Talk:List of AMD Ryzen processors/Archive 1

Latest comment: 1 year ago by AP 499D25 in topic Table redesign
Archive 1Archive 2

Suggestion: Re-use existing templates

From Zen (microarchitecture):

=== Desktop processors ===
{{AMD Ryzen 1000 Series}}
=== Mobile APUs ===
{{AMD Ryzen Mobile}}
=== Desktop APUs ===
{{AMD Ryzen Desktop Processor with Radeon Vega Graphics}}

From Zen+:

===Desktop===
{{AMD Ryzen 2000 Series}}

From List of AMD accelerated processing unit microprocessors (linked in Zen+):

==="Picasso" (2019)===
{{AMD Ryzen Mobile 3000 series}}

This means that you don't have to copy the contents of these templates, but instead you can just put the name of these templates inside double curly brackets as seen above. See WP:Templates for more information on how to use them.

--Pizzahut2 (talk) 15:26, 23 January 2019 (UTC)

The pages now properly use {{AMD Zen+ based mobile APUs}} . {{AMD Ryzen Mobile 3000 series}} has been kept as a branding template in case it is needed. Trigenibinion (talk) 10:05, 23 January 2020 (UTC)
The pages now properly use {{AMD Zen+ based desktop CPUs}} instead of {{AMD Ryzen 2000 Series}} because the table includes two brandings. Trigenibinion (talk) 10:16, 23 January 2020 (UTC)

Structure of Article: Architecture vs Market based

Currently all product tables are contained twice in this article, eg. AMD V1000 series is under "Zen based" and under "Embedded". I think we should try to avoid such duplication and stick to one theme. So either top level is arch and in each arch are all the markets (desktop, mobile, emb) OR top level is market and below are the different archs. Thoughts?

I would prefer the market based scheme, since a reader might be interested in only one market and this way can nicely see the progression in that segment. Whereas with the other scheme (arch based) one would need to "jump" between sections. Wikiinger (talk) 20:14, 23 June 2019 (UTC)

EPYC isn't Ryzen

I see that a table of EPYC processors is included here. The thing is, EPYC isn't Ryzen. So should it be removed or should the article be renamed? 87.75.117.183 (talk) 20:14, 24 June 2019 (UTC)

This was my mistake apparently. You're right, EPYC shouldn't have been added. Pizzahut2 (talk) 11:07, 25 June 2019 (UTC)
EPYC is technically Ryzen, with different branding and minor modifications in circuitry. Erkin Alp Güney 19:25, 16 September 2021 (UTC)
@Erkin Alp: EPYC is not technically Ryzen, it is technically Zen/Zen+ etc. You are effectively saying the 40 core Xeon Platinum 8380 is technically a 2 core i3 1000G1. It is definitely not, it is both Ice Lake. Like that, Ryzen is the marketing name of the consumer product, EPYC the marketing name of the professional/server product.2A02:A452:5E54:1:F08E:62EB:41AB:ED3E (talk) 21:14, 30 September 2021 (UTC)

What is going on here?

First, what is this with 2000 series APUs listed as first gen Ryzen and 3000 series APUs being listed as second gen? Also, why is there Athlon APUs listed in a Ryzen article? --uKER (talk) 23:31, 27 August 2019 (UTC)

"what is this with 2000 series APUs listed as first gen Ryzen and 3000 series APUs being listed as second gen?"
AMD mixes generations. The 2000 series uses APUs with 1st gen Zen cores, while the CPUs without IGP are 2nd gen (Zen+). Same happens with the 3000 series.
"why is there Athlon APUs listed in a Ryzen article?"
They are part of the same template. Someone would have to split it into Athlon and Ryzen, and then check and correct all articles which use this template. Some templates only have a single Athlon model, in which case I guess it's not worth the effort.
Alternatively copy the contents of such mixed Athlon/Ryzen templates into the article (omitting the Athlon APUs), but in this case someone would have to keep the copy updated.
Pizzahut2 (talk) 11:00, 28 August 2019 (UTC)
"Alternatively copy the contents of such mixed Athlon/Ryzen templates into the article (omitting the Athlon APUs)"
Done this now for the template which has 4 Athlon APUs.
Pizzahut2 (talk) 11:27, 28 August 2019 (UTC)
There are other Athlons on the page if you want to continue maintaining this. Trigenibinion (talk) 10:36, 23 January 2020 (UTC)
Done. Trigenibinion (talk) 14:37, 23 January 2020 (UTC)
The Zen are not in the Athlon list. Trigenibinion (talk) 14:40, 23 January 2020 (UTC)
Done. Trigenibinion (talk) 15:18, 23 January 2020 (UTC)

Adding codenames and manufacturing process in title

Hello,

Most of the lists of CPU from AMD and Intel have at least the manufacturing process and the codename as title of the tables. Example: List_of_AMD_Sempron_microprocessors. This page seems to be the only one not to have that. I proposed it but it has been dismissed. If anybody wants to reuse this information and to present it in a better way, see my contribution of today. LThecross (talk) 16:24, 12 April 2020 (UTC)

Mainstream/Performance/Enthusiast division

I'm willing to remove these from all the templates because

  1. AMD does not use this nomenclature (check the corresponding web pages)
  2. These pricing tiers are relevant maybe for US and some 1st world countries
  3. They are very subjective
  4. They are extremely current - i.e. no one will call an FX-63 CPU from 2004 an enthusiast CPU

Artem S. Tashkinov (talk) 18:42, 20 December 2020 (UTC)

I agree on all of these points. Adding another reason: none of the other Ryzen templates do this and we should strive for consistency. - Wikkiwonkk (talk) 23:35, 14 April 2021 (UTC)
Except for your first point, I don't agree:
  1. AMD might not use it (I didn't check), but it does allow for a quick classification by the reader.
  2. Those aren't price tiers, but performance tiers.
  3. The naming might be, but the performance isn't subjective.
  4. Obviously the "enthusiast" is to understand relative to the other CPUs in the table, which will still be the case in 20 years or so..
With that being said, I do like the sections in those big tables because they does make them easier to read, IMHO. However I agree the naming is "kind of" subjective, so how about renaming the sections to Ryzen 3 series, Ryzen 5 series, etc.. instead of Entry-level, Mainstream, etc.? This way we can keep the sections, but don't introduce unofficial categories. Wikiinger (talk) 12:09, 8 May 2021 (UTC)
PS: And I would like to move the "HEDT"-Part (Threadripper CPUs) to a separate table, because IMHO those aren't Desktop, but rather Workstation parts.

Ryzen 5 Pro 3350G

I'm missing this one. --Gallograph (talk) 12:43, 1 April 2021 (UTC)

Incorrect PCIe lanes for Renoir

See block diagram leaked from AMD here: https://www.igorslab.de/en/manufacturers-when-gaming-notebooks-in-the-clamp-ryzen-4000-apus-in-the-bandwidth-limit-tiger-lake-comes-first-with-only-4-cores-2/

Renoir Mobile chips have 20 PCIe lanes (8+4+4+4), not 16. I don't know how to fix this.

Instruction set and other information

All, and I mean ALL microprocessor lists I know of, always listed supported instructions and similiar information before each model table. All up to Bristol Ridge (when looking to list of Athlons). Since Zen, nothing, nowhere, frankly, all the tables are more and more useless in late years (look at what happened to list of Xeons).

