ROCm

September 28, 2026 by hgpu

Xtrace: High-Fidelity GPU Intra-Kernel Tracing via Binary-Level Instruction Splicing

Zhuobin Huang, Kai Zhang, Weihao Cui, Hongshi Tan, Liang Luo, Christopher Dewan, Shen Li, Bingsheng He

View

Download (PDF)

Tags: AMD Radeon Instinct MI300X, ATI, Computer science, LLM, nVidia, nVidia B300, nVidia H100, Performance

September 28, 2026 by hgpu

DeepSeek-V4-Flash on AMD gfx90a: Correctness Recovery and Inference Performance Engineering

Siming Huang

View

Download (PDF)

Tags: AMD, AMD Radeon Instinct MI250, Computer science, HIP, LLM, Performance

September 20, 2026 by hgpu

AutoTuneBench: Trustworthy Measurement for Agent Auto-Tuning of LLM Serving Engines

Li Chen

View

Download (PDF)

Source codes

Tags: Computer science, CUDA, LLM, nVidia, nVidia GB10, Package, Performance

September 20, 2026 by hgpu

PrefixBench-H100: Characterizing Prefix Reuse and Time-to-First-Token in H100 LLM Serving

Omkar Shewale, Deepak Kumar, Divakar Kumar Yadav

View

Download (PDF)

Source codes

Tags: Computer science, CUDA, LLM, nVidia, nVidia H100, Package, Performance

September 20, 2026 by hgpu

Concurrency Response of Plain Global Loads on the NVIDIA H100

Somashekar Manjunath, Rahul Ramachandra M

View

Download (PDF)

Tags: Computer science, CUDA, Hardware Architecture, nVidia, nVidia H100, Performance

August 30, 2026 by hgpu

Enhancing the Performance Analysis of NCCL GPU Collectives

Jurij Cerar

View

Download (PDF)

Tags: Computer science, CUDA, nVidia, nVidia H100, Performance, Tesla T4, Thesis

July 13, 2026 by hgpu

Fearless Concurrency on the GPU

Melih Elibol, Jared Roesch, Isaac Gelado, Eric Buehler, Michael Garland

View

Download (PDF)

Tags: Computer science, CUBLAS, CUDA, nVidia, nVidia B200, nVidia GeForce RTX 5090, Performance, Python, Rust

June 17, 2026 by hgpu

KEET: Explaining Performance of GPU Kernels Using LLM Agents

Joshua H. Davis, Klaudiusz Rydzy, Srinivasan Ramesh, Aadit Nilay, Daniel Nichols, Swapna Raj, Nikhil Jain, Abhinav Bhatele

View

Download (PDF)

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

May 11, 2026 by hgpu

Revealing NVIDIA Closed-Source Driver Command Streams for CPU-GPU Runtime Behavior Insight

Yuang Yan, Ian Karlin, Ryan Grant

View

Download (PDF)

Tags: Computer science, CUDA, nVidia, nVidia A40, Performance

May 3, 2026 by hgpu

Evaluating CUDA Tile for AI Workloads on Hopper and Blackwell GPUs

Divakar Kumar Yadav, Tian Zhao, Deepak Kumar

View

Download (PDF)

Source codes

Tags: Computer science, CUBLAS, CUDA, LLM, nVidia, nVidia B200, nVidia H100, nVidia RTX PRO 6000, Package, Performance, Triton

May 3, 2026 by hgpu

KernelSkill: A Multi-Agent Framework for GPU Kernel Optimization

Qitong Sun, Jun Han, Tianlin Li, Zhe Tang, Sheng Chen, Fei Yang, Aishan Liu, Xianglong Liu, Yang Liu

View

Download (PDF)

Source codes

Tags: Computer science, CUDA, LLM, nVidia, nVidia A100, Package, Performance, PyTorch

March 15, 2026 by hgpu

Brian2 Atlas

A Unified Intermediate Representation and Execution Architecture for Heterogeneous and Distributed Neural Simulation

AgentGarten: Code Worlds for Evolving Agents

KernelLens: Automated Triton-to-C++ Compiler for Enterprise Inference Runtimes

KernelLens: Deploying Triton Kernels to C++ Inference Engines

KernelZero: Co-Evolving Proposer and Coder for Continuously Improved GPU Kernel Generation

Introduction to Computer Vision

ACT: Accelerator Compiler Toolkit

Automatically Generating ML Compiler Backends from Tensor Accelerator ISA Descriptions

pytest-gpu-proof: A Simple Pytest Extension for Local GPU Code Testing

pytest-gpu-proof: Enabling Cloud-CPU Continuous Integration for GPU Code with Local GPU Attestation

Accelerating the Mitigation of LLM Inference Nondeterminism Across GPU Architectures

AutoTuneBench: Trustworthy Measurement for Agent Auto-Tuning of LLM Serving Engines

nvsass-disassembler

Automated Instruction Encoding Synthesis for Modern GPU ISA Compression

See all packages

* * *

high performance computing on graphics processing units: hgpu.org

Microarchitectural Memory Bandwidth Saturation, KV-Cache Paging Dynamics, and Time-to-First-Token Latency: A Comparative Benchmark of vLLM, TensorRT-LLM, and FlashAttention-3 on NVIDIA Hopper H100 versus AMD Instinct MI300X

Xtrace: High-Fidelity GPU Intra-Kernel Tracing via Binary-Level Instruction Splicing

DeepSeek-V4-Flash on AMD gfx90a: Correctness Recovery and Inference Performance Engineering

AutoTuneBench: Trustworthy Measurement for Agent Auto-Tuning of LLM Serving Engines

PrefixBench-H100: Characterizing Prefix Reuse and Time-to-First-Token in H100 LLM Serving

Concurrency Response of Plain Global Loads on the NVIDIA H100

Enhancing the Performance Analysis of NCCL GPU Collectives

Fearless Concurrency on the GPU

KEET: Explaining Performance of GPU Kernels Using LLM Agents

Revealing NVIDIA Closed-Source Driver Command Streams for CPU-GPU Runtime Behavior Insight

Evaluating CUDA Tile for AI Workloads on Hopper and Blackwell GPUs

KernelSkill: A Multi-Agent Framework for GPU Kernel Optimization

Recent source codes

Brian2 Atlas

AgentGarten: Code Worlds for Evolving Agents

KernelLens: Automated Triton-to-C++ Compiler for Enterprise Inference Runtimes

KernelZero: Co-Evolving Proposer and Coder for Continuously Improved GPU Kernel Generation

Introduction to Computer Vision

ACT: Accelerator Compiler Toolkit

pytest-gpu-proof: A Simple Pytest Extension for Local GPU Code Testing

Accelerating the Mitigation of LLM Inference Nondeterminism Across GPU Architectures

AutoTuneBench: Trustworthy Measurement for Agent Auto-Tuning of LLM Serving Engines

nvsass-disassembler

Most viewed papers (last 30 days)