Jump to content

Talk:Adreno

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

Release date column?

[edit]

I feel that it would be beneficial to add a column for release dates for these chips, anyone else agree? Aaronfranke (talk) 17:38, 2 September 2016 (UTC)[reply]

"Adreno Nxx series"-lines could read as "Adreno Nxx series (2017Q1-2021Q3)" too. — Preceding unsigned comment added by 188.238.221.151 (talk) 23:54, 23 November 2016 (UTC)[reply]


Table bar could be floating or repeated down page, it makes comparing features difficult

[edit]

--RofthoraxGmailCom (talk) 18:10, 13 October 2020 (UTC)[reply]

Adreno 430 GPU specs

[edit]

The information that the GPU has 500MTriangles/s fillrate is a rumor from the page and it also says that Snapdragon 810 has Krait64 CPU and supports LPDDR3. They are not truth. So it should delete data of ALUs, fillrate, and flops.--Maricolle1993 (talk) 03:35, 19 January 2015 (UTC)[reply]

Magazine

[edit]

It looks like the thing was written like magazine in some sections...... Please correct any if you like. --202.71.240.18 10:42, 13 April 2007 (UTC)[reply]

move to "Adreno"

[edit]

The article should be move to Adreno Doors5678 (talk) 13:18, 19 April 2012 (UTC)[reply]

Yes, I agree too, the article's title should be "Adreno" or "Qualcomm Adreno". --Faramarz♚♔♚ 10:05, 23 March 2013 (UTC)[reply]

a few corrections/notes

[edit]

hey, this is the freedreno guy. I noticed a couple issues in the table that I'd thought I'd point out:

a3xx is not vliw (althought is still ofc unified shader model). It is a scalar architecture with pipelining handled (mostly) explicitly by the shader compiler... by which I mean various different categories of instructions, the compiler needs to be aware that the result is available a certain # of instruction slots later. I've written up some documentation at https://github.com/freedreno/freedreno/wiki/A3xx-shader-instruction-set-architecture

a4xx appears to be the same shader core as a3xx with some small tweaks. At this point, I don't have a4xx hardware yet, so I can't conclusively say anything. But qcom has included a4xx support in their binary driver releases for a while, and it is possible to trick the blob driver into thinking it is running on a4xx hardware and intercept cmdstream. From this, what I see is that a4xx has basically the same shader core with some small tweaks (there are a few cases where fewer delay slots are needed between instructions). There are quite a lot of changes outside of the shader core, but with only one small difference (larger immediate bitfield for catagory 0 instructions... ie. a4xx can do larger jump's) it looks like shaders compile for a3xx should run correctly on a4xx. (There are, however, quite a lot of changes outside of the shader core, mostly related to support for full dx11 pipeline.)

As far as GLES versions, all a3xx and later should in theory support opengles 3.0. Whether gles3 drivers are shipped on a particular device is a different matter.

Also, I'd suggest to drop the 'Video Codecs' columns. Video codecs are a different IP block outside of adreno.

And I'm not sure about the 1.2 in the 'OpenGL' column for a4xx. That doesn't seem right. At least I've not seen anything where qcom advertised desktop opengl support (but if they wanted to, it seems like a3xx can at least do (or emulate) most of opengl 3.0 if not all). — Preceding unsigned comment added by Robdclark (talkcontribs) 03:39, 14 May 2014 (UTC)[reply]

Merge with Snapdragon?

[edit]

I have a conflict of interest / affiliation with Qualcomm. I know that someone must have worked very hard on this article at some point, but I'm not sure the page should exist. Most of the content seems to violate What Wikipedia is Not as an indiscriminate list of products, features and specifications. There's also some Crystal Ball, where it refers to 2009 as future date and most of the proper encyclopedic material is unsourced.