Why? Behemot (talk) 21:48, 19 August 2021 (UTC)

New Resources

This is my CPU and a CPU not listed yet: AMD Ryzen 5 4600G — Preceding unsigned comment added by 66.102.195.102 (talk) 18:49, 6 November 2021 (UTC)

It's there. I didn't add it, but I did just update it to reflect it becoming available to consumers. Wikkiwonkk (talk) 18:20, 17 March 2022 (UTC)

New Zen 2 APUs CPUs

AMD has thrown us a curveball by releasing two Renoir Zen 2 APUs without IGPs, making them CPUs with 4000 series names (press release): the Ryzen 5 4500 (US $129) and the Ryzen 3 4100 (US $99). They are monolithic (no chiplets), socket AM4, with much less L3 cache. So how are we going to shoehorn these into the CPUs table? - Wikkiwonkk (talk) 18:49, 17 March 2022 (UTC)


New Zen 3+ APUs

[1]

References

CPU/APU Table Style: merge core column

From Zen 4 on AMD will always include a graphics core in there CPUs (APUs). With the current style of the Zen APU tables, the tables will get uncomfortable wide IMHO (cf Ryzen 4000G series (which is missing the Chiplets and PCIe support columns). Hence I propose to merge the Cores (threads) and core config column. The numbers displayed are IMO suited for the the same cell, since they represent the same data. It would look something like this:

Model Release date
and price
Fab Chiplets CPU GPU PCie
support
Memory
support
TDP
Core
config[i]
Clock rate (GHz) Cache Architecture Config[ii] Clock Processing
power[iii]
(GFLOPS)
Base Boost L1 L2 L3
Ryzen 3 4300GE[note 1] July 21, 2020
(OEM only)
TSMC
7nm FinFET
4 (8)
1 × 4
3.5 4.0 32 KB inst.
32 KB data
per core
512 KB
per core
4 MB GCN 5th gen 384:24:12
6 CU
1700 MHz 1305.6 DDR4-3200
dual-channel
35 W
Ryzen 3 4300G[note 1] 3.8 65 W
Ryzen 5 4600GE[note 1] 6 (12)
2 × 3
3.3 4.2 8 MB 448:28:14
7 CU
1900 MHz 1702.4 35 W
Ryzen 5 4600G[note 1] April 4, 2022
US $154
3.7 65 W
Ryzen 7 4700GE[note 1] July 21, 2020
(OEM only)
8 (16)
2 × 4
3.1 4.3 512:32:16
8 CU
2000 MHz 2048 35 W
Ryzen 7 4700G[note 1] 3.6 4.4 2100 MHz 2150.4 65 W
  1. ^ Cores (threads) and core complexes (CCXs) × cores per CCX
  2. ^ Unified Shaders : Texture Mapping Units : Render Output Units and Compute Units (CU)
  3. ^ Single-precision performance is calculated from the base (or boost) core clock speed based on a FMA operation.
  1. ^ a b c d e f Model also available as PRO version (4350GE, 4350G, 4650GE, 4650G, 4750GE, 4750G), released 21. July 2020 for OEM only.

What do you guys think, or other ideas to keep the table width reasonable?
--Wikiinger (talk) 14:54, 3 June 2022 (UTC)

I think it's a good idea, but the way it's written looks a bit cluttered to me. Maybe you could make it clearer by using the typical xC/yT notation that seems to be widely used:
6C/12T
(2 x 3)
As for the GPU columns, I would suggest taking the architecture column out and putting a sentence with a link to it at the top of the table, something like "Integrated GPUs are based on the GCN 5th generation architecture." The GFLOPS figures don't seem particularly important and could be removed too. Zerranto (talk) 22:39, 7 June 2022 (UTC)
Indeed it is a bit cluttered and using common notation is a godd idea. However I would like to avoid parenthesis. Regarding the GPU-Archtiecture see section below. The GFLOPS column is a good indicator how "powerful" (in comparison to the other models) the GPU is, so I would keep it. Lets see how this would look like:
Model Release date
and price
Fab Chiplets CPU GPU PCie
support
Memory
support
TDP
Core
config[i]
Clock rate (GHz) Cache Architecture Config[ii] Clock Processing
power[iii]
(GFLOPS)
Base Boost L1 L2 L3
Ryzen 3 4300GE[note 1] July 21, 2020
(OEM only)
TSMC
7nm FinFET
4C / 8T
1 × 4
3.5 4.0 32 KB inst.
32 KB data
per core
512 KB
per core
4 MB GCN 5th gen 384:24:12
6 CU
1700 MHz 1305.6 DDR4-3200
dual-channel
35 W
Ryzen 3 4300G[note 1] 3.8 65 W
Ryzen 5 4600GE[note 1] 6C / 12T
2 × 3
3.3 4.2 8 MB 448:28:14
7 CU
1900 MHz 1702.4 35 W
Ryzen 5 4600G[note 1] April 4, 2022
US $154
3.7 65 W
Ryzen 7 4700GE[note 1] July 21, 2020
(OEM only)
8C / 16T
2 × 4
3.1 4.3 512:32:16
8 CU
2000 MHz 2048 35 W
Ryzen 7 4700G[note 1] 3.6 4.4 2100 MHz 2150.4 65 W
  1. ^ Cores C / threads T count and core complexes (CCXs) × cores per CCX
  2. ^ Unified Shaders : Texture Mapping Units : Render Output Units and Compute Units (CU)
  3. ^ Single-precision performance is calculated from the base (or boost) core clock speed based on a FMA operation.
Wikiinger (talk) 17:20, 8 June 2022 (UTC)
I don't like that the C/T letters are the same size as the numbers. Makes it hard to determine the count on quick glance. Making the letters small improves the situation a bit, still unsure... Also the column is wider than in the first draft, still narrower than two though. Wikiinger (talk) 17:30, 8 June 2022 (UTC)

Handling single entry columns (Fab, Socket, Memory Support, L1/L2 Cache,...)

Also I think the Fab columns should be removed. The fabrication details would be better served with a full sentence description at the top of each table. Zerranto (talk) 22:45, 7 June 2022 (UTC)

@Zerranto I see you already went (partly) ahead with your suggestion... :-/
I agree to get rid of single entry columns (such as Socket), but with the Fab column it is actually a mixed bag. Since AMD often mixes architecture within same model branding. (see, Template:AMD_Ryzen_3000_mobile_APUs, Template:AMD_Ryzen_Mobile_5000_series and especially with GPU models). So this way the tables are getting inconsistent, which I would like to avoid. Socket is a prime candidate, also Memory and maybe even L1 and L2 cache? Anyhow it would be nice to come up with a good design here, before mass editing a bunch of tables... :-( Wikiinger (talk) 17:45, 8 June 2022 (UTC)
I'm sorry if I've been too presumptive. But honestly I couldn't see any reason to keep the Fab column in the table. For one thing, the fab process will always be the same for all the processors in any particular table, due to the way that the tables are grouped by architecture rather than purely by model number (for example the 1600 AF, which is Zen+ despite being in the 1000 series, is grouped with the 2000 series CPUs). In addition the details have to be abbreviated to fit in the table, whereas outside of it they can be written out in full. Finally I checked other lists of long-running CPU brands, such as List of Intel Core i7 processors, and they all had the process node details in the page headings, rather than in the tables. So I thought I might as well go ahead without any need for debate. But if you want them back, I will restore them.
As for your other points, I agree that the L1 and L2 cache columns are good candidates for removal. I don't think removing Socket or Memory would work though. Threadrippers have different sockets and support more memory channels than the desktop CPUs, and they're listed in the same tables. You could remove the Socket column from the mobile CPU tables though; since they aren't upgradable the type of socket they use isn't really relevant. Zerranto (talk) 19:02, 8 June 2022 (UTC)
Yes, please revert your edits until we've come up with a good design & consensus here. I really would apreciate this.
a) "due to the way that the tables are grouped by architecture rather than purely by model number (for example the 1600 AF, which is Zen+ despite being in the 1000 series, is grouped with the 2000 series CPUs)" which is wrong IMHO. The other template names are clearly named by the model branding and since AMD likes to mix models with different architectures...
b) Re: Threadripper these can easily be split into there own table. Which would be more fitting anyways.
-- Wikiinger (talk) 15:39, 10 June 2022 (UTC)
OK, I've done it. I can't agree with the bit about moving the AFs to the 1000 series table though. Since the tables are grouped by architecture, moving them would mean the tables are no longer consistent since there would be Zen+ CPUs in the Zen section. The only way it could work would be to change the groups to be the numbered series instead, which would actually be more confusing for the reasons you mentioned.
But I think separating the Threadrippers out into a separate section would make a lot of sense. If you agree I'll go ahead with it. By the way I made a sandbox version of one of the tables, which should be easier to test new designs on rather than putting them in the talk page. Zerranto (talk) 19:17, 10 June 2022 (UTC)
@Zerranto @Wikiinger
I have a question regarding the Threadripper processor entries in the tables. I too strongly suggest that they should be split up from the mainstream desktop CPUs rather than combined together into one. However, if you take a look at the Ryzen 3000 series desktop processors table, there are two kinds of Threadripper - the non-Pro, and the Pro series. Each of them have their own common features different from one another, such as non-Pro TR has only quad-channel memory while Pro TR has octa-channel.
So should the Pro and non-Pro TR still be both together in one table, or should they both be split up as well?
Cheers. — AP 499D25 (talk) 01:57, 19 December 2022 (UTC)
Furthermore the templates are used in a lot of places, such as the architecture articles. So having a "full blown" sentence about the architecture doesn't make sense there.
IMHO this should be a small table (for layout purposes only, no borders), where the common facts can be seen at a quick glance. Wikiinger (talk) 17:59, 8 June 2022 (UTC)

