Talk:AMD Accelerated Processing Unit

From Wikipedia, the free encyclopedia
Jump to: navigation, search
WikiProject Computing / Hardware (Rated C-class, Low-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.
C-Class article C  This article has been rated as C-Class on the project's quality scale.
 Low  This article has been rated as Low-importance on the project's importance scale.
Taskforce icon
This article is supported by Computer hardware task force (marked as Mid-importance).


Please expand on my information about the fusion GPU/CPU processor if you can. --Xernous

The information on Richland being part of the 2012 APU line is not right. Richland is a chip scheduled for Q1/Q2 2013, which can be seen as a trinity 2.0. Kaveri is slated for something like Q4 2013 and could slip to 2014. I am worried this page actually deters consumers from buying fusion, because it is so confusingly arranged. — Preceding unsigned comment added by (talk) 13:21, 27 January 2013 (UTC)

The information on the left bar at the top of the article confuses me. It looks like a fusion apu is a gpu, so much info on the gpu models, but almost no infor on the cpu models/specs. If i only looked at that bar i would think Fusion was just a gpu. — Preceding unsigned comment added by (talk) 13:41, 27 January 2013 (UTC)

GPU refresh rate[edit]

GPUs in computers tend to be upgraded two to four times more often then the processor, so it would be painful to limit the GPU by bolting it down to a CPU. This would imply that AMD wouldn't make a complete CPU/GPU package, in order to avoid being outrun by Intel, which is already ahead with the Core 2 series. But combining the two to speed up gaming, and the like is still a juicy target. So I think, and this is supported by this line of thought, and a couple sources (obviously still too much speculation for me to feel comfortable plopping it in the article), that its still gonna happen, but not in the way currently described in the article. I think it will be implemented in a way that the CPU doesn't lose clock cycles handing the graphics portion of the work to the GPU, or will ready the data for better processing so that the GPU runs faster. Like identifying it, setting it aside, reading the data, then sending enhanced data to the GPU, without losing too many clock cycles. - Thekittenofterra 00:38, 12 February 2007 (UTC)

I'm not quite sure what you're saying. The GPU and CPU will be two separate fully functional processors, just sharing one package (and possibly/probably the same die), alleviating the need for the PCI Express bus for graphics communication. The CPU therefore won't lose any more clock cycles when the GPU is working then it does currently. I see this product as being more for integrated markets then enthusiast though - the total package would likely have under a quarter the memory bandwidth of your average graphics card at time of release. Themania 10:49, 17 March 2007 (UTC)
AMD and ATI are renound for releasing enthusiast products into the market. Fusion will probably be aimed more towards the intergrated markets at launch (to be safe), but this doesn't stop AMD releasing an enthusiast version (after all, there's no need to poll the PCI Express, this saves time alone). You're right in the matter that the GPU chip (RenderX) isn't locked to the CPU though, they can (as far as I'm aware) work independantly still. Fusion just allows the CPU and GPU to interact at a much faster speed (bandwidth available alone is increased quite a bit). It'll be interesting to see what AMD have up their sleeve(s). Rasengan 23:07, 31 March 2007 (UTC)
Its true that people who upgrade their computers tend to upgrade their GPUs much more frequently than their CPUs, but its only a tiny minority of computer users who upgrade their computers at all, and AMD's fusion parts aren't aimed at these customers. These people will, hopefully for AMD, just get a FX-8140 and a Radeon hard or something like that. Symmetry (talk) 14:36, 20 April 2012 (UTC)

Out of date?[edit]

"According to the [CNN] article, further details will surface in December of 2006."

Has there been any newer information? 21:25, 28 June 2007 (UTC)

[1] Can't find the originals for the images on this post, but it's an updated roadmap from what is currently shown.ShilliamWakespeare (talk) 22:06, 9 April 2011 (UTC)

Broken References[edit]

"AMD's "Fusion" Process to merge CPU and GPU" ([2])

The link appears to be broken... not the greatest page to use for a citation 20:54, 31 July 2007 (UTC)

nehalem graphics[edit]

this article says that nehalem will have a graphics core on die, nehalem says it will be off die. which is right? Bob A 02:33, 20 September 2007 (UTC)

That article has not been updated since Beijing IDF, and recent IDF has just ended, just give sometime. BTW, I trust that The Inq is right about that. -- 05:48, 21 September 2007 (UTC)


I feel this section should be removed as there is no proof this is true. There are a few very important things overlooked in this section and the speed increase is made on the assumption that the PCI-e bus is in fact a bottleneck to today's graphics performance.