Normally in this situation we could find a few good sources and stub the article for now, but in this case, Adreno does not appear to meet GNG, because it is not "the subject" of multiple, reliable, in-depth sources. Rather, the available sources are actually about Snapdragon and discuss Adreno within them. Therefore, per WP:ORGVANITY (disclosure, I wrote that essay), I think it would be more appropriate to follow the judgement of the source material by creating an Adreno sub-section on the Snapdragon page, rather than creating a separate article on Adreno, whereas the sources do not treat it as an independent subject.

CorporateM (Talk) 20:19, 20 June 2014 (UTC)[reply]

sorry oppose while it is AFAIK true that since the division was bought by Qualcomm, the Adreno-SIP block have only been found on the dies of Snapdragon SoCs. Still, it is a distinct product, that should have a distinct article. But the article definitely could be better, see e.g. Qualcomm Hexagon. User:ScotXWt@lk 13:38, 22 June 2014 (UTC)[reply]
User:ScotXW I pinged a couple editors after posting this and I'd like to give them time to weigh-in, but in the meanwhile, I wanted to see if we both agreed that these pages should not have these large tables of products and technical specs? I could update/correct the spec tables, but it would literally just be copy/pasting a bunch of content from Qualcomm's spec-sheets and it seems like an indiscriminate collection of information to me. CorporateM (Talk) 20:05, 23 June 2014 (UTC)[reply]
The table is there to facilitate comparison with Nvidia, AMD, Vivante Corporation, etc. Of, some tables need to be filled with – referenced – information. I hope in time, passers-by will fill the gaps. I doubt, somebody would do both: create a new table AND fill the gaps. User:ScotXWt@lk 20:13, 23 June 2014 (UTC)[reply]
  • Comment - I agree that both this article and the Snapdragon (system on chip) article suffer from an excess of lists and tables. They don't look very encyclopedic, and seem in violation of WP:NOTMANUAL, where they resemble technical manuals more than articles. It's to the point where both articles are difficult to read because it's a struggle to find the article in between all of these tables. And keep in mind, I'm an IT professional, so I can't imagine how these articles would look to someone not in the tech field. -- Atama 20:52, 23 June 2014 (UTC)[reply]
    I agree with you. So? I think this articles are very useful to have around. Do you, as an "IT professional" agree? People interested in Hannah Montana may not agree, but then again, they would probably stay away from this talk page. The WP hasn't been a welcome place since a very long time now. The WP is still useful though, in its current state, but I could imagine something much more useful, and contribute there. User:ScotXWt@lk 07:10, 24 June 2014 (UTC)[reply]
Regarding the proposed merge, since my suggestion is that there is not much encyclopedic material on this page cited to reliable secondary sources to salvage for merging to the Snapdragon page, I think AfD would be a good place to get more discussion and a vote. Although I do think the pages should be merged, you have also presented a reasonable argument that if a substantial portion of Adreno's history pre-dates its integration with Snapdragon, then there may be a sufficient amount of information that is not relevant to Snapdragon to warrant creating a separate article. My understanding (and I am not an expert) is that pieces of acquired technology were integrated into Adreno, but "Adreno" didn't really exist except as a part of Snapdragon. Just certain pieces of its intellectual property did and were being used for other purposes by other companies under other product names.
However, I think what user:Atama is referring to is trimming down this article of specification tables regardless of whether the page is merged/deleted. I think the question is not whether the information is useful, but whether Wikipedia is the appropriate place for it. CorporateM (Talk) 15:48, 24 June 2014 (UTC)[reply]
Yes, that is what I meant. It's not the presence of the tables that's a problem, it's how they overtake the articles. It's good info but I don't think it all needs to be on here. I'll give an example, look at P5 (microarchitecture). It's not up to GA status, but it's rated B class, so it's not awful. There are tables with technical jargon but they supplement the article content. They don't dominate the article. -- Atama 16:51, 24 June 2014 (UTC)[reply]
A quick Google search reveals this list of just five models. I think editors typically see product grids like that to be promotional, but perhaps the community is more forgiving on technical subjects. CorporateM (Talk) 17:21, 24 June 2014 (UTC)[reply]
  • Comment - coming here after CM left me a note. I agree that the article is a mess in it's present state and for me as a reader who has never heard of it (which is the audience we should really aim for) the tables are intimidating. It's more like a catalogue than an encyclopedia article. As in all cases, we should follow the sources that discuss it - not necessarily the same as being useful. If it they include some content in tables, then maybe we should too, but if the current content can't be found in reliable sources then it shouldn't be here. I think it would be better to discuss a merge once any sources have been identified and the content of the article adapted accordingly. If at that point it is clear that there isn't enough information for a standalone article then it can be merged into Snapdragon (system on chip). SmartSE (talk) 21:05, 25 June 2014 (UTC)[reply]
I think that's a good standard to use. Limit the tables, and see how much of an article can be made using reliable sources, rather than technical specs to fill up tables. If it's impossible to expand this beyond a sub-stub then it might be better off being a section in the Snapdragon article. If, however, it's possible to have the beginnings of an article then keep this here and give it a chance for expansion as more coverage is produced and/or found. -- Atama 21:19, 25 June 2014 (UTC)[reply]
Not to beat a dead horse, but I think deferring to the judgment of the source material brings us full-circle to the original issue. For example, if I do a few quick searches for articles in Electronic Engineering Times[1][2][3][4][5] and Electronic Design News[6] not only do the sources not include spec tables, but they do not treat Adreno as a separate, independent subject. CorporateM (Talk) 22:30, 25 June 2014 (UTC)[reply]
[edit]