4600G not OEM only anymore

Ryzen 5 4600G and others are not OEM only, they come with the full AMD Box with Cooler and warranty https://www.amd.com/en/press-releases/2022-03-15-amd-launches-the-ultimate-gaming-processor-brings-enthusiast-performance https://www.gsmarena.com/amd_announces_new_affordable_ryzen_5000_and_4000_series_processors-news-53575.php Terraaer (talk) 23:44, 13 July 2022 (UTC)

Spacing with noinclude

When using noinclude, make sure to put the opening <noinclude> tag immediately after the last bit of content to be transcluded. If you put it after a line break, that line break will also be transcluded, resulting in excess paragraph breaks in the target article. Hairy Dude (talk) 11:30, 5 October 2022 (UTC)

APU definition

AMD marketing has not been using the term for some years now, but there's a precise definition: it is an HSA-style processor that contains a CPU and a GPU part. GPU processing power has nothing to do with it. The Ryzen 7000 are APUs, not CPUs. And even AMD still knows this: https://www.amd.com/en/support/apu/amd-ryzen-processors/amd-ryzen-9-desktop-processors/amd-ryzen-9-7950x Trigenibinion (talk) 12:16, 30 November 2022 (UTC)

Table redesign

Hello all!

I am re-doing the style and layout of all the Ryzen CPU and APU templates, due to the following issues and concerns:

  • The tables (at the time of writing this) are done in a way that significantly prioritises their appearance, over their purpose (the purpose is to give the reader an organised, sorted, tidy collection of the CPU models so that they can be compared at a glance).
    • The choice of coloured horizontal branding "rows" (e.g. the Ryzen 3/5/7 or Mainstream/Performance/Enthusiast rows) are quite baffling. The use of colour seems unjustified. In a table that contains green "Yes" and red "No" cells for example, the use of web-friendly colours there is reasonable, as the colour coding/keying significantly improves readability. However, over here it appears to be purely for "beautification" purposes, rather than improved legibility through colour coding. I have no issues with the general use of colour for theming and to make a table less "boring" (e.g. see Template:Samsung Electronics for a good example of that), but when it induces potential readability issues by unnecessarily causing patterns of high-contrast colour (e.g. you read the table from top to bottom and it goes orange-grey-orange-grey-orange-grey; unnecessary in the sense that it's only being done for "stylistic" purposes and not to improve communication through colour keying, as per the green 'yes'/red 'no' example) it is good reason to believe that the use of colour, in that manner, is unreasonable.
    • The horizontal branding rows/headers / whatever you want to call them, cause the otherwise mergeable columns of same information, to be broken up, leading to information having to be duplicated below the branding row.
    • Furthermore, the rows also break the flow of reading columns from top to bottom, for example if you want to compare L3 cache sizes and clock speeds between Ryzen 5 and Ryzen 7 processors...
  • So because of the branding rows, there is an excessive amount of duplicated data. For example, in the Ryzen 3000 desktop CPUs table, "TSMC 7FF" is repeated six times. Together with the inappropriate use of colour, this further hurts the readability of the table.
  • L1 and L2 cache in their total amounts, are not useful information. The L1 and L2 caches are per-core, they are not shared, and so having more of it in total, does not improve performance per se. Conversely, L3 caches are shared among cores or clusters of cores, depending on the processor, and having more L3 cache (per CCX) can have significant performance implications, e.g. Ryzen 5 5500 vs 5600, and the Ryzen 7 5800X vs. 5800X3D.
  • On the note of cache sizes, perhaps the way the cache amounts are stated in the tables, could be improved further. For example for the Ryzen 9 3950X, technically it actually contains two sets of 2x16MB of L3 cache, rather than a singular 64MB. So maybe it could be written as "64MB (line break) 2×16MB (line break) per CCD" for that example.
  • The individual SKUs in the tables are ordered in a lowest-to-highest-end fashion. It is quite nonsensical, it's a bit analogous to a car dealership having its cheapest basic car at the forefront of its showroom and then its best flashiest car model hidden away at a corner a long walk away. That aside, most companies list their products in a high-end-to-low-end fashion. That includes even AMD, which if you look at this website, towards the bottom where they list the processors, they too have it in that order. An even better argument, the vast majority of Wikipedia articles about Intel processors, such as this one, have the processors listed in a high-to-low-end order. It is only with AMD processors like Ryzen, FX and Phenom that the CPUs are listed in a low-to-high-end fashion. Why the order of the CPU SKUs have to be different with a different brand of CPU is beyond me. Let's keep it all one way for all the CPU tables, regardless of make.
  • By having relatively common information such as CPU socket and process node in the table, rather than in a bullet list above the table, they widen the table with clutter for something that can be easily explained in the list.

So in summary, the tables are being re-done in a way where all of the relatively constant information (e.g. socket, fabrication process) are put into a "common features" bulletpoint list above the table, and the table contains the variable data such as core count and clock speeds. The branding rows will be replaced with branding columns instead, so that the otherwise mergeable columns like L3 cache and TDP are not broken up, and the products listed in the tables will be inverted so that it is in a high-end-to-low-end order, .

Another small touch is that the tables will be made sortable so that if the reader does wish to view the CPUs in a low-end-to-high-end order, they can do so with a click of the "Branding and Model" header.

Further note, the Threadripper CPUs will be split up from the consumer/mainstream Ryzen desktop CPU tables. The Threadripper CPUs are in a completely different segment, with different pricing category, performance category, socket, platform, I/O etc from the consumer CPUs, and it makes little to no sense to have them combined together. There have already been previous discussions by other editors about splitting up the Threadripper CPUs from the consumer desktop CPUs.

The new table layouts take inspiration from the beautifully simple and easy to read Intel CPU tables like Raptor Lake, as well as ideas from a previous discussion of the same topic at Talk:Zen 4.

I remain open to any further suggestions and feedback regarding how the tables could be styled and laid out, as well as what information shall be in them. If you disagree with anything above, feel free to discuss why. Note that significant changes like putting the common info in the table may potentially only be made after a conclusion is reached in favour of it.

Thank you all and have a good time everyone.

AP 499D25 (talk) 09:09, 24 December 2022 (UTC)

I have realised that the "Common features" list at the top breaks the flow of text on text-heavy articles such as Ryzen, while it is appropriate to have on list articles like List of AMD Ryzen processors. Huge apologies in advance to everyone for the potential disruption caused by this.
One idea I have is to make the common features list "switchable" and then add some sort of parameter to the template, so that when it is called, that list will be displayed if yes is specified to that parameter, and not displayed if no is specified.
AP 499D25 (talk) 11:26, 24 December 2022 (UTC)
I wrote "Discuss at Talk:List of AMD Ryzen processors#Table redesign before making major changes" in the edit summaries, however I meant to say discuss before reverting, rather than discuss before making major changes. By all means, if you have a big improvement to make to the tables, go for it.
I'm willing to make some sort of compromise between the current table layout and the old one if needed.
AP 499D25 (talk) 12:18, 24 December 2022 (UTC)
This is magnificent work and excellent reasoning. Thank you so much! Artem S. Tashkinov (talk) 18:44, 24 December 2022 (UTC)
I don't mind the change of the primary source references to external links, however the other reason why I'm not bothered with doing it besides exhaustion, is that I've seen a concern about it over on another talk page, particularly the "prone to link rot" part. Usually I like to 'settle' concerns about a widespread change before going ahead with it, if there's already been some opposition to it.
Also, for the common features list issue above, my current idea is to move it out of the template and instead have it article dependent, i.e. on 'list of processor' articles I would put the full features list, however on articles like Ryzen that already have some of the common features explained elsewhere, only some of the items from the list would be written.
AP 499D25 (talk) 12:54, 25 December 2022 (UTC) edited 13:12, 25 December 2022 (UTC)
Link rot equally affects links and references, so I'm not sure how and if references are better. :-) Besides we have a web archive bot which can deal with the issue. Artem S. Tashkinov (talk) 06:04, 27 December 2022 (UTC)

Discussion

There are so many issues with the tables that this will be a lengthy comment. There may be a few problems with the old table style but they did not need an entire redesign after years. I imagine this comment will probabkly be brigaded by people with too much time on their hands for daring to simply disagree with them.

