GPU Glossary

TABLE OF CONTENTS

Device Hardware

CUDA (Device Architecture)

Streaming Multiprocessor

Special Function Unit

Load/Store Unit

Tensor Memory Accelerator

Streaming Multiprocessor Architecture

Texture Processing Cluster

Graphics/GPU Processing Cluster

Device Software

CUDA (Programming Model)

Streaming ASSembler

Parallel Thread eXecution

Compute Capability

Cooperative Thread Array

Thread Block Grid

Thread Hierarchy

Memory Hierarchy

CUDA (Software Platform)

CUDA C++ (programming language)

NVIDIA GPU Drivers

CUDA Driver API

NVIDIA Management Library

CUDA Runtime API

NVIDIA CUDA Compiler Driver

NVIDIA Runtime Compiler

NVIDIA CUDA Profiling Tools Interface

NVIDIA Nsight Systems

CUDA Binary Utilities

Performance Bottleneck

Arithmetic Intensity

Memory Bandwidth

Arithmetic Bandwidth

Warp Execution State

Pipe Utilization

Issue Efficiency

Streaming Multiprocessor Utilization

Warp Divergence

Branch Efficiency

Memory Coalescing

Register Pressure

/perf/issue-efficiency

?

Something seem wrong?
Or want to contribute?

Click this button to
let us know on GitHub.

What is issue efficiency?

Issue efficiency measures how effectively the warp scheduler keeps execution pipes busy by issuing instructions from eligible warps .

An issue efficiency of 100% means every scheduler issued an instruction on every cycle, indicating at least one eligible warp on each cycle. Values below 100 % signal that, during some cycles, all active warps were stalled - waiting on data, resources, or dependencies - so the scheduler sat idle and overall instruction throughput fell.

Building on GPUs? We know a thing or two about it.

Modal is an ergonomic Python SDK wrapped around a global GPU fleet. Deploy serverless AI workloads instantly without worrying about quota requests, driver compatibility issues, or managing bulky ML dependencies. Deploy on GPUs

Peak Rate Streaming Multiprocessor Utilization