Here are a few key points that are not made with regard to fusion...

  • The GPU is now sharing die space with the CPU which means manufacturers will have to cut down on the GPU processing units or produce very large die. This is not favourable as larger die waste more space on a wafer and are also more prone to defects in manufacturing. A larger die will require more power and produce more heat also.
  • Current GPU's have memory bandwidths of up to 141.7 GB/s dedicated for their use.
  • That memory is also their own private memory for storage and internal use.
  • New GPU's bring faster graphics on PCI-E buses suggesting that the PCI-e bus is not bottlenecking GPU performance
  • GPU's are massively parallel devices which "crunch" data using their various units.
  • The display connector is on the GPU board so while data may go from CPU to GPU the data is then displayed through the GPU display port. The process is already efficient.
  • The current section gives the impression that the only change happening is that the CPU and GPU are coming together, also to be connected via a faster link giving the impression that this is all that is happening without mention of any other trade-offs.
  • The bus speed is negligible unless the GPU has so much processing power it saturates it.
--In reality, on an E350 "desktop" system, in games the system IS CPU limited (aside from 1920x1080)! Then at 1080 it is a combo of both since the HD 6310 is little more than a HD 5450, just doesn't have the processing power to render at a reasonable frame rate. Also from the looks of it, the Brazos platform APUs (Ontario/Zacate, etc) do seem to just be a marriage of a Bulldozer or two (for dual cores) and a GPU, with a little Northbridge thrown in; Llano looks to be more than that. Mind you I'm not an expert, nor do I know the latter bits to be fact at all (6310=5450, and there being no huge sacrifices made to the GPU segment). This is all just what I'm gathering from looking at the Die shot, which is of course was doctored a little by AMD to not give away key architecture bits, and specs/technical data. Formula350 (talk) 15:51, 5 May 2011 (UTC)

To elaborate further on the last two points I'll use an analogy using an internet connection. Imagine you have a set of data that needs to processed. A workload such as folding@home is a good example of such data. You have a computer that will take 5 hours to process that data. The data is 10mb in size and you have an internet connection that can upload that data to another machine in 10 minutes. Another computer in the world can process that data in 2 hours. Despite the internet connection being very slow compared to internal system memory and the other machine being maybe the complete opposite side of the world it is still quicker for you to send it to the other side of the world have it processed there and sent back. This is similar to how GPU's work. They have their own memory for storing data so that data is not constantly being transferred across the PCI-e bus.

If we were to move that other machine closer to our machine or say upgrade the internet connection speed the time could be even quicker. But if that process involved other trade-offs such as the other machine having half the processing power the time would be slower.

There is also no mention of the logical reason behind integration. Cost saving. If all the processing equipment a mainstream user needs can be put onto a single chip production costs are reduced significantly. For example the process of testing chips only requires one chip to be tested as opposed to say 4 or 5. A single production line is easier to manage than many.

Despite what I have said above fusion will likely be faster than current integrated solutions because the graphics processor is nearer to the system memory controller (integrated shares system memory). Also fusion systems will likely use DDR3 giving the GPU more memory bandwidth. —Preceding unsigned comment added by Djgandy (talkcontribs) 12:51, 17 July 2008 (UTC)

I have copy-edited the section to try to keep more neutral, but it will probably removed due to lack of reliable source to verify the information and future benchmarks would confirm or deny the speed increase on CPU and GPU processing. But what about adding the information that you talked above? That would really helps to keep the article more neutral and informative. Carlosguitar (Yes Executor?) 22:59, 2 August 2008 (UTC)


Now that the Fusion software has been announced (see, I Believe this this page needs to be written from scratch since the assumption didnt come true. —Preceding unsigned comment added by (talk) 18:27, 21 September 2008 (UTC)

This article is a mess! -.- I found it to be a confusing and rather jumbled read. I think the whole thing needs a rewrite and an update as well. (talk) 08:19, 13 March 2011 (UTC)

