Talk:List of Nvidia graphics processing units

Edit request

Hello, could someone who knows how please update the 9 series table to show that a 4gb GTX 950 is available? I can't seem to figure out how to do it without making a horrible mess of the table. Source: https://www.amazon.co.uk/MSI-Nvidia-Gaming-Graphics-Card/dp/B00UHHXXH8 Thank you! — Preceding unsigned comment added by 82.44.70.123 (talk) 07:31, 27 February 2019 (UTC)

Who ruined the tables ?????

These tables are now a mile long sideways. This wasn't the case a while ago. Now i have to scroll left and right ? They were much denser initially and fit perfectly in one page. I someone with a 4k screen editing these ? Just cause it fits your 4k screen doesn't mean it's usable for the rest of the planet. — Preceding unsigned comment added by 90.253.76.35 (talk) 17:31, 3 May 2018 (UTC)

I concur. I recently did an edit on some tables removing API data from them and placing it at the top of the corresponding sections since the info is the exact same for the entire product stack with no variations. Someone reverted it. The reason I did it in the first place is that given no deviation, those columns seem unnecessary and only serve to make the table wider than it needs to be.Jtenorj (talk) 16:47, 1 October 2018 (UTC)

I think those wikipedia editors don't know that phones exist and that not everyone has a 4k display, i wrote this on a 1080p display, also, 4k correctly only refers to DCI 4K used in cinemas, the consumer "4k" is 3840x2160, it's slightly narrower than DCI 4k. — Preceding unsigned comment added by RhebucksOnRyzen (talk • contribs) 07:24, 9 March 2022 (UTC)

missing Nvidia products

Quadro Plex, Quadro NVS, Quadro FX mobile. Alinor (talk) 15:41, 21 January 2010 (UTC)

ION and ION2, Geforce desktop chipsets, Geforce laptop chipsets.Alinor (talk) 05:50, 28 March 2010 (UTC)

Done. Alinor (talk) 14:08, 6 February 2011 (UTC)

NVS 450 is still missing. Thanks in advance -- LAZA74 (talk) 17:48, 4 January 2018 (UTC)

Geforce 6800 XE is missing (it has DDR2 128-bit, 256MB VRAM, config: 8:3:8:8, 275 MHz core, 266 Mhz memory) https://www.techpowerup.com/gpu-specs/geforce-6800-xe.c1819 -- 5 January 2020 — Preceding unsigned comment added by 85.216.150.61 (talk) 15:46, 5 January 2020 (UTC)

All the Nvidia mining GPUs are missing. Not adding them right now since my Google-fu isn't good enough to get good sources in the time I had to spend on building the table:

Model	Launch	Code name	Fab (nm)	Transistors (billion)	Die size (mm2)	Bus interface	Core config	Clock speeds			Fillrate		Memory				Supported API version				Processing power (GFLOPS)			TBP (Watts)	SLI support
								Base core clock (MHz)	Boost core clock (MHz)	Memory (MT/s)	Pixel (GP/s)	Texture (GT/s)	Size (GiB)	Bandwidth (GB/s)	Bus type	Bus width (bit)	DirectX	OpenGL	OpenCL	Vulkan	Single precision (Boost)	Double precision	Half precision
P106-090	July 31, 2017	GP106	16	4.4	200	PCIe 3.0 x16?	768:48:48	1354	Unknown	Unknown	Unknown	Unknown	3	Unknown	GDDR5	192	Unknown	Unknown	Unknown	Unknown	Unknown	Unknown	Unknown	Unknown	No
P106-100	June 19, 2017						1280:80:48	1506	Unknown	Unknown	Unknown	Unknown	6	Unknown		192	Unknown	Unknown	Unknown	Unknown	Unknown	Unknown	Unknown	Unknown	No
P104-100	December 12, 2017	GP104		7.2	314		1920:120:64	1607	Unknown	Unknown	Unknown	Unknown	4	Unknown	GDDR5X	256	Unknown	Unknown	Unknown	Unknown	Unknown	Unknown	Unknown	Unknown	No
P104-101	January 5, 2018						2560:160:64	1506	Unknown	Unknown	Unknown	Unknown	4	Unknown		256	Unknown	Unknown	Unknown	Unknown	Unknown	Unknown	Unknown	Unknown	No
P102-100	February 12, 2018	GP102		12	471	PCIe 3.0 x4?	3200:200:80	1582	Unknown	Unknown	Unknown	Unknown	5	Unknown		320	Unknown	Unknown	Unknown	Unknown	Unknown	Unknown	Unknown	Unknown	No
P102-101	Unknown						3200:200:80	1557	Unknown	Unknown	Unknown	Unknown	10	Unknown	GDDR5	320	Unknown	Unknown	Unknown	Unknown	Unknown	Unknown	Unknown	Unknown	No

85.228.100.181 (talk) 00:07, 21 February 2021 (UTC)

• missing NVIDIA CMP 170HX mining card (based GA100 GPU) https://videocardz.com/newz/nvidia-cmp-170hx-mining-card-with-ga100-gpu-has-a-massive-heatspreader 123.243.193.71 (talk) 09:45, 25 November 2021 (UTC)

Title

The following discussion is an archived discussion of the proposal. Please do not modify it. Subsequent comments should be made in a new section on the talk page. No further edits should be made to this section.

