Talk:Simultaneous multithreading

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia

[Untitled][edit]

"In 2005, [Pentium 4 multithreading] security concerns were made public by Colin Percival, Cache missing for fun and profit demonstrating that malicious threads can monitor the execution of other threads."

I think this is an interesting fact that User:62.15.117.39 brought up in this revision, so am putting it here. But I agree with User:69.134.163.109 that it's not necessarily a detail for an encyclopedia article. -- Furchild 22:11, Jun 26, 2005 (UTC)

Oh. Another link that is more informative: Hyper-Threading Considered Harmful. -- Furchild 22:59, Jun 26, 2005 (UTC)

NB: SMT can always lower performance by forcing a thread to share resources. Imagine the situation of a thread which requires all ROB entries for a modern MPU to execute without stalls, the minute SMT is turned on, it will begin stalling. SMT *usually* increases performance, but it is not guaranteed and can lower performance

Also, there are no MPUs that have shipped with more than 2 simultaneous threads (POWER5 and Pentium 4). Other MPUs from Sun and Raza Microelectric use coarse grained multithreading. Since instructions cannot execute from multiple threads at the same time, it is not simultaneous.

Montecito[edit]

Montecito uses SMT: http://www.doc.ic.ac.uk/~phjk/AdvancedCompArchitecture/PastPapers/2003-2004-MEng3Test.pdf you can read at 1.b: "In a proposed simultaneous multi-threading (SMT, also known as hyperthreaded) Itanium 2 design, a single CPU core is extended with two program counters, and two register sets, so that it can execute two different threads at the same time. Referring to Figure 1 (page 46), identify which parts of the design would have to be changed, and explain briefly what would have to be done." --134.155.99.41 07:53, 24 November 2006 (UTC)[reply]

"The processor was never released, since the Alpha line of processors was discontinued when Compaq acquired DEC."

Compaq bought DEC in 1998 and Alpha was discontinued in 2001, so saying the above is grossly misleading.

Diamondville isn't out of order[edit]

As The Register points out

http://www.reghardware.co.uk/2008/04/03/idf_inside_silverthorne/page2.html

Intel is launching a new type of CPU that uses HyperThreading in place of out-of-order execution so some mention should be made about this.

Hcobb (talk) 18:01, 22 April 2008 (UTC)[reply]

Another early example[edit]

I haven't the references for this but as far as I know the ICL Series 39 machines had a 'B-pipe' which was exclusively for system use. It did the initial handling of interrupts, scheduled task switches, and did virtual to real address translations for the I/O. Individual micro instructions were taken from the B-pipe when there was a gap in the main pipe - or when necessary for example after an interrupt it happened the other way round. Dmcq (talk) 10:05, 28 September 2008 (UTC)[reply]

Article needs reorganization?[edit]

I've seen SMT presented in a few different courses and textbooks, and what they usually do is view it from the superscalar utilization angle, define two kinds of waste - "horizontal" and "vertical", and then show how SMT reduces both kinds of waste. IMHO, this presentation is easier to understand. (This is also the method that Tullsen, Eggers and Levy adopt in their seminal paper introducing SMT)

I'd like to re-organize the article in that fashion. Are there any thoughts/comments/concerns on this? --Pramod 06:34, 3 January 2009 (UTC) —Preceding unsigned comment added by Pramod.s (talkcontribs)

I have added a section that briefly describes the disadvantages of SMT, but I agree that a thorough rewrite of the article would be good. Now that multi-core processors have become so common, it needs to be explained more clearly what the difference is between having two threads in the same processor core or two cores on the same processor chip. Maybe the term “processor“ is ambiguous here, as some people call the whole chip a processor? Afog (talk) 18:41, 4 November 2009 (UTC)[reply]

Cache Thrashing Description?[edit]

Does this really have anything to do with multithreading?

"Programs written before multiprocessor and multicore designs were prevalent commonly did not optimize cache access because on a single CPU system there is only a single cache which is always coherent with itself. On a multiprocessor system each CPU or core will typically have its own cache, which is interlinked with the cache of other CPU/cores in the system to maintain cache coherency. If thread A accesses a memory location [00] and thread B then accesses memory location [01] it can cause an intercache transaction particularly where the cache line fill exceeds 2 bytes, as is the case for all modern processors."

If you have a single core implementation with multithreading then you have "a single cache which is always coherent with itself" and the description of thrashing between multiple caches is moot. And if you have a multiprocessor implementation even without multithreading then you have this problem. As such, it seems more like it belongs in a multiprocessor and/or coherency article.

75.25.154.173 (talk) 23:47, 4 June 2010 (UTC)[reply]

MIPS multithreading not SMT[edit]

The MIPS 34K and 1004K soft CPUs implement a type of barrel-processing MT. These CPUs are single-issue (not superscalar) so they do not implement SMT. Dyl (talk) 03:08, 12 October 2010 (UTC)[reply]

Disambiguation[edit]

A disambiguation page should be created for the abbreviation SMT. I'd appreciate if anyone more experienced than me could do that. — Preceding unsigned comment added by Guissoares (talkcontribs) 20:42, 15 March 2011 (UTC)[reply]

Other implementations[edit]

How about adding:

SPARC64 VI (2 cores with 2-way VMT+CMP), SPARC64 VII (4 cores with 2-way SMT), SPARC64 VII+ (4 cores with 2-way SMT). Newer designs by Fuijtsu for HPC applications doesn thave SMT: SPARC64 VIIIfx (8 cores without SMT, but with HPC extensions), SPARC64 IXfx (16 cores without SMT)

Intel Larrabe GPU (4 or 8-way SMT)

All GPUs since introduction of programmable shaders have SMT built-in (with as high as 512 threads per core). — Preceding unsigned comment added by 91.213.255.7 (talk) 16:15, 7 December 2011 (UTC)[reply]

efficiency solution[edit]

This term is used in this article without clearly explaining what it means. (What is being optimized here? Cost, power, size, time?) Hcobb (talk) 01:18, 20 January 2013 (UTC)[reply]

Cleanup[edit]

Hey everyone. What type of cleanup this article requires? InfocenterM (talk) 13:50, 29 July 2013 (UTC)[reply]

The section "Modern commercial implementations" has very few references. JeffV 15:40 02 October 2013 (UTC)

External links modified[edit]

Hello fellow Wikipedians,

I have just added archive links to 2 external links on Simultaneous multithreading. Please take a moment to review my edit. If necessary, add {{cbignore}} after the link to keep me from modifying it. Alternatively, you can add {{nobots|deny=InternetArchiveBot}} to keep me off the page altogether. I made the following changes:

When you have finished reviewing my changes, please set the checked parameter below to true or failed to let others know (documentation at {{Sourcecheck}}).

This message was posted before February 2018. After February 2018, "External links modified" talk page sections are no longer generated or monitored by InternetArchiveBot. No special action is required regarding these talk page notices, other than regular verification using the archive tool instructions below. Editors have permission to delete these "External links modified" talk page sections if they want to de-clutter talk pages, but see the RfC before doing mass systematic removals. This message is updated dynamically through the template {{source check}} (last update: 18 January 2022).

  • If you have discovered URLs which were erroneously considered dead by the bot, you can report them with this tool.
  • If you found an error with any archives or the URLs themselves, you can fix them with this tool.

Cheers.—cyberbot IITalk to my owner:Online 06:16, 28 February 2016 (UTC)[reply]