Agreed, this page probably has the worst structure I have ever seen on wikipedia. — Preceding unsigned comment added by (talk) 13:26, 27 January 2013 (UTC)

"Competition" a cut'n'paste from Atom's page?[edit]

All of the competition section seems to be discussing how various other vendors compare to Intel's Atom, not AMD's fusion - I'd guess the section was lifted from the Atom page and needs a rewrite. (talk) 17:51, 30 June 2011 (UTC)

July 25th "leak" is very likely bogus[edit]

There have been two "leaked" slides that have been attributed to the website and some of the details from these slides have made it onto this page. These slides are likely bogus and here's why.

First, only pages 4 and 12 have been leaked. If you're going to fake a report, then you're going to do as little work as possible and just fake two pages. If you're leaking information, you want to release everything and we'd have the entire report.

Second, there are a number of minor presentation mistakes that are unlikely to come from AMD. Such as "July, 2011" (the comma is extraneous). And on page 12, the text runs out of the bottom of the bubble.

Third, it has "AMD Confidential - NDA Required" stamped across each slide. I've never seen a legitimate slide with that across the front. Usually, if the NDA is that important, they will stamp the NDA information at the top or bottom of each slide, but never across the information that they're trying to present. The way it's done, it simply serves to obscure the information and I can't imagine AMD doing that.

Fourth, technically, it makes no sense. AMD has a history of supporting their sockets for a long time. AM3+ just came out (and in fact, there are no AM3+ processors yet). AM3 has been the flagship socket for over 2 years now and to suggest that AM3+ could be replaced within six months is silly. Further, the FMx series sockets are for integrated graphics, but the flagship processors don't have on-board graphics for two reasons - first, power users always supply their own and second, they don't want to give up the valuable chip real estate to a GPU when it could be used by a CPU. (talk) 03:06, 26 July 2011 (UTC)

More understandable overall format[edit]

I hope someone can summarize everything and also make it into a more understandable format for the whole page. Coz even though i have knowledge in computers, i find it hard to understand alot of stuff in the page. and i find that there is alot of excess stuff. For example the "Comparison of CPU core power" table, it is basically pointless. (talk) 08:46, 13 August 2011 (UTC)

Clarify what's the point of this APU[edit]

The article has a rich table showing that this APU has horrible performance (20 times slower than Core i7), and then only much later it mentions that the power consumption is comparable to Intel Atom; reading the table again shows that performance is then much better than the Atom processors which are mentioned (which, however, are not the current ones). The introduction needs to mention and emphasize that these APUs are designed to be energy-efficient. --Blaisorblade (talk) 18:46, 2 January 2012 (UTC)

I edited the introduction to reflect the above comment, feel free to improve it.--Blaisorblade (talk) 18:46, 2 January 2012 (UTC)

Needs updating[edit]

Shouldn't the A8-3820 be in the list of Llano APUs? — Preceding unsigned comment added by (talk) 07:18, 13 February 2012 (UTC)

3-cores / 4-cores Llano APUs[edit]

Shouldn't the A6-3500 be the only 3-cores Llano APUs?
I'm pretty sure A6-3600, A6-3620, A6-3650 and A6-3670K are 4-cores APUs. (talk) 00:12, 5 March 2012 (UTC)

There seem to be 2 Llano sections - I cant work out why. Is there duplication? Fig (talk) 22:10, 30 January 2013 (UTC)

new Trinity APUs released[edit] — Preceding unsigned comment added by (talk) 10:08, 15 May 2012 (UTC)

2012 platforms[edit]

Today the new A10-5800K was announced - with an Radeon HD 7660D GPU! This would be an Southern Island and not Nothern island as it is announced in the table... — Preceding unsigned comment added by (talk) 17:26, 4 June 2012 (UTC)

Also the E2-series is missing. Neználek (talk) 13:15, 11 July 2012 (UTC)

A LOT of the new platform information is missing. Existing tables should be updated as well to reflect new offerings in those process nodes. Added the 'update' tag until article has been improved. 2dFx (talk) 18:53, 5 October 2012 (UTC)

AMD C-70[edit]

