Uniform memory access: Difference between revisions

Revision as of 04:33, 1 February 2008

Latest revision as of 00:00, 26 March 2025

(85 intermediate revisions by 67 users not shown)

Line 1:

{{Short description|Parallel computing memory architecture}}

{{misleading|date=November 2007}}

'''Uniform memory access''' ('''UMA''') is a [[shared-memory architecture]] used in [[parallel computer]]s. All the processors in the UMA model share the physical memory uniformly. In an UMA architecture, access time to a memory ___location is independent of which processor makes the request or which memory chip contains the transferred data. Uniform memory access computer architectures are often contrasted with [[non-uniform memory access]] (NUMA) architectures. In the NUMA architecture, each processor may use a private cache. Peripherals are also shared in some fashion. The UMA model is suitable for general purpose and [[time sharing]] applications by multiple users. It can be used to speed up the execution of a single large program in [[real-time computing|time-critical]] applications.<ref>{{cite book |title=Advanced Computer Architecture |author=Kai Hwang |isbn=0-07-113342-9}}</ref>

'''Uniform Memory Access''' (UMA) is a [[computer memory]] architecture used in [[parallel computer]]s having multiple [[processor]]s and probably multiple{{fact|date=November 2007}}<!-- is there no clear definition of UMA? --> [[Random access memory|memory chip]]s.

==Types of architectures==

All the processors in the UMA model share the physical memory uniformly. Peripherals are also shared. Cache memory may be private for each processor. In a UMA architecture, accessing time to a memory ___location is independent from which processor makes the request or which memory chip contains the target memory data. It is used in [[symmetric multiprocessing]] (SMP).{{fact|date=November 2007}}<!-- Does that imply, that ASMP never uses UMA? This would mean ASMP is always NUMA... -->

There are three types of UMA architectures:

* UMA using bus-based [[symmetric multiprocessing]] (SMP) architectures;

* UMA using [[crossbar switch]]es;

* UMA using [[multistage interconnection networks]].

==hUMA==

Uniform Memory Access computer architectures are often contrasted with [[Non-Uniform Memory Access]] (NUMA) architectures. UMA machines are, generally, harder for computer architects to design,{{fact|date=November 2007}}<!-- SMP with shared bus is harder to design than NUMA? --> but easier for programmers to program, than NUMA architectures.

In April 2013, the term hUMA (''heterogeneous uniform memory access'') began to appear in [[AMD]] promotional material to refer to [[CPU]] and [[GPU]] sharing the same system memory via [[Cache coherence|cache coherent]] views. Advantages include an easier programming model and less copying of data between separate memory pools.<ref>{{cite web |author=Peter Bright |url=https://arstechnica.com/information-technology/2013/04/amds-heterogeneous-uniform-memory-access-coming-this-year-in-kaveri/ |title=AMD's "heterogeneous Uniform Memory Access" coming this year in Kaveri |website=[[Ars Technica]] |date=April 30, 2013}}</ref>

==See also==

[[Category:Computer memory]]

* [[Non-uniform memory access]]

[[Category:Parallel computing]]

* [[Cache-only memory architecture]]

* [[Heterogeneous System Architecture]]

==References==

{{computer-stub}}

{{reflist}}

{{Parallel computing}}

[[de:Uniform Memory Access]]

[[Category:Computer memory]]

[[Category:Parallel computing]]