The result of the proposal was moved. --BDD (talk) 17:57, 31 January 2014 (UTC)

Comparison of Nvidia graphics processing units → List of Nvidia graphics processing units – This is one of several articles that I think are all mistitled for the same reasons. Please see the centralized discussion at WikiProject_Computing. Someone not using his real name (talk) 16:33, 13 January 2014 (UTC)

• Procedural close discussion spread over 4 talk pages. If you want a multimove discussion, please use the appropriate template and close this discussion, if you want separate discussions, please close your centralized discussion -- 70.50.148.122 (talk) 06:55, 14 January 2014 (UTC)

The above discussion is preserved as an archive of the proposal. Please do not modify it. Subsequent comments should be made in a new section on this talk page. No further edits should be made to this section.

Re-introducing "Features" tables

There used to be a table with the heading "Features" from the GeForce 6xxx to 200 series, below the main information table, the last one being here: List_of_Nvidia_graphics_processing_units#GeForce_200_series. This table's columns have varied over time - for example, the Features table for the 8xxx series lists everything from different types of SLI support (rather redundantly in two colums), PureVideo support, and CUDA support; whereas the former only mentions SLI and PureVideo. I propose that we re-introduce the features table, and make it comprehensive and consistent.

As for motivation and a reason for why I'm here, I'm coming from the Nvidia NVENC article. I noticed that the article did not have a list of which GPUs supported NVENC and which didn't. In the talk page someone suggested that instead of editing the NVENC article itself, I create a table on this article instead. Considering that previous generations have had information regarding PureVideo added to them, I feel like NVENC would be a great complement to that.

However, since I want consistency, I feel like it might be a good idea to change some of the more recent entries - the GeForce 900, 10 and Volta series GPUs all have a column for SLI support. Since there's a precedent already for having that information in a table, I think it might be good to remove it from the main table and put it back into the features.

The table would look something like this:

Model	Features
	Scalable Link Interface (SLI) (including SLI HB)	PureVideo	Nvidia NVENC	CUDA Compute ability
GPU 1	No	No	No	No
GPU 2	4-way / 2-way HB	Feature Set A/B/C...	1st/2nd/3rd... Generation	1.0/1.1...

As you can see from the links I've embedded into the table, most of this information can and will be sourced from elsewhere in Wikipedia - except of course for NVENC (for which the official support matrix will be used). I think it's important, however, to have all of the basic information of which GPUs support what technology to be included on this page. It may prompt readers, unacquainted with the technologies listed, to continue reading about them by clicking the links available on the tables.

I'd like some input from other editors before I go ahead and make some changes, so I await your replies. TiredOcean (talk) 21:31, 27 December 2017 (UTC)

Feature table summary

I saw in the ATi/AMD GPU page they have a table with the Radeon GPUs and their feature sets on a higher level point of view. Was wondering if people would think if something like that would be good to add to this page. I made a mock-up on my sandbox

ShuffyIosys (talk) 16:28, 18 March, 2018 (UTC)

GeForce 7900 GX2 vs GeForce 7950 GX2

Hey Guys, I can't see a difference between those two cards. Is that correct? Best regards, 84.144.205.205 (talk) 07:00, 11 August 2018 (UTC)

Volta

Volta section probably should be moved down entirely to workstation section. GV100 is already dubbed there. Elk Salmon (talk) 07:58, 26 August 2018 (UTC)

What about the three devices MX150 / GT1030 / GP 108

I'm confused, I see three devices but can't identify them in the tables. 0x10de 0x1d01 0x10de 0x1d10 0x10de 0x1d12 — Preceding unsigned comment added by 68.101.121.218 (talk) 03:33, 12 September 2018 (UTC)

Fab

Those numbers, that are part of techprocess brand, does not represent anything that can be measured with a given number. It's just an abstract number based on density delta. Elk Salmon (talk) 19:17, 25 September 2018 (UTC)

Huh? The use of nanometers to measure "minimum feature size" of a manufacturing process is well established. From PCMag:

Historically, the feature size referred to the length of the silicon channel between source and drain in field effect transistors (see FET). Today, the feature size is typically the smallest element in the transistor or the size of the gate.

--Vossanova ^o< 22:31, 25 September 2018 (UTC)

Since FinFET it's just a virtual number calculated from density improvement rate. 12FF is just a brand name. It's not a measurement. It does not measure anything. And if you would refer to fin width. It's less than 10 nm in 16FF. 12FFC is a rebranded 16FF++++ process with 12FFN being Nvidia optimization. Density varies through process refreshes. Elk Salmon (talk) 18:33, 3 December 2018 (UTC)

For the GeForce 20 series I cannot find a confirmed source saying that the (16FFN ++) node is used. I will add a citation needed for this until a reliable source can be found. All of the sources which I was able to locate referenced the (12FFN) node.

--Aedazan ^o< 13:12, 28 September 2018 (AEST)

Hey, while changing the table a bit I also changed the "Fab" part. I didn't see this discussion till afterwards though.. Sorry

A Fab is normally the building (complex) where the chips are produced. What we are talking about here is just called "technology" or "process", which I used. There is also no real technical reason for these Nanometer numbers anymore. It's mostly marketing. Appending "(nm)" makes therefore no sense.

While I also can't find a direct confirmation of Nvidia about Turing using "12nm", it is pretty logical. Turing is pretty similar architecturally and also has similar density to Volta (Divide transistors through die size).We know Volta is made in "12nm" from eg. Anandtech