Hello fellow Wikipedians,

I have just added archive links to 2 external links on Adreno. 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 to let others know.

checkY An editor has reviewed this edit and fixed any errors that were found.

  • 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 05:15, 20 January 2016 (UTC)[reply]

[edit]

Hello fellow Wikipedians,

I have just modified one external link on Adreno. Please take a moment to review my edit. If you have any questions, or need the bot to ignore the links, or the page altogether, please visit this simple FaQ for additional information. I made the following changes:

When you have finished reviewing my changes, you may follow the instructions on the template below to fix any issues with the URLs.

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: 5 June 2024).

  • 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.—InternetArchiveBot (Report bug) 04:19, 27 June 2017 (UTC)[reply]

Draft

[edit]

This page is about a graphics processing unit that is exclusively sold as a component of Qualcomm Snapdragon chips. In 2014, I suggested we might want to consider whether the Adreno page should be deleted or merged into Qualcomm Snapdragon. @Atama: and @Smartse: said it is hard to say looking at the current article, which is flooded with tables. They wanted to see a cleaned-up version of the Adreno page in prose with good sources to get a better idea. Sometimes the best way to see if something is notable (or in this case whether it conforms to WP:LENGTH for separate articles) is to see what comes out when you try to write about it with good sources.

I actually wrote a cleaned-up prose-style version in 2014; then left it collecting digital dust! Although it is four years old now and therefore a bit out-dated, it would be a massive improvement over the current article. I have pasted it into Talk:Adreno/draft for consideration as a replacement of the current article, followed by consideration for whether it should be merged to the Snapdragon page. CorporateM (Talk) 15:13, 10 February 2018 (UTC)[reply]

This guy copped a clue. I, too, thought the whole article could be junked except for the lead and History sections. Technical details belongs in a product spec sheet, not wikipedia. The Product line section reads like advertising script. What's missing here, and maybe user corporateM has that knowledge (I have some of it, too), is a whole set of sections on the architecture: an overview followed by a section on each component. These are the substance of the article, and you have to know something to write something. Someone browsing the current article comes away without knowing anything about what Adreno is or how it works. Sbalfour (talk) 15:58, 11 November 2018 (UTC)[reply]
A large part of the problem here is that there's virtually zero meaningful information about the Adreno architecture in the public domain. What I know comes from privileged sources, and part or most of it may be trade secrets. Sbalfour (talk) 20:08, 11 November 2018 (UTC)[reply]