It appears that the AMD website is no longer a reliable source of information. The Processor specs for the E2 series Brazos chips state a 32nm process, when the chips are actually fabricated using a 40nm. The AMD shopping page manages to get this correct. All 3rd party sources agree that it is a 40nm process. However, there appears to be no consensus on the detail of the C-70 chip. The AMD processor spec page says that it has a 7290 graphics core, while the shopping page says it is a 6290 core. 3rd party sources are in disagreement too, with some stating 7290, and some stating 6290. We need someone with some detailed technical knowledge of this chip to provide input. — Preceding unsigned comment added by (talk) 15:31, 23 December 2012 (UTC)

Names and Terms[edit]

AMD Fusion (APU) is now a deprecated term. But it is not replaced by Heterogeneous Systems Architecture (HSA). AMD has shortened the term into AMD APU - check HSA is a different story, and is truly not the name of AMD's series of APU. HSA is a software infrastructure maintained by the HSA Foundation, which is formed by several key SoC, IP and chip vendors like AMD, ARM, Qualcomm, Imagination, Samsung, MediaTek, TI, etc. The HSA stuff worths its own page, instead of being treated as AMD's own property. - Watdapuff (talk) 08:08, 5 January 2013 (UTC)

work needed[edit]

I was going to make some edits to bring the article more up-to-date, but the article is just such a mess that I don't know where to start. I think we need to put in a section for current fusion processors, but beyond that, it needs reorganization. it looks like someone just dumped a bunch of new articles, and put in some tables but just left them lay wherever. adding cleanup tag. Aunva6 (talk) 07:00, 20 January 2013 (UTC)

how about making this page reflect on the strategy and point behind fusion, and split the platforms and processor models up to their own pages? I feel the E/E2/C/Z series apus are as comparable to trinity and llano as ivy bridge/sandy bridge is to atom. Maybe divide this page into low-power and performance sections? The only thing making these two architectures equal is the iGPU architecture and the marketing name! — Preceding unsigned comment added by (talk) 13:31, 27 January 2013 (UTC)

not a bad idea. this weekend, i'll work on splitting it into sections to start. probably make a lead as well. Aunva6 (talk) 19:57, 13 February 2013 (UTC)
turns out much of the article was duplicated in List_of_AMD_Accelerated_Processing_Unit_microprocessors. I have removed all duplicated material, and I still plan to add alot more information this weekend. and thanks for the help codepro Aunva6 (talk) 05:10, 14 February 2013 (UTC)

I have reorganized the page to combine similar sections and to list platforms and features by year. The article is still a mess and it looks like it has no connection to List of AMD Accelerated Processing Unit microprocessors at all. I suggest that most of the specifications should be removed to the latter list and the sections on current features and the roadmap should be rewritten in prose. --Dmitry (talkcontibs) 22:05, 22 February 2013 (UTC)

yeah, I agree. I might have forgotten to put a link to the list in when I removed the tables. there is a link in the references section. Aunva6 (talk) 22:15, 22 February 2013 (UTC)

The list of APUs has some entries for cores and platforms which are not really mentioned in the main article...

I see that a rewrite was ongoing from February 13. Unfortunately converting bare lists of features to bare table of features does not work, and neither did removing the specs of several processors to the list article and not mentioning them again.

Ideally this article should be structured as follows:

  • history of the terms "Fusion", APU and HSA
  • general notes on HSA and its roadmap, system-on-a-chip features, introduction of A-series, etc.
  • several sections on current and future platforms and their overall features and included technologies, arranged by core name - not the "model year" as it is now
    • these sections should have a brief list of processor families based on each platform/core
  • Detailed specs should be removed to the list article; however desktop processors with integrated graphics do not really belong there, since they have their own articles - so combining List of AMD Accelerated Processing Unit microprocessors with List of AMD A microprocessors might be a better idea.