1. The order that CPUs are put in is utterly baffling. All other tables for CPUs and GPUs from AMD and Nvidia are put in a simple ascending order by name which is easy to understand and is quicker to find a specific CPU in a long list. This table seems to have imported the style from Intel tables that have ranked CPUs from "best" to "worst". This approach is using a completely subjective metric that can be measured by so many things: core count, price (low to high), price (high to low), release date, base clock, boost clock, iGPU base clock speed, iGPU boost clock speed, amount of cache, memory speed, power usage (low to high), power usage (high to low), benchmark scores, and on and on. The point of a table on Wikipedia is to simply present the information in an easily understandable, non-biased order, not somewhere for some random person to subjectively rank CPUs based on one of a million metrics which essentially just amounts to an opinion outlet.

Vendors themselves list products from best to worst where best is simply the fastest. That's always been the case. No idea why WP needs to divert from this practice. It's extremely simple and quite logical. Artem S. Tashkinov (talk) 10:37, 30 December 2022 (UTC)
Vendors rankings of their own products is completely irrelevant and there is no need to just repeat corporate marketing. Subjective bias does not belong in tables. It is extremely difficult to find a specific CPU in a table that has been organised by "quality" compared to one simply organised by name in ascending alphabetical or numerical order. APD4711 (talk) 13:11, 1 January 2023 (UTC)

2. The branding column is unneccessary and hurts accessibility, not helps it. Firstly, a branding column increases the width of the table while also complaining that the old table was too wide. There needs to be a clear break between each product segmentation, particularly for people that are vision impaired. Like you mentioned with Samsung, there are some CPU/GPU brands like Intel Core and Arc use different colours for different segements. The blue colour on a Core i9 box is noticably darker than the blue used on a Core i3 box. Including these different segmentation colours in clear separating branding rows greatly help people with vision impairments. It is not applicable in this case with Ryzen as it uses the same colour for each segment. Having a clear branding row separating segments is more beneficial with a long list of CPUs rather than having a column rowspanned for 13 rows. Which table is narrower, A or B?

Why there has to be a clear break at all? Product segmentation is purely artificial and extremely contemporary. What used to be a "9" class product ten years ago is now a throw away thing sold for a hundred bucks on Ebay. I see no reason to use coloring at all. It's distracting, it doesn't convey any [additional] information except for, again, a madeup segmentation. To be honest, it's actually redundant for all I know and I'd not include it. Yes. The Core i7 7700K CPU nowadays is slower than the much cheaper Core i3 12400F CPU, so what's the use of these segments at all? Besides, extra segment rows break sorting.
Table A
Branding and Model Cores
(threads)
Clock rate (GHz) L3 cache
(total)
TDP Chiplets Core
config[i]
Release
date
MSRP
Base Boost
Ryzen 9 7950X[1] 16 (32) 4.5 5.7 64 MB 170 W 2 × CCD
1 × I/OD
2 × 8 Sep 27, 2022 US $699
7900X[2] 12 (24) 4.7 5.6 2 × 6 US $549
Ryzen 7 7700X[3] 8 (16) 4.5 5.4 32 MB 105 W 1 × CCD
1 × I/OD
1 × 8 US $399
Ryzen 5 7600X[4] 6 (12) 4.7 5.3 1 × 6 US $299
Table B
Model Cores
(threads)
Clock rate (GHz) L3 cache
(total)
TDP Chiplets Core
config[ii]
Release
date
MSRP
Base Boost
Ryzen 9
7950X[5] 16 (32) 4.5 5.7 64 MB 170 W 2 × CCD
1 × I/OD
2 × 8 Sep 27, 2022 US $699
7900X[6] 12 (24) 4.7 5.6 2 × 6 US $549
Ryzen 7
7700X[7] 8 (16) 4.5 5.4 32 MB 105 W 1 × CCD
1 × I/OD
1 × 8 Sep 27, 2022 US $399
Ryzen 5
7600X[8] 6 (12) 4.7 5.3 32 MB 105 W 1 × CCD
1 × I/OD
1 × 6 Sep 27, 2022 US $299

References

  1. ^ "AMD Ryzen 9 7950X". AMD. Retrieved August 30, 2022.
  2. ^ "AMD Ryzen 9 7900X". AMD. Retrieved August 30, 2022.
  3. ^ "AMD Ryzen 7 7700X". AMD. Retrieved August 30, 2022.
  4. ^ "AMD Ryzen 5 7600X". AMD. Retrieved August 30, 2022.
  5. ^ "AMD Ryzen 9 7950X". AMD. Retrieved August 30, 2022.
  6. ^ "AMD Ryzen 9 7900X". AMD. Retrieved August 30, 2022.
  7. ^ "AMD Ryzen 7 7700X". AMD. Retrieved August 30, 2022.
  8. ^ "AMD Ryzen 5 7600X". AMD. Retrieved August 30, 2022.

3. Actually useful columns have been removed without reason. If there is a column, for L3 cache, you also need to include the full information like L2 and L1 cache. L1, L2 and L3 cache varies between CPUs so it needs to be specified. The 7700X has a different amount of L1, L2 and L3 than the 7900X or 7950X. Not including L1 and L2 cache while including L3 is like including CPU boost clock but not base clock. Another column that has been removed is core configuration and chiplets. Core configuration is actually incredibly useful for Epyc tables as the number of CCDs for example dictates the amount of L3 cache. Ryzen and Epyc CPUs are very similar in being built on modular chiplets so it makes sense for consistency purposes to include core configuration on the Ryzen tables. Memory support is useful for comparison when Ryzen tables are put side by side. For example, DDR5 will mature throughout the years and it is a guarantee that Zen 5 or Zen 6's memory controller will support higher DDR5 speeds than 5200M/Ts, just as Intel's Raptor Lake supports DDR5-5600 while the previous generation Alder Lake supported DDR5-4800. Memory speeds change throughout generations so it should be included so that it can be compared to other generations. The number of memory channels also needs to be specified as Ryzen desktop CPUs may support 2 channels while the upcoming Threadripper 7000 will support 4-channels or 8-channels. On the other hand, there are a few columns like fab and iGPU processing power that can be actually removed.

Nothing has been removed, as the duplicated/repeating data has been moved out of the the tables. What's the point of listing L1/L2 caches/memory support which are the same for all the CPUs? It's exactly the same per core for the entire architecture.

4. The order of columns was unnecessarily changed like the release date and price being moved to the end of the table when it was perfectly fine at the start. It also makes more sense to put the chiplet and core config columns beside the cores/threads column.

Wikipedia lists already released to the market products. The release dates thus do not convey any valuable information.

5. The essay should not be included in the template which should be exclusively reserved for the table itself. If it is necessary, it should be added to the page itself and not the template because the template can be used in different circumstances when a wordy list of descriptions would not be appropriate. A long, wordy set of bullet points defeats the entire purpose of a table and it's long description of CPUs . The logical conclusion of putting more and more details into a long essay is that the table will be essentially useless at 3 columns wide and reading a long tedious block of text means that information cannot be easily found. If you want to remove as many columns as possible from the table, you may as well go the whole way and just delete the entire table and write a 4 paragraph essay describing the details of each CPU individually.

So, let's make tables 3000 pixels wide and include every CPU metric under the Sun? Where's a WP rule which states that the template must not contain a bullet point list? Let's see, https://ark.intel.com/content/www/us/en/ark/products/134599/intel-core-i912900k-processor-30m-cache-up-to-5-20-ghz.html contains around 50 data points. Let's include all of them right? Even the ones which are common for all the CPUs in the arcitecture? Is AMD better? https://www.amd.com/en/products/cpu/amd-ryzen-9-7950x - at the very list 24 data points.