I think the chart should remain. I agree it could be edited down but the major parts are good. I would even say you could just make it Collapsible so it still there but will not take over the full page. ContentEditman (talk) 15:16, 14 November 2018 (UTC)[reply]

I do not understand the reasoning behind removing the technical specs table from this article. I am not aware of any other source for such information, and I reference it often. If the format of the page is the issue, could the table be relocated and linked here? I'm not after trade secrets, but many of these specs can be independently derived, and crowd-sourcing seems like an ideal model to maintain this. Chimbraca (talk) 02:11, 15 November 2018 (UTC)[reply]


Split up again

[edit]

Separated out Imageon and Adreno to provide space and context to expand on Imageon history. This also will give more freedom to merge Adreno with qualcomm's Snapdragon if so desired. Though I personally think Adreno merits an independent page, at present it just lacks the kind of encyclopedic info that would be of high value. pinchies (talk) 08:24, 26 October 2019 (UTC)[reply]

About Adreno gpu total theoretical fp32

[edit]

I asked qualcomm about their adreno gpu's total gigaflops in fp32 and they didnt answer my question but they mentioned third party vendors can provide it. So i asked samsumg technical support the total fp32 of adreno 740 and they responded '3481.6~theoretical gflops for adreno 740' For adreno 750 4435.2~ gflops When i asked about links they said it's internal confidental information and i accepted it. I did ask samsung to provide the total gflops of snapdragon 8 gen 4 igpu and when they release they're release their new smartphone if they use it. They answered they will reconsider it and hopefully they provide it the next time.

If you dont believe me you can ask samung yourself ,but try to ask qualcomm because qualcomm asked me what university im at and since i dont study,i couldn't answer their question Techmin651 (talk) 07:07, 29 July 2024 (UTC)[reply]

