From Wikipedia, the free encyclopedia
Jump to: navigation, search
WikiProject Computing / Hardware (Rated Start-class, Mid-importance)
WikiProject icon This article is within the scope of WikiProject Computing, a collaborative effort to improve the coverage of computers, computing, and information technology on Wikipedia. If you would like to participate, please visit the project page, where you can join the discussion and see a list of open tasks.
Start-Class article Start  This article has been rated as Start-Class on the project's quality scale.
 Mid  This article has been rated as Mid-importance on the project's importance scale.
Taskforce icon
This article is supported by Computer hardware task force (marked as High-importance).


Whoever originally wrote the first section sucks at researching things. Misuse of the word link was abundant. They should have read the overview and the specifications of the actual technology found here. I mean, RETARD! I've come back to this page so many times because it made so little sense. I finally got the intelligence to look it up somewhere not editable by the public. Now it is clear as day. I hope my touch ups help somebody else who has had the same issue with that section. — Preceding unsigned comment added by (talk) 21:51, 17 September 2011 (UTC)


The HyperTransport website claims up to about 12 GB/s transfers, does that mean that it has been expanded to 64-bit (32-bit each way)? —Mulad 20:32 9 Jun 2003 (UTC)

Latency of Hypertransport[edit]

I am wondering whether there's any white paper out there that gives a number in nano seconds about the latency of the HT. It is somehow strange that everybody is talking just about the bandwith. The same is true for the QPI from Intel. To compare both, shouldn't latency matter very much? Cheers Ned

HT as an FSB replacement[edit]

Does HT qualify as an FSB replacement? My understanding (and that of the FSB article) is that an FSB is the connection between the CPU and the MCH, which is not what HT serves as - it serves as an external connector for the CPU to a device of any desired functionality with an HT link (could be a Northbridge/MCP, could be another K8, could be a co-processor as per cHT links...). I want to strike "HT as an FSB replacement" and replace with "HT as an FSB augmentation." HT does not carry data to memory, which is the defining charachteristic of that particular bus (otherwise it becomes "just another interconnect out of the CPU" so to speak).

Actually I wouldn't say that carrying data to memory is the defining feature of a FSB, instead I would say that a FSB is the main connection from a CPU to the rest of the world. The FSB article itself says as much when it lists the RAM as just one of the things that are connected to over a FSB. I would rather see this subtlety added to the paragraph. Perhaps something like: Many of the processors which utilize HyperTransport including the Opteron and the BCM1480 have an integrated memory controller. In these uses, HyperTransport does not serve as the local connection to memory as in the classical definition of a FSB. However, the HyperTransport does serve as the connection to the memory directly connected to other Opterons (or BCM1480's) when these devices are used in NUMA applications. Brholden 05:33, 16 August 2006 (UTC)

HT is not a bus[edit]

"An electrical bus (sometimes spelled buss) is a physical electrical interface where many devices share the same electric connection. This allows signals to be transferred between devices (allowing information or power to be shared). A bus often takes the form of an array of wires that terminate at a connector which allows a device to be plugged onto the bus." from Wikipedia article bus.

HT is a point-to-point serial interconnect, not a bus. A bus is inherrently parallel, (if my understanding is correct) and shared (HT is not is point to point). With consensus (or after three days) I will change all references to "HT bus" in this article to "HT Interconnect."