Hopefully it makes more sense. --Dmitry (talkcontibs) 22:28, 22 February 2013 (UTC)

i merged the pages, and this page is still being cleaned up. still has a ways to go yet. thanks for the input. Aunva6 (talk) 22:59, 22 February 2013 (UTC)
Thank you. I have moved similar A-series entries closer together in the list to facilitate future clean-up, and I will undertake to make some preliminary explanation text based on AMD presentations and other Wikipedia articles.
One more point. There is much conFusion caused by the recent retroactive rebranding of AMD Fusion due to trademark dispute. Even if AMD has entered a private agreement and their sales partners are probably obliged to comply by contract, this is not binding on other parties such as Internet news sites and Wikipedia.
There is absolutely no need to go through every article and replace every wikilink to AMD Fusion, like it happened here: [3] [4] [5], as this only creates further conFUSION. All these products and technologies were branded FUSION at the time of their release, the relevant news sources will continue to refer to the Fusion brand (as they are not really going to edit their older new stories and replace all mentions of Fusion with APU, HSA or whatever arbitrary name AMD chooses now), so for now we should retain the Fusion name where possible, duly mentioning that a rebranding took place in Jan 2013. --Dmitry (talkcontibs) 08:41, 23 February 2013 (UTC)
perhaps, although the common name in usage for these processors has been accelerated processing unit for quite a while. Aunva6 (talk) 09:06, 23 February 2013 (UTC)

new upcoming APUs[edit] -- (talk) 14:04, 26 January 2013 (UTC)

major edits[edit]

since major edits are happening, please use the {{Inuse}} tag when doing a major revision, so we can avoid conflicting edits.

BS information[edit]

alot of the information was contradictory to more reliable sources. i've removed some of it, antd it's probably going to be a far sized rewrite by the time i'm done Aunva6 (talk) 01:47, 15 February 2013 (UTC)

to-do (again)[edit]

ok, i am going to figure out what needs to be done to this article. feel free to add anyhting you think of.

  • convert feature table to prose. it is really not what the article needs, especially one that is as big as what we have.
  • convert or remove smaller table. some of the smaller ones are probably better off being removed in their entirety
  • sort through the information on the processors; see what is misplace, what is legit, what is out of date, and what is simply bogus.
  • expand the history and HSA sections, those sections really need more information. the HSA section was entirely from an article I was going to create, but couldn't find enough info for.
  • copy edit. much of the article pretty much just splatters information all over each section, and needs to be more cohesive for starters.

again, if you see anything I didn't catch, post it! Aunva6 (talk) 00:55, 14 March 2013 (UTC)

Clean up time[edit]

This wiki really is a mess. I am going to try and clean up and sort everything out. Help is of course appreciated. Kapitaenk (talk) 05:42, 18 March 2013 (UTC)

should've seen it before me and codepro started the cleanup. keep in mind that there is a list article as well, avoid overlap. try to use prose as much as possible for the feature sections, and the HSA and history sections cou8ld use expansion. Aunva6 (talk) 04:32, 19 March 2013 (UTC)
Yeah, I checked the history, the article was unreadbale a few months ago. :) Anyway, my first priority is clean up. After that I will expand the HSA section and eventually create a new article for it (if somebody doesn't do so first).
I am also not totally satisfied with the sorting I have chosen for the "platforms" section, I will work it out over the next few days though. Kapitaenk (talk) 05:13, 19 March 2013 (UTC)

I feel like all of AMD's articles are a bit of a mess, honestly. Should Bulldozer be separated into Bulldozer, Piledriver, Steamroller, etc (and Bobcat into Bobcat and Jaguar) as the Intel pages are? Strata8 (talk) 09:33, 19 March 2013 (UTC)

True, most of the AMD articles are a mess.
And yes, the Piledriver, Steamroller, Excavator and Jaguar architectures should have their own articles. Kapitaenk (talk) 21:16, 19 March 2013 (UTC)
lol, I made piledriver, and it got proded almost immediately... but yeah, they really deserve separate articles, as much as sandy and ivy bridge do. -- Aunva6talk - contribs 04:06, 24 March 2013 (UTC)
sorry guys, I am taking a break from this article. There is quite a bit of work to be done and it is a lot for one single person (me). I might start editing again next week some time. Kapitaenk (talk) 05:55, 28 March 2013 (UTC)
no problem. there may be more editors coming to help, but we shall see. -- Aunva6talk - contribs 14:27, 28 March 2013 (UTC)

I have reorganised the article to attempt to improve the flow of information. DElliott (talk) 19:42, 25 April 2013 (UTC)

Which structure?[edit]