Hello and welcome to Wikipedia.
Things don't work out this way on here, whatever is on Wikipedia must be cited and backed up by reliable sources, that Samsung rep you spoke to has absolutely no idea what he is talking about.
Adreno 740, 750 and X1 have the same number of cores & ALUs, each has a total of 1536 ALUs, the Adreno X1 is advertised by Qualcomm to have a 4.6 GFlops FP32 at 1500MHz, so in order for the 8G2 FG to get 3481.6 FP32 GFlops it needs to run ~1133MHz and the 8G3 FG needs to run at ~1433MHz which we all now is false due to clockspeed being known to be much lower at 719MHz & 1000MHz respectively. Alawadhi3000 (talk) 19:50, 29 July 2024 (UTC)[reply]
look first of all, you need to check your cheeps and cheese link about adreno x1 using a varaint of adreno 740, because ethier the author did a typo which i clearly think not, he clearly says a variant of adreno 730 of snapdragon 8 gen 1+ family.
Speaking of reliable source i respect that but your links,sources doesn't meantion any total alu about adreno 740 and adreno 750 and the only respected source was anandtech but they dont do any deep analysis like before.
So etheir you speak to qualcomm about details of their igpus or leave them empty even the gflops part of adreno 650,660,740 and 750 because you have no proof of data your given.
It's only qualcomm which i encourage you to do that to ask as researcher in a verified university or their third party vendors like samsung,xaomi etc...
The adreno x1 comes straight from qualcomm and i thank them for that, but the other igpu's that i mentioned,like i said,i dont see any reliable data from your links nor do i see qualcooms links meantioning adreno 740 and adreno 750 total alu and gflops, so you should remain their coloums empty with your geuses until qualcomm proves you otherwise.
You keep on editing the same way, we're going to have ourself a big issue and hope other readers stand up to you and im going to let wikipedia decide if your right or wrong Techmin651 (talk) 07:34, 30 July 2024 (UTC)[reply]
I never said the X1 is a variant of the Adreno 740, you have reading or comprehension problems, the X1 is a variant of the 730 and not the 740 or 750 is because both the 730 and X1 lack hardware raytracing whereas the 740 & 750 support hardware raytracing. The X1 appears to be a scaled up Adreno 730 just like the C&C author said. The number of ALUs between the 740, 750 & X1 are the same and this is confirmed via dieshots plus other sources.
You know that if you have the number of ALUs and the clockspeed its easy to calculate/confirm the theoretical FP32 output right?
Snapdragon 865 dieshot shows that the Adreno 650 has 3 clusters, 3 clusters = 768 ALUs, clockspeed is 587MHz in the 865 or 670MHz in the 865+ and 870.
2*768*0.587 = 901.632 FP32 GFlops for the Snapdragon 865.
2*768*0.670 = 1029.12 FP32 GFlops for the Snapdragon 865+ and 870.
Snapdragon 888 dieshot shows that the Adreno 660 also has 3 clusters, 3 clusters = 768 ALUs, clockspeed is 840MHz.
2*768*0.840 = 1290.24 FP32 GFlops for the Snapdragon 888.
Snapdragon 8 Gen 2 Analysis (incluing dieshot) shows that the Adreno 740 has 6 clusters, 6 clusters = 1536 ALUs, clockspeed is 680MHz or 719MHz in the For Galaxy version.
2*1536*0.680 = 2088.96 FP32 GFlops for the Snapdragon 8 Gen 2.
2*1536*0.719 = 2208.768 FP32 GFlops for the Snapdragon 8 Gen 2 for Galaxy.
Snapdragon 8 Gen 3 dieshot shows that the Adreno 750 has 6 clusters, 6 clusters = 1536 ALUs, clockspeed is 903MHz or 1000MHz in the For Galaxy version.
2*1536*0.903 = 2774.016 FP32 GFlops for the Snapdragon 8 Gen 3.
2*1536*1.000 = 3072 FP32 GFlops for the Snapdragon 8 Gen 3 for Galaxy.
The last line on your reply in just comical, this will be the last time I just revert your disruptive edits, any missing ref will be added to the article. Alawadhi3000 (talk) 18:00, 30 July 2024 (UTC)[reply]
Was it you that edited adreno x1 using a varaint of adreno 740 and if you did, you need to correct it.As for your other links, i see the link to the adreno 750 igpu that author wrote something 5548 gflops in fp32 "not fp16"
As for the other links as i said,it's better that comes from qualcomm because some it is still geusswork, like the total number of alu.
You know what i dont care, i geuss i had higher expectation and thought the only thing holding adreno igpu's back was memory bandwith and overheating.
That guy who succeded running red dead redemption 2 on snapdragon 8 gen with fan cooling was a big leap Techmin651 (talk) 19:42, 30 July 2024 (UTC)[reply]
No I wasn't the one who edited it from being a variant of A730 to A740, you can redo that edit if you want.
Qualcomm doesn't share any info on its GPUs these days, sometimes they don't even share the model number of the GPU so we have to get it from source code, drivers, apps ...etc, the only recent exception was the X1 where they did a deep dive into the specs.
The author of the 8G3 dieshot saying the FP32 throughput is 5548GFlops looks like he is suggesting that the Adreno 750 has now dual issue ALUs like NVIDIA Ampere, Ada Lovelace and AMD RDNA3 architectures. I really doubt this claim as you can't tell that from a dieshot, so better to play it safe and keep it single issue until we have more proof.
Mobile GPUs are mainly limited by TDP, these mobile chips usually run at 5W-10W which is not enough to run a 1536 ALU GPU even at max clocks, you can tell that the big issue wasn't memory bandwidth when you compare CPUs that has octa channel memory controller don't perform much better even though they have double the memory bandwidth. Alawadhi3000 (talk) 20:37, 30 July 2024 (UTC)[reply]