The HT site itself calls it a bus...[1] Can more than one HT link be called a bus? -Ravedave 22:14, 19 July 2006 (UTC)
HyperTransport lives in the gray area between serial and parallel interconnects. It uses a mixture of the techniques of both. It is fine to remove some of the uses of "bus", but the one to not remove is as its use as a Front side bus of microprocessors. HyperTransport use as the front side bus of the Opteron/Athlon/Turion microprocessors is the most important application of HyperTransport. The term Front side bus has the broader meaning as the main interface coming out of a microprocessor, independent of its specific implementation. If the word "bus" was removed from this spot, the article would lose meaning. Less important, but along the same lines, the term Computer bus has the additional meaning of a memory-mapped interface independent of the specifics of the interface. (forgot to sign this a while back Brholden 05:33, 16 August 2006 (UTC))
The article Front side bus concedes that HT is not an FSB "Not technically a front side bus." I understood that FSB was the connection between the CPU and MCH, and there was a lot of dicsussion during the unveiling of the K8 marchitechture that indicated the "FSB" (as it were) would be on-die. The FSB article links to a back side bus article, which makes reference to connections between the CPU and cache, but given that storage is hierachachal, I don't believe this is accurate (in a theoretical sense, not necessarily an electrical one). However, I think that because electrical engineering terminology reflects a very definitive and static reality, it's important that we (as an excyclopedia) are absolutely clear on these topics. Gray area is very bad.
Yes, and no. An alternate definition of 'bus' is often a set of signals that go from one IC to another. So it may be one "data line" (well, in HT that line is two physical differential traces but I digress) along with its control signals, and we could call it a bus. It gets into grey areas when you start bonding serial channels together into a "bus" but you have no exact data phase alignment between the channels. Is that a bus? I'd say yes - it has many of the similar restrictions as a traditional bus, but some are more relaxed. I just tweaked the article a touch, avoiding the word bus. — RevRagnarok Talk Contrib 03:28, 16 August 2006 (UTC)


Just for Info : HT is the Trademark for Hyper Transport HTT or HT-Tech or HT-Technology is the Trademark for Hyper Threading Technology Denniss 10:23, 14 Dec 2004 (UTC)

Link width correction[edit]

Yes, the maximum supported link width is 32-bit in each direction.

I did some slight changes to the article.

--Eltiel 03:45, 6 September 2005 (UTC)

Added "Overview" header[edit]

This is a nice article about a nice technology. But (couldn't you tell that there was a "but" coming?) the first section is rather long, putting the contents box well down the page. So I added an "Overview" header after the first paragraph. I hope that this makes the article even better. Chris Chittleborough 07:03, 10 November 2005 (UTC)

HyperTransport an SGI technology?[edit]

Several articles I have read claim that hypertransport was originally an sgi technology. Is there any basis to this? I will not add to this article but I think that if proven to be correct, sgi should be given credit in the article. -- 21:53, 29 December 2005 (UTC)

It was based on the Digital Equipment Corporation's EV7 bus architecture. [2] —Preceding unsigned comment added by (talkcontribs)


I know that there are 3 versions (1, 2, and amazingly 3) of Hyper transport, they should probably be in the article. -Ravedave 04:35, 17 May 2006 (UTC)

This seems to have been corrected (see beginning of Overview) Ppchailley 10:45, 30 June 2006 (UTC)

Intel HyperTransport[edit]

I thought that Intel is already using HyperTransport via certain nVidia chipsets?

Check out the following pages. —The preceding unsigned comment was added by Ofunniku (talkcontribs) 13:29, 9 March 2007 (UTC).

A number of non-Intel chipsets use HyperTransport between the devices of the chipset. Some of these chipsets can be used with Intel processors. Brholden 18:47, 9 March 2007 (UTC)
I'm sorry. I forgot to sign my previous post. I'm kinda new to the discussion pages on wikipedia :) So, will this information be written to the main article? I think it's pretty important. -Ofunniku 5:11, 29 March 2007 (UTC)

Intel does use hypertransport but not as a replacement for the front side bus. It is used occasionally for linking dual and multicore processors together.

Do you have any sources for this claim ? --Denniss 02:31, 19 June 2007 (UTC)
That is highly unlikely. The only instace that HyperTransport ever appeared on Intel platforms is on nVidia chipsets, as the recent nForce chipsets are designed with HyperTransport in mind. Their Intel-compatible chipsets retain most of the features found on the AMD side. Ofunniku 13:06, 27 July 2007 (UTC)

Server makers[edit]

I'm not sure listing server makers in this article makes much sense. I'm pretty sure if you're building a server that uses AMD Opteron processors, you have to use Hypertransport, and this article is probably not the best place to try to maintain a complete listing of everyone who makes servers that take Opteron processors. For that matter, I think this article would need to list everyone who makes a motherboard that takes Athlon64 and Athlon64 X2 and Athlon64 X4 processors, too (Asus, Tyan, and HP come to mind as examples who make motherboards / desktop PCs that I've personally worked with that I'm pretty sure have Hypertransport.)