I deliberately structured this page by architecture-type and List of AMD Accelerated Processing Unit microprocessors by machine-type so we could discuss how the information should be structured. Does anyone have an opinion on which works better? DElliott (talk) 12:53, 15 May 2013 (UTC)

looks good for this, i'm trying to find a decent sized image of the new logo, but all i've gotten is a crop from the newegg ad... -- Aunva6talk - contribs 17:47, 5 June 2013 (UTC)
In that case, what are your thoughts on converting List of AMD Accelerated Processing Unit microprocessors to show the information by architecture type?DElliott (talk) 15:18, 10 June 2013 (UTC)
sounds like a good idea. especially since that would pretty much make it organized by time as well... -- Aunva6talk - contribs 16:37, 5 July 2013 (UTC)
DrSeehas, do you have any thoughts on this change? DElliott (talk) 08:27, 7 August 2013 (UTC)
You can organize the information by "application" (desktop, mobile, ...) like it is now or by architecture type (your proposal). Maybe two lists? I personally prefer the latter, but I suppose I am not the typical Wikipedia user...
Could you please have a look at List of AMD CPU microarchitectures. Is Llano Family 12h or Bulldozer (Family 15h)? --DrSeehas (talk) 08:49, 7 August 2013 (UTC)
Sorry, can you rephrase that question? DElliott (talk) 13:23, 7 August 2013 (UTC)
I try: Is "Llano" AMD Family 12h or AMD Family 15h?
And another question: Is there an AMD Family 13h? --DrSeehas (talk) 13:45, 7 August 2013 (UTC)
<> suggests Llano is 12h. I've never heard of 13h. — Preceding unsigned comment added by DElliott (talkcontribs) 14:15, 7 August 2013 (UTC)
Thank you. So I did it nearly right. What irritates me: "Llano" is supposed to be Bulldozer also, but Bulldozer is in this article Family 15h? I think, Llano should be exchanged with Bulldozer (AMD Family 12h) and Bulldozer should be Piledriver (Family 15h). --DrSeehas (talk) 15:02, 7 August 2013 (UTC)
I believe that Llano is 12h (derivative of 10h), Bulldozer is 15h (Piledriver Steamroller Excavator), Bobcat is 14h, Jaguar is 16h (Puma) probably would be better shown with a family tree ClearBlueSky85 (talk) 08:33, 12 September 2015 (UTC)


Making the distinction between VLIW5, VLIW4, and GCN is probably too confusing for readers and I'm not sure why it matters. The mainstream press articles don't tend to talk about it those terms. [6] [7] is the only reliable source I can find that makes reference to this distinction. — Preceding unsigned comment added by DElliott (talkcontribs) 11:14, 2 August 2013 (UTC)

Since when have facts become "confusing" to the (dumb?) reader? In case you do not know why it matters, you should pursuit a job in marketing. User:ScotXWt@lk 22:49, 6 May 2014 (UTC)
We have article to Tesla, Fermi, Kepler, and Maxwell but we merely vaguely distinguish between AMD's VLIW5, VLIW4 and GCN microarchitectures. If anything, we need much more information. And it is there, but nobody with that knowledge would waste 5 minutes in the wikipedia. User:ScotXWt@lk 22:56, 6 May 2014 (UTC)

Heterogeneous System Architecture[edit]

Heterogeneous System Architecture redirects to this article and it should not, it should rather have an own article. And then another one for the HSA Foundation. Even if Intel and nvidia have not joined the party (yet?), HSA is quite a "big deal" (look at the founding members, look at benchmarks) and I wonder how it hasn't received more attention in the Wikipedia. I won't write the article alone. Anybody else? User:ScotXWt@lk 22:53, 6 May 2014 (UTC)