What this unnecessary redesign misses is that the fundamental purpose of a table is to provide comprehensive information at a glance, not to fill up white space on a page with only a few columns. Trying to make the table narrower doesn't make sense when actually useful information is being sacrificed and the table only includes the most basic, surface level details like clock speed and price. I am fine with some changes being made to the old table but I am yet to see a solution that actually addresses these issues. APD4711 (talk) 17:47, 29 December 2022 (UTC)

Comprehensive, not extremely redundant. Artem S. Tashkinov (talk) 10:37, 30 December 2022 (UTC)
First things out of all, I just wanted to say thank you for taking the time to express your thoughts in full on the matter. I wanted to hear a full list of reasons why the old table design could be strongly preferred.
"I imagine this comment will probabkly be brigaded by people with too much time on their hands for daring to simply disagree with them." - don't worry! After all, we are trying to reach a consensus here.
--------------------------------------------
1: Order (hierarchy) of CPUs
"This table seems to have imported the style from Intel tables that have ranked CPUs from "best" to "worst"." - the reason for doing this was that the Intel CPU tables, for many many years since their beginning, were done in a highest-to-lowest-performance order, while the AMD tables in general, come a bit later after that. It was done based on how the Intel tables were seemingly well-established for many years. I figured the oldest and longest-lived consensus shall take place. It's worth noting I did not make any sort of "personal analysis" here, I simply flipped the entire table upside down, i.e. made the items descending order, without shuffling any items around.
On the topic of order of items, one thing I disagree with is making the items alphabetical order. Take i3-6100T for example, it is inferior to i3-6100 in performance, yet the alphabetical sorting would put it after it. Another example is i7-10700K vs 10700KF. When you read the table from top to bottom it becomes quite bamboozling, as the performance and features order of the CPUs seem quite randomised. I'm okay with changing the order from descending back to ascending again, but alphabetical order doesn't make any good sense to me. The order of CPUs by performance is something that has been established for long time on Intel tables with almost no issue.
--------------------------------------------
2: Branding column vs row
What can I say, there are quite possibly a thousand different ways that a table could make use of screen width. Release date and pricing could be combined or made separate, L1 and L2 cache could be included in or moved out of table, etc.
I agree that the use of branding rows does save some table width. One major issue that you may or may not have noticed however, is that it causes otherwise mergeable columns, like the 32MB L3 cache between R5 7600X and R7 7700X, to be forcefully split up, leading to duplication of that same information. This point only becomes clearer and more emphasised as AMD releases more CPU SKUs and we add them to the table.
If you know a thing or two about website design you might know that the use of colour should be minimal, in a way that improves readability, and not used in an "overarching" way that negatively affects the readability of information.
One more issue with branding rows, is column-to-column reading. Oh, what's that "5.3" number for. I look up, suddenly there's a big orange row that has no vertical guidelines in it to guide my eyes up. So I have to look up more closely. This may very well slow down reading for some people...
What do you think of vertical brand columns, like the table layout 1 in this sandbox? It's minimal use of colour and table width.
--------------------------------------------
3: Removal of some columns
Like mentioned above, there are all sorts of ways a table could make use of screen space. I removed L1 and L2 cache columns because once again, they are per core, they are not shared, so having more of it in total does not mean faster CPU per se, so it is very misleading. Now, I didn't mention this in the original post above due to time and word length considerations, but GPU core config and processing power GFLOPS columns were also removed because most people probably don't care about shader count, ROP count, processing power, that kind of stuff about the iGPUs. As Rando717 put it in a previous discussion on Talk:Zen 4, " Nobody is gonna be like "Look! Ryzen 5 4680U with 60 more shaders and 1344 FLOPS compared to 1152 on Ryzen 5 4600U." ". It's also slightly misleading as a 1.8TFLOPs integrated graphics is not going to match a 1.8TFLOPs discrete GPU in every scenario, due to memory bandwidth differences.
Basically I'm spending the table width on what matters to the average people, and removing what the majority of readers would likely find useless, with the help of some consensus from the previous discussion.
Do you think the average, non-technically-minded reader is going to care about trivial things like L1 and L2 cache size totals?
Idea: if there has to be L1 and L2 cache columns, instead of writing 512KB, 8MB for the R7 7700X for example, how about we put it as 8x64KB, 8x1MB instead? It makes more sense from a technical point of view. I've taken note that tables for older AMD CPUs like FX or Phenom are like that.
"Another column that has been removed is core configuration and chiplets." - Pay close attention, I did not remove them, but rearrange them to the right side of the table, after TDP but before price/release date.
Inclusion of DDR memory support and speeds in table - you actually have a very good point here! It may definitely be good to move some items from the bulletpoint list back into table if needed for emphasis.
--------------------------------------------
4: Arrangement of columns
"when it was perfectly fine at the start." - Tbh I've hated this table layout since its very beginning.
Anyways, the relocation of some of the columns was done also with the consideration of the average reader wanting to learn more about tech. Core count is very significant information, definitely major implications on multi-threaded perf. What is very likely not so important however, is chiplet and core configuration. They are highly technical information, with very little performance implication. Do you think the average person learning about CPUs is likely to care about those things as much as the core count and GHz of the CPU?
Similar reasoning for release date and price too. What proportion of people are actually wanting to know about those things first, rather than the performance and main specs of the CPU?
--------------------------------------------
5: Bulletpoint list above the table
This point is something I highly agree with. As I mentioned in a comment earlier on, I have realised that including a list in the template results in disruption in text flow on some articles such as Ryzen, as they already cover some of those items somewhere else. I was intending to move them out of the templates and have them article dependent quite soon anyway.
--------------------------------------------
I may consider starting a request for comment (RfC), as the number of editors willing to be involved in this topic seems very low, not enough to be able to achieve conclusion effectively, and this is a wide issue that affects not only AMD CPU tables, but also potentially Intel CPU tables, and all the GPU tables too, with disputes happening across all venues. I shall retract from any further edits to any tables while discussion is in progress.
Good table design should not only be easy to read, but also scale well with number of items in table, regardless if it's only 4, or if it's 25 items.
AP 499D25 (talk) 04:30, 30 December 2022 (UTC)
In my opinion your new tables are much harder to read. I reference these tables regularly and am pretty disappointed to see they've all been changed.
Just because you didn't find value in some of the information presented in the tables doesn't mean others didn't. Some of your ideas have merit but overall these changes are for the worse. Pathogen-David (talk) 10:00, 31 December 2022 (UTC)
Hello, thanks for joining in the discussion regarding the table changes.
Could you clarify what are the issues with the current tables?
For example, the relocation of some items from inside the table to a bulletpoint list above the table?
It's worth noting the only items I actually removed (deleted) from the tables are the integrated GPU configuration and processing power GFLOPS columns. Common features such as socket have been moved out of the table into the list at the top, and some columns like chiplet and release date have been relocated. AP 499D25 (talk) 10:44, 31 December 2022 (UTC)
Could you please clarify what's now harder to read? The tables have become narrower and concise with the duplicated data moved out of them. "overall these changes are for the worse" - again this needs clarification 'cause it really sounds like an issue of "tastes differ" while we are discussing the merits of not including everything under the sun in them thus making them unyieldingly large and wide. Artem S. Tashkinov (talk) 09:59, 1 January 2023 (UTC)
They are harder to read because the CPUs have been put into some personal ranking by "quality" rather than simply by using a standard ascending order by name, there is no clear separation between segments which is particularly important woith longer lists, and the order of columns are seemingly at random. I am yet to hear any consideration from fans of the unreadable redesign how they don't hurt accessibility for vision impaired people. There is no reason to remove most of the actual details just to make the table "narrower" because doing so just defeats the entire purpsoe of a having a table which is to provide information. This redesign was utterly unneccessary, makes having tables at all redundant since the information they provide is not comprehensive or is just the most button of the barrel, basic, surface level possible. The logic of wanting to remove actual useful information just to make a table narrowwer ends with a useless table with 3 columns while a long tedious essay provides extensive descriptions of each individual CPU. The old table design didn't need to be entirely removed and replaced by something worse. APD4711 (talk) 12:57, 1 January 2023 (UTC)
There's nothing "personal" for listing products from fastest to slowest and exactly the same way as vendors list them. If anything it allows to find products faster and easier. Nowhere the tables are listed according to "quality", it's just performance.
"the order of columns are seemingly at random" No, they are not. They list the most important data points starting with the branding, then goes the number of cores, then go maximum frequencies, then goes the L3 cache. Everything else is redundant and doesn't require any order and can be reordered if you like so.
"don't hurt accessibility for vision impaired people" - previous tables were 1000% unusable for visually impared people because finding informating in tables containing over two dozen columns is impossible. A bullet list listing common features on the other hand? A breeze to get acquainted with.
What exactly is "unreadable redesign"? Was the previous 2000 pixel wide design where corresponding rows and columns became completely impossible because tables became humongous?