So I think it would be better to just make it clear that using those processors effectively requires Hypertransport, and the articles about those processors can contain lists of manufacturers if there's a need to make such a list. JNW2 03:00, 31 October 2007 (UTC)

re list[edit]

agreed that this is not the place for a comprehensive list of HT server manufacturers, but feel that including a few examples for the sake of illustration is OK as long as it's acompanied by a link to a more compehensive list elsewhere. —Preceding unsigned comment added by (talk) 13:11, 28 November 2007 (UTC)

where's the switch[edit]

the article seems to indicate that HT works like a packet switched network but it's not clear how or where the switching takes place. explanaiton welcome. —Preceding unsigned comment added by (talk) 13:13, 28 November 2007 (UTC)

of course if no switch is required an explanation of why not would be fitting.

put differently, if HT works like a network an explanation of the network topology would be helpful. —Preceding unsigned comment added by (talk) 15:01, 28 November 2007 (UTC)

comparison to other interconnects?[edit]

might be interesting to add some comparisons to other high performance cpu/memory interconnect architectures (ie, the DEC interconnect mentioned above, SGI's ccNUMA system, sun's fireplane, HP superdome, the crossbar switches that have appeared in some RISC workstations, etc). —Preceding unsigned comment added by (talk) 13:21, 28 November 2007 (UTC)

Added link to new article on Elastic interface bus--Nowa (talk) 01:55, 21 July 2008 (UTC)

Fair use rationale for Image:HyperTransport logo resized.jpg[edit]

Nuvola apps important.svg

Image:HyperTransport logo resized.jpg is being used on this article. I notice the image page specifies that the image is being used under fair use but there is no explanation or rationale as to why its use in this Wikipedia article constitutes fair use. In addition to the boilerplate fair use template, you must also write out on the image description page a specific explanation or rationale for why using this image in each article is consistent with fair use.

Please go to the image description page and edit it to include a fair use rationale. Using one of the templates at Wikipedia:Fair use rationale guideline is an easy way to insure that your image is in compliance with Wikipedia policy, but remember that you must complete the template. Do not simply insert a blank template on an image page.

If there is other fair use media, consider checking that you have specified the fair use rationale on the other images used on this page. Note that any fair use images lacking such an explanation can be deleted one week after being tagged, as described on criteria for speedy deletion. If you have any questions please ask them at the Media copyright questions page. Thank you.

BetacommandBot (talk) 23:10, 22 December 2007 (UTC)

HyperTransport being compared to PCI?[edit]

Is there any reason that the article compares the clockspeeds of HyperTransport to PCI? Even if that was a fair comparison, why not compare it to a more contemporary connection such as PCI-E? —Preceding unsigned comment added by (talk) 05:29, 2 June 2008 (UTC)

IBM PPC970 Chipset Apple note edit dispute[edit]

This text was added to the Line 65 notation of IBM's PPC970 chipsets as implementations:

as co-designed and used by Apple in the Power Mac G5

This citation doesn't belong in the article. The chipsets were included as IBM, an editor subsequently added a citation which indicated Apple had assisted IBM in design work. While this information is true, it 1) appears in the article for PPC970 (the only other place where the chipsets are mentioned in detail) 2) is excessive detail for a simple list of implementations (which is to say, the purpose of the list is simply to denote, not to describe. For detail on the implementation, one should follow the link onward to the complete article on the implementing device) and 3) isn't relevant to the article itself (adds no information about Hypertransport) or the section of the article (implementations).

The editor who added the quote inappropriately engaged in an edit-dispute with another editor by re-doing changes instead of proceeding to the article's discussion page to settle the issue through intelligent discussion.

