User Guide

Intel® VTune™ Profiler User Guide

ID 766319
Date 7/13/2023
Public

A newer version of this document is available. Customers should click here to go to the newest version.

Document Table of Contents

Untyped Reads Coalescence

Metric Description

Transaction Coalescence is a ratio of the used bytes to all bytes requested by the transaction. The lower the coalescence, the bigger part of the bandwidth is wasted. It originates from the GPU Data Port function that dynamically merges scattered memory operations into fewer operations over non-duplicated 64-byte cacheline requests. For example, if a 16-wide SIMD operation consecutively reads integer array elements with a stride of 2, the coalescence of such a transaction is 50%, because half of the bytes in the requested cacheline is not used.