"This redesign was utterly unneccessary" I can say the same about the previous attempt to include everything under the Sun in the tables thus making them impossible to view on even desktops. Some tables previously required more than 2000 pixels in width to see them fully.
"not comprehensive or is just the most button" The goal of the tables is to provide data points which are different between products. The goal was not to include arbitrary important to you data points. I've already voiced my concerns about that over ten times - you've never adressed them. I'm sorry but I know exactly who you are. You previous redesign threw in new exactly the same data points willy-nilly because you alone thought they were imporant.
"wanting to remove actual useful information just to make a table narrowwer" exactly.
"while a long tedious essay" which is 20 times easier to read than navitaing through the table which is 2500-pixel wide.
"The old table design didn't need to be entirely removed and replaced by something worse." The design before you intervened was OK'ish but could have been improved which has been achived through conensus. You on the other hand without reaching any consensus and while using sock-puppets try to push your own view of the "best" design against any common sense or logic. Please demonstarate existing comprehensive Wikipedia product tables which include multiple data points which are exactly the same across all the listed products. Your redisign included literally up to a dozen of such data points. Artem S. Tashkinov (talk) 19:13, 1 January 2023 (UTC)
I'll say it one more time that I did not make any sort of "personal ranking" ranking here, rather, simply flipped the table contents so they are in descending order instead of ascending.
--------------------------------------------
"there is no clear separation between segments which is particularly important woith longer lists." - I don't understand why this is needed? Look at Intel CPU tables and especially look at AMD/Nvidia GPU tables, they don't have any "segment rows" to separate segments of the many listed GPUs at all and yet almost no one ever complains about poor readability or "not enough separation".
--------------------------------------------
"the order of columns are seemingly at random" - tell me why chiplet, core config, release date, original launch price are as much important information as the main specs of the CPU such as the core count and GHz? When you buy a new CPU you don't ever choose one by how many chiplets there are or what the core config is, you buy one by how many cores it has and what the GHz of the CPU is.
The pricing becomes irrelevant information within a year after the CPUs' release anyway, as the CPUs will often get discounted below the original launch price especially as the next gen parts arrive. After that, who cares about the original launch price of a CPU?
It's a good reason why the Intel tables have the pricing all the way towards the right.
--------------------------------------------
"I am yet to hear any consideration from fans of the unreadable redesign how they don't hurt accessibility for vision impaired people." - fine I'll take both table designs to someone who knows a lot about accessibility and web design rules and have them comment on which is the objectively better one, in terms of readability.
AP 499D25 (talk) 14:06, 1 January 2023 (UTC)
Since your redesign is objectionable and there was no consensus on it before it was added, can you please revert all tables to the old style? There currently is not support for the rdesign. APD4711 (talk) 13:01, 1 January 2023 (UTC)
While I admit the current table layout probably isn't going to stay forever and some further changes may be made to it in the foreseeable future, I do not believe that the current table design is fundamentally flawed enough to warrant complete reversions. The fact that things like "Some of your ideas have merit" and "There may be a few problems with the old table style" have been said indicate that it is not and there are definitely some things that needed to be changed regarding the old design. But I'll believe it and change course when numerous other editors jump in and express total opposition for the new design and not just you :) .
"There currently is not support for the rdesign." - I did not make this new table layout all by myself, it was made with the help of a few points and achieved consensus from a previous discussion over at Talk:Zen 4. The discussion there shows that there is not just one, but at least three editors who are in opposition of the old table layout and were in support of a redesign or major change. Have a read through that discussion mate, there were not just one, but several people there proposing for big changes to be made to the tables. Pretty much no one there expressed support for the old design.
You may have noticed another person has jumped into the discussion above to complain about this table redesign. I am just waiting for them to expand upon and clarify what exactly are the readability issues with the new design. Just saying "these changes are for the worse" isn't specific enough.
Btw, the old table layout (the one with the coloured branding rows, and every info in table instead of some in a list above) has been applied numerous times to Intel CPU tables before in the past only to be reverted by not just one, but several editors. So that's yet another editor at least in opposition of that table design.
I'll also say it yet again that currently there's probably not enough editors involved in this discussion here to form effective and proper conclusion. I'll wait a few days more to see if anybody else joins in, and then ping in a few editors around to really start the discussion.
--------------------------------------------
You have made some very good points in the previous reply though (e.g. list at top of table should not be in template), I have heard your voice and believe those points make strong sense from reader's perspective, and so I will be intending to make the following changes to all the tables soon:
  • Move common features list above table out of template and make it article dependent
  • Move DDR memory support info back into table
  • Move socket info back into table
  • Move PCIe lanes info back into table
I may also consider putting the L1 and L2 cache columns back, except they won't be expressed as totals, they will be written as "[core count] x [cache per core]". So for example 8x1MB L2 cache on 8-core R7 7700X.
AP 499D25 (talk) 14:42, 1 January 2023 (UTC)
I see zero issues with templates having bullet points. AFAIK not a single WP rule says it's not advisable.
I'm very much against including the information which is again exactly the same for all the products in the table including the socket, L1/L2 caches, RAM support and PCIe lanes (AFAIK AMD CPUs of the same lineup are exactly the same in this regard). Right now the tables look perfect IMO. Artem S. Tashkinov (talk) 19:19, 1 January 2023 (UTC)
Well that didn't last very long...
Anyways, I think I'll go ahead with just the first point only, and leave everything else alone (for now).
I won't be putting too much weight on APD4711's sayings, will wait for in-depth input from at least two other editors before making any other changes to the tables.
AP 499D25 (talk) 01:16, 2 January 2023 (UTC)
Again, let me start a new discussion right here for Mr. APD4711 and his previous six banned accounts. Please definitelively answer these questions:
1. Which additional data points should be included in the tables vs how they are presented as of today January, 1st, 2023?
2. On what grounds these duplicating data points should be included and who decides that?
3. What's the exact point of showing exactly the same data points across all the products in the table? How do these duplicating data points aid anyone exactly?
Let's check, e.g. the Zen 4 uArch: https://www.amd.com/en/products/cpu/amd-ryzen-9-7950x + https://www.techpowerup.com/cpu-specs/ryzen-9-7950x.c2846 + https://en.wikichip.org/wiki/amd/ryzen_7/7700x - where do we stop exactly?
What about we follow you PoV and include everything because you personally believe it's all important and necessary for visually impaired people? Do you really believe these people need to know that e.g. such CPUs feature "word size: 64 bit (8 octets, 16 nibbles)". Why not include it as well? No? Then what about generating the tables which show differing features of CPUs and leave everything else outside of the tables? Um, exactly like they look like as of today? Artem S. Tashkinov (talk) 19:32, 1 January 2023 (UTC)
You genuinely seem like an emotional child when an adult suggests that the table not be three columns wide. Take some time to learn what the purpose of a table actually is and learn the correct order that numbers and letters go in while you're at it. APD4711 (talk) 20:15, 1 January 2023 (UTC)
Since you are so aversed to the very concept of a table provinding useful information, I wrote an extended essay for each CPU that you might prefer:
  • Ryzen 9 7950X, the bestest CPU in the WORLD!!!!, has 16 cores and 32 threads, with a base clock of 4.5 GHz and a boost clock of 5.7 GHz. It has 64MB of L3 chache, don't worry about the other cache as you don't have the attention span for it. It cost $699 at launch but don't worry about when it was actually launched because that is not relevant information.
  • Ryzen 9 7900X has 12 cores and 24 threads with as base clock of 4.7 GHz and a boost clock of 5.6 GHz. Stating the amount of cache it has is irrelevant because it si dUpLicAteD InForrmAtIoN that will lose your short attention span.
  • Ryzen 7 7700X has 8 cores and 16 threads with 32MB L3 cache
  • Ryzen 5 7600X, the most terrible of the Ryzen 7000 series, has just 6 cores and 12 threads with a base clock of 4.7 GHZ and a boost clock of 5.3 GHz
APD4711 (talk) 20:24, 1 January 2023 (UTC)
"Ryzen 9 7950X, the bestest CPU in the WORLD". Currently there's no such statement in WP. At the same time it is the fastest in MT workloads among the released Zen 4 desktop CPUs. Only under rare circumstances it can be slower than 7700X in ST scenarious but exceptions are exceptions.
"don't worry about the other cache" L1/L2 caches are exactly the same for all the Zen 4 CPUs and the information is provided. Also, since it's the same it's the L3 cache (and its configuration) which affects performance. L1/L2 has zero performance impact. The operating frequency provides it in addition to the L3 cache.
You're correct in that Cores * Operating frequencies * L3 cache ultimately decide the CPU performance and that is what actually matters. Other data points you're looking for are irrelevant both in terms of their performance impact and importance to the end user. Also these other data points are indeed included. You have not answered any of my questions which ultimately means that you have zero arguments for what kind of data to include, and why include the redundant data in the first place. Thank you for your input. The consensus for leaving the tables as they are now sticks. Artem S. Tashkinov (talk) 21:04, 1 January 2023 (UTC)
Lastly you 1) edit articles without reaching a consensus 2) You leave insulting summaries 3) You continue to think that your opinion trumps everyone else's without providing any arguments. This is not how WP works. Some of your other edits are indeed valuable and it's sad to see you go. Artem S. Tashkinov (talk) 21:13, 1 January 2023 (UTC)

For the record:

  • I support sorting the tables by increasing compute power/increasing core count/increasing brand level. This go with sorting the sections with chronological order, so the farther a reader go in the page, the more powerful processors are. Also in line with Special:permalink/1129303504 & Special:permalink/1130911085, with the sorting of numbers (in general), with Timeline of computing 2010–2019.
  • Either inside or outside the tables, i oppose mentioning the total amount/quantity of L1 cache and L2 cache (as opposed mentioning the amount/quantity of L1 cache and L2 cache per core).
  • I agree with the other changes made by AP 499D25.

Visite fortuitement prolongée (talk) 10:14, 3 January 2023 (UTC)

Thanks for the input. I may go ahead with certain major changes like reorder the CPUs if they get enough support. AP 499D25 (talk) 14:29, 3 January 2023 (UTC)
All the tables are sortable now so I guess we could put to rest any sort mode disagreements. Artem S. Tashkinov (talk) 19:10, 3 January 2023 (UTC)

Wall of text

People have not been following this. You went and removed lots of information from lots of tables. All the GFLOPs figures in APUs are gone. Trigenibinion (talk) 03:56, 4 January 2023 (UTC)

If GFlops between products in the table vary I see no reason not to have them. As for "lots of information from lots of tables" - please be more specific. AFAIK, almost everything has been preserved but moved out to the bullet point lists which are easier to navigate than extremely wide tables. Artem S. Tashkinov (talk) 09:44, 4 January 2023 (UTC)
Yes, there are big differences in GFLOPS between models. Those numbers guide you in selecting and APU and the information is present for GPUs. Also the GPU config should not have been removed either, as that is what is used to compute the processing power. Trigenibinion (talk) 15:06, 4 January 2023 (UTC)
I may consider adding them back seeing the support here, thank you. I just don't want the tables to include every single nitty gritty of trivial info and turn into WP:INDISCRIMINATE. AP 499D25 (talk) 01:20, 5 January 2023 (UTC)
CU (compute units) and frequency give more than enough understanding of what kind of performance you're looking at. They are a lot easier to read than purely theoretical GFLOPS which are interesting to 0.001% of people on Earth and also no not convey actual performance. E.g. in terms of GFLOPS AMD has been leading for more than a decade yet NVIDIA GPUs with fewer GFLOPS run faster. I prefer not to see them here. Artem S. Tashkinov (talk) 10:41, 5 January 2023 (UTC)
It seems the box/part numbers and cooler models were removed previously. Trigenibinion (talk) 15:27, 4 January 2023 (UTC)
The part number and cooler model columns probably weren't there in the first place on some of the tables as the old tables didn't have them either. AP 499D25 (talk) 01:16, 5 January 2023 (UTC)
Cooler models were never here in the first place and this is such trivia I find its importance highly dubious. As for part numbers - these are interesting maybe to large retailers. Artem S. Tashkinov (talk) 10:41, 5 January 2023 (UTC)
Many people just want to get the boxed cooler. Knowing which one is bundled is important. The coolers were there before in APU and Ryzen tables. Trigenibinion (talk) 21:19, 5 January 2023 (UTC)
1) Bundled AMD coolers have always been barely sufficient and quite loud, so most people replaced them with ~$30 coolers which work better (quieter and keeping the temps low). 2) When you're looking for a CPU to buy you see if the cooler is bundled and its description on the web store page. 3) In many places the AMD CPUs which are meant to be sold with a stock cooler are sold without it as an OEM product, so again, the product listing is more important that whatever WP is listing. I'm not against specifying a cooler but to me it looks extremely redundant and almost misleading. Artem S. Tashkinov (talk) 11:02, 6 January 2023 (UTC)
I have a bit of an idea for how we should handle all this trivial info like included box cooler, iGPU GFLOPS, and such. How about we include it on the list articles such as List of Ryzen processors, but on articles like Ryzen, Zen 4, the tables are kept down to the essential important information?
This will require duplicating / making copies of the templates though, one to use on list articles and the other on normal / non-list articles, which is a disadvantage. Currently most of the templates are re-used on as many articles as possible, which is a good thing in terms of updating and maintaining them.
I've noted that some articles for older AMD CPUs like Phenom and Athlon are done in a similar way, the main articles contain just the essential info whereas the list articles contain nearly all the detail and trivia about each processor model. Except, templates aren't used on the articles for older CPUs.
AP 499D25 (talk) 12:21, 6 January 2023 (UTC)
This is not trivial info, especially GFLOPS. The tables are now dumbed down to Intel's level. Trigenibinion (talk) 12:47, 6 January 2023 (UTC)
Another issue with GFLOPS numbers besides what Artem S. Tashkinov mentioned, is performance numbers between architectures. A one teraflop TeraScale iGPU is not going to perform the same as a one teraflop Vega iGPU, and especially neither is it going to perform same as one teraflop RDNA GPU. Due to vast architectural differences and improvements, even if a newer GPU's GFLOPS number is the same as an older gen one, it's still going to perform quite a bit faster, due to those improvements.
The HD 5770 1GB graphics card in my old desktop has a TFLOPS number of around 1.3. I bet that thing doesn't hold a candle to even a 1.0 TFLOPs modern Vega or RDNA 2 iGPU. AP 499D25 (talk) 00:14, 8 January 2023 (UTC)
There are some new features in new architectures, but GFLOPS show you the innate power of the GPU, especially in OpenCL. Removing such information is a disservice when there is such wide variation. Or do you now want to remove those figures also from the GPU pages? The figures from RT and Tensor are more problematic given that they are not so general purpose. Trigenibinion (talk) 00:00, 9 January 2023 (UTC)
One should not only look at the GFLOPs but also the memory bandwidth (which for iGPUs is OK to show in terms of memory channels and MT/s). More important than looking at API feature levels would be video capabilities. Trigenibinion (talk) 00:07, 9 January 2023 (UTC)
That may be true for Intel coolers. AMD ones have typically been decent, just not designed for overclocking. Trigenibinion (talk) 12:46, 6 January 2023 (UTC)


PLEASE GET THE GFLOPS BACK !!!

Hi everybody. I created an account just to be able to say this: whoever decided on its own to delete all GFLOPS info was wrong!!! Allow me to explain:

  1. I've been coming to this page for quite a long time now. Especially in the last year or so. And one of the main specs that I like to compare every single time is the IGP theoretical power in GFLOPS. It's a very good raw indicator of the difference between, say, a 6600H and a 6800H. And it's one of the main reasons that I've been coming to this page: because everything was there, concisely, precisely, in nice tables with nice figures. Today I've lost about 20mins, suddenly not seeing the GFLOPS info anymore, trying to get it back somehow... until I finally understood that somebody had decided to just delete it for no reason! Who would do such a thing? The info was already there, it wasn't doing any wrong to anybody, so why on Earth delete it???
  2. Said editor was talking about "the average user". Do you really think that "the average user" would come to Wikipedia and compare CPU specs in tables? NO! The average user wouldn't even do that. People coming to this page are "above average" users, and to such people, info such as GFLOPS really matters.
  3. I don't care about technical discussions about the importance or the relevance of such info. I won't enter such discussions, because that's not the subject here. It's not up to Wikipedia editors like Artem S. Tashkinov to decide on their own whether an info should matter or not to "the average peole". That info was already here, so it should remain publicly available instead of being unilaterally censored, period.
  4. I will strongly support IMMEDIATELY REVERTING to previous tables where GFLOPS were clearly indicated. I don't care about the tables format or any eye-candy. I only care about useful info such as GFLOPS remaining on this page, as it should be. There's no reason to do otherwise, so please, no more discussing on that matter.