I will remove the citation above on 18 October at 24:00 UTC unless further conversation builds an opposite consensus.Sahrin (talk) 13:34, 17 October 2009 (UTC)

Eh, no big deal. If two people really don't want it there and one does, the one loses. (talk) 09:55, 18 October 2009 (UTC)
This is actually not how we work. Our goal is to build an information consensus and act with unified intention - not to conduct a vote. If you have a different perspective, please share it with us so that we may enhance our understanding. Sahrin (talk) 15:34, 18 October 2009 (UTC)
Well, I originally hadn't put much consideration into the appropriateness of adding this note to this particular section of the article. I was actually more surprised than anything that neither Apple (one of the HT Consortium co-founders) nor the Power Mac G5 (one of the first practical applications of the technology) had been mentioned anywhere in the article. Even the article on the Power Mac G5 itself still lacks information on its use of HyperTransport, which I find unusual.
I'd be much more satisfied with a dedicated section under Applications for HyperTransport, describing its use in the G5 in more detail, but it's beyond my expertise level on the subject to add it myself. For example, what little research I have conducted shows me, a layman, that Apple's implementation does not fall squarely into any of the existing sections there, and I would not even know what to title it. That's why I initially took the shortcut route of inserting that note at the first mention of related information.
Anyway, you can remove it, and I won't re-add it. (talk) 22:51, 18 October 2009 (UTC)

Asterisk in HT Frequency Specs section.[edit]

Under HT f Specs section, their is an asterisk at the end of one of the column headers in the table: Max. Bandwidth at 32-Bit unidirectional* There does not seem to be a reason for this that I could find. No other mention of unidirectional or 32 bit or anything has an asterisk, and there doesn't seem to be a corresponding one elsewhere on the page to indicate a footnote or additional information regarding the asterisked-text. There is a bulleted block of text below the table in question, but it doesn't seem to relate clearly to the asterisked-column, though I could be wrong as I don't understand this material well. Could somebody with a better grasp of things than me figure out what the asterisk is for and make it more clear if necessary? Thanks. ************************************************--Δζ (talk) 21:14, 14 March 2010 (UTC)

GiB vs GB[edit]

Fixed the misapplied GiB in link bandwidth. Since the GHz measure is 10^9, the resulting bandwidth measure is also 10^9, not 2^30. Plaidomatic (talk) 23:34, 27 September 2011 (UTC)

Lead is presently a mess[edit]

The current specification HTX3.1 remains competitive for 2014 high speed ...

The text is hard to read (there's at least a missing parenthesis, in the vicinity of an embedded hard link and a perhaps poorly judged mdash), which unfortunately only helps to obscure that this paragraph goes trotting off to talk about front-side bus issues, without the predominant FSB application of HT having yet been tabled. — MaxEnt 22:01, 18 June 2015 (UTC)

Incorrect and Misleading Bandwidth calculation[edit]

The section Links and rates reads: "theoretical transfer rate is 25.6 GB/s (3.2 GHz × 2 transfers per clock cycle × 32 bits per link) per direction, or 51.2 GB/s aggregated throughput"

The bandwidth of a system bus is calculated as: (Frequency x bus width). In the aforementioned calculation, the formula should read

3.2 / 2 * 32, not, 3.2 * 2 * 32

The frequency 3.2 GHz is assuming the DDR connection is included in the real frequency, as is when RAM DIMM speeds are published on retail boxes.

This is proven using product and quotient provided in the sentence. In order to achieve a bi-directional (aggregated) max bandwidth of 51.2 GB/s, with a 32 bit wide link, a clock frequency of 1.6 GHz (3.2/2) is required. As 1.6 * 32 = 51.2, leaving a unidirectional bandwidth of 25.6 GB/s.

So, the results that are currently stated, show the bandwidth of a HyperTransport link with a real clock frequency of 1.6 GH/z.

In conclusion, if the article wishes to retain the results of 51.2 GB/s for bi-directional max bandwidth, unidirectional max bandwidth being 25.6 GB/s, the formula needs to be changed. Or, list the real clock frequencies, half of what they currently are. If the formula wishes to reflect the bandwidth utilizing the fact that HyperTransport is a DDR connection, then the results need updating.

Assuming DDR clock frequencies:

Frequency with DDR = (real_frequency x 2)

Frequency with DDR = 1.6 x 2 = 3.2

Max bi-directional bandwidth w/DDR = (3.2 x 32) = 102.4 GB/s (aggregated) Max unidirectional bandwidth w/DDR = (max_bi-directional_bandwidth_with_DDR / 2) = 102.4 / 2 = 51.2 GB/s