I was the one who redirected. the problem is, there are really no sources on the HSA Foundation. trust me, I tried. User:Aunva6/HSAdraft. if you can come up with some, feel free to edit the stub draft I linked -- Aunva6talk - contribs 23:21, 6 May 2014 (UTC)
I actually came looking here looking for HSA, and I'm a bit disappointed that the section about Kaveri only discusses technicalities. What I found in terms of sources is
I do think we should have a short article about this: it's an interesting and innovative idea (unlike the dry, benchmark and spec-based stuff in most articles about chips) that received a lot of tech press coverage. Aunva6, would you mind if I requested a move of your draft to mainspace over the redirect? QVVERTYVS (hm?) 08:38, 23 May 2014 (UTC)
I went ahead and put the DB-move on the redirect. thanks for asking. -- Aunva6talk - contribs 19:34, 23 May 2014 (UTC)
Besides the AMD's marketing department known for being, hm, lame, the Wikipedia should care for "technicalities", i.e. articles that explain how stuff works (not benchmarks). E.g.
ARM is part of HSA Foundation yes, and big.LITTLE is heterogeneous in an unusual sense(?) but not part of HSA I would think. comp.arch (talk) 22:00, 23 May 2014 (UTC)
Hi, I just moved the page BOLDly (hope Dsimic who made the redirect, doesn't mind - only edit history, don't see what the issue is). Note I changed "HSA is widely used in system-on-chip devices, such as tablets" (link is dead) as I believe it to be untrue, that is not using this HSA architecture. Have GPUs generally used the same address space? And HSA is about more than that. I assume the whole "Heterogeneous System Architecture" can be moved then. Including its sub-section(?). I've moved whole sections before with or without putting in merge templates before. I think the main thing is putting in "Copied from" in the heading for copyright reasons. Am I missing something? My first time dealing with this db-move template that was put in (inappropriate, incorrectly?) over the redirect. comp.arch (talk) 22:00, 23 May 2014 (UTC)
Hello there! It's really great that this redirect is now turned into an article! I'll see to contribute more content or some improvements to the article in the next day or two. — Dsimic (talk | contribs) 22:20, 23 May 2014 (UTC)

Heterogeneous System Architecture - hype?[edit]

"operation of systems that contain multiples processing units, CPUs, GPUs, DSPs and/or any other type of ASICs" was unsourced. I found this. I might not be up-to-speed, my theoretical understanding is that compiling C (or some subsetted version) to FGPAs is possible with some limitations. That is amazing. I'm not sure how that applies to HSA, as in, is it being done with HSA currently. Similar for ASICs. That is, isn't it really only being done with CPUs and GPUs currently and the rest is a dream? I think we should say that, what is being used now and works and what is hoped for separately. Note also, potentially more than two types of processors are enough for HSA, right?

In the table: "CPU and GPU now share the power budget. Priority goes to the processor most suited to the current tasks." What does the second sentence actually mean? I guess if total power budget is reached, CPU and/or GPU is throttled. Does computation move from one or the other? How exactly? ARM big.LITTLE does that, but is not part of HSA(?). That is heterogenious power-wise but same ISA. ARM Holdings is however part of HSA Foundation (what does that mean.. currently?). comp.arch (talk) 16:04, 20 May 2014 (UTC)

it's more the concept of integrating more parts onto the same chip or die, like the memory controller, northbridge, etc. the HSA Foundation is a partnership between several corporations, including ARM, AMD, and a couple others. -- Aunva6talk - contribs 19:37, 23 May 2014 (UTC)
HSA was totaly neglegted in the Wikipedia, and now suddenly that I authored some stuff around it, so it may seem to you like a hype. But it's not. Steamroller-based hardware is available and so is the PlayStation 4. Linaro is on-board, so even without more Java-crap and HSAIL we should see some software benefiting from the already implemented concepts of HSA. User:ScotXWt@lk 21:26, 23 May 2014 (UTC)
Not at all do I think HSA is hype in general, been following it and thanks for writing about it. Just beware that heterogeneous computing is more general concept than HSA and is much older than multi-core. I got your attention with the heading :) But note I was really talking about HSA hype mostly in relation to FGPAs. But maybe it's working already? comp.arch (talk) 22:00, 23 May 2014 (UTC)

AMD Fusion feature overview[edit]

The AMD APU feature chart needs to be updated. It states that Richland APUs are based on GCN 1.1 when they are in fact based on VLIW4.

ARM server architecture (2016): Seattle[edit]

It seems that the ARM/Seattle section should be move to a AMD CPU page. There is speculation, without citation, that AMD may release an ARM based APU. And the only citations included are very old. The section itself is, in fact, out of date as Seattle has been released since January as the A1100 series. Dbsseven (talk) 22:40, 15 August 2016 (UTC)