Thank you all for your attention, and sorry about the rant, but I think it was necessary. --Jose de la Mancha (talk) 19:00, 12 January 2023 (UTC)

Hi there, thanks for expressing your thoughts on this matter.
I get that you feel strongly about the removal of integrated GPU GFLOPS and config columns, and want them to be brought back.
However, there are a few key points worth mentioning here:
  • Wikipedia works on consensus. That means, even if 2 people highly support this change to happen to an article, if 20 other people oppose that change, then the opposition takes form as the consensus. As it is now, only one other person has expressed their support for the re-addition of GFLOPS and GPU config. So I'll wait for at least one more editor to jump in and support the addition before making the change. Additionally, if another editor comes in and opposes the changes, then some further discussion will be needed before we get to making the changes.
  • The GPU config and GFLOPS info weren't just removed arbitrarily / randomly. Although I didn't point this out clearly above, one of the major issues with the old table layouts were that they were taking up way too much screen width for what little useful, actually worthy and non-redundant information they provided. Socket? Well, that's the same for all the processors, so why not move it out into a list above the table instead. A few things just had to give way here to make the tables legible on smaller screens and devices (e.g. tablets, phones), and there was actually already a previous discussion at Talk:Zen 4 about removing those columns. Combine the two together, and it was a good reason to go ahead with the removal.
  • There is no need to revert the entire table layout change to get the additional GPU info back, when all that's needed to be done, is just add a few columns and copy some bits of info from the old table from the template's history.
  • Lastly, there exists a Wikipedia rule called Wikipedia is not an indiscriminate collection of information. Tables should contain essential, relevant information to the average reader, such as core count, GHz, and cache size in the case of CPUs, and should not contain an excessive amount of trivial information that clogs up significant table space. If it's something that only 0.4% of readers will understand / find important, and it hurts the readability of the table, then it should not be included. Unless there's a strong consensus for doing so.
So in summary, the GPU config and processing power info weren't just removed for no reason. There was already a previous discussion elsewhere about it for removing it, together with the need to cut down table bloat, it was done with valid reason. For it to be added back again, I would like to see consensus from numerous others in favour of it.
Hope that helps a lot.
AP 499D25 (talk) 02:22, 13 January 2023 (UTC)
Hi there, thanks for a very articulate reply, that's always appreciated :) Rest assured that I get your points. I hope you'll get mine.
  1. I hear your argument about table layouts taking too much screen width. But other solutions remain. You could put the info outside the tables (in bullet points, etc). Or just create additional tables below the tables. If the problem really is width, then surely there won't be any problem with height, right ?
  2. Even better, just create a single extra table at the end of the page, only with all the GFLOPS info for all IGPs. People needing that particular info will get it, and other people will never have to look at that table. Problem solved and everybody would be happy. My point is : just find whatever alternative layout suits you, but please keep that info somewhere (anywhere) in the page. That sounds like a reasonable request to me.
  3. I also hear your argument about Wiki not being an indiscriminate collection of info. But IGP raw power in GFLOPS is not exactly trivial. Take my previous example : how are you going to quickly compare a 660M and a 680M? Did you know that the 5700U IGP is actually much more powerful than the 6600H IGP? How could you quickly tell that without the GFLOPS ? They're easy to read, and they give a perfectly good raw/global estimate of the difference between two IGPs. We're not trying to compare them to NVidia or whatever. Just to each other, that's all (which should be one of the purposes of the page : not only comparing CPUs, but also comparing IGPs).
  4. Besides, the info was already in this page, and people with me were actually using it. Maybe you weren't aware that people like me were finding such info useful, just because you didn't find it useful.
  5. To me, removing non-trivial technical info just for the sake of presentation / table width can never be a good thing. Tech should always be above eye-candy.
  6. Lastly, I agree with the consensus part. I have just read the previous Talk:Zen 4 discussion, but (forgive me if I'm wrong) I didn't see any real consensus, just two opinions from Artem and yourself. That makes 2 people supporting that change, and now 2 other people opposing it (in just a few days/weeks since that change happened). Sounds like rather balanced so far. And divided opinions shouldn't be enough for removing the info in the first place. To me, once some info is on Wikipedia and has been for quite a long time, then removing it should at least require some kind of democratic majority, not just even scales (and enough people supporting change, not just a couple of them). It should be about percentages : 50-50 ? Don't change. 80-20 ? Go ahead. You get my point. ;)
Thanks again. Jose de la Mancha (talk) 04:01, 13 January 2023 (UTC)
While I would have waited for one more editor to hop in and support the change before making it, due to the objective, logical, and well-explained reasoning here, I will be going ahead with the returning of the GPU processing power column without hesitation.
One question: did you find the GPU core config (e.g. TMUs, ROPs) also of strong value by any chance?
AP 499D25 (talk) 13:56, 13 January 2023 (UTC)
First of all, I absolutely thank you for being a logical person and behaving like such. In my mind, logic must always prevail, and in any argument, I'm always ready to say "OK, you're right" when logic ends up being on the other person's side.
Sorry again for my previous ranting, probably due to the fact that I was quite upset with the deleted info and my wasted time. Besides, I didn't know who would reply, or if somebody would even reply. Unfortunately, few people in today's world are ready to sustain a logical discussion and find some common ground with the other person. You are one of those few, so thank you for that. There may be hope in humanity after all. :)
To answer your question in the most honest way possible, while in theory all tech info has its relevance, I have to confess that for a quick comparison between same vendor IGPs, my eyes go first to GFLOPs. I know it's not ideal, but like I said it's a number (easy to read) and it gives a good estimation. So thank you for bringing it back. Maybe some experts need additional info for precise comparison purposes. I personally don't think I need it.
We could think of additional info (like real test scores : 3DMark, Time Spy, etc). That would be even better than GFLOPS. But since such scores may slightly vary between two identical IGPs, there probably are better websites for that, and Wikipedia can remain for official manufacturer specs, including GFLOPS. Jose de la Mancha (talk) 16:53, 13 January 2023 (UTC)
Small update: I see the GFLOPS are back, so thanks again for that.
I see that they're missing for the 6xxx and 7xxx though.
If I remember well, that info was present in past versions of the page at least for 6xxx series. 2A01:E34:ECA6:D370:CCD4:DD55:4C8B:7CBD (talk) 14:25, 14 January 2023 (UTC)

GPU config still missing

Now that you're putting the GFLOPS back, you also need to restore the GPU config, as the processing power is calculated from that information. Trigenibinion (talk) 14:13, 15 January 2023 (UTC)

Would that really be necessary? I don't mind, but to me only the calculated result matters.
E.g. when a CPU that can be overclocked has a stock core frequency of X GHz, we don't necessarily mention its voltage, even though that voltage has a direct influence on the frequency and thus is necessary to its calculation.
Same thing for RAM sticks, etc.
PS : the GFLOPS are still missing for 6xxx and 7xxx mobile CPU IGPs. Jose de la Mancha (talk) 00:53, 16 January 2023 (UTC)
Yes, it is necessary because otherwise there's no way to verify the GFLOPS numbers or to calculate new ones. Also, the config also contains varying quantities of texture units and ROPs, which are needed for other calculations and which give a better idea of a GPU's performance. The clocks are published, voltages and GFLOPs generally not. There's no obvious relationship between voltage and frequency to derive the latter from the former, while the GFLOPS are are the result of a multiplication involving the shader count. Now after this mess, we're not even sure if the values put back are correct. Trigenibinion (talk) 01:23, 16 January 2023 (UTC)
Thanks for your concerns, I will return the iGPU config columns also. AP 499D25 (talk) 03:29, 16 January 2023 (UTC)

Dates -> DTS

Would be nice if the dates in the tables were converted from plain English to the { { dts } } template. Thanks. Artem S. Tashkinov (talk) 11:17, 5 January 2023 (UTC)

Will be doing that. Yet another task to cure my boredom. AP 499D25 (talk) 12:14, 5 January 2023 (UTC)
Your boredom has destroyed a lot of work. Trigenibinion (talk) 12:50, 6 January 2023 (UTC)
Archive 1Archive 2


Cite error: There are <ref group=note> tags on this page, but the references will not show without a {{reflist|group=note}} template (see the help page).
Cite error: There are <ref group=lower-roman> tags or {{efn-lr}} templates on this page, but the references will not show without a {{reflist|group=lower-roman}} template or {{notelist-lr}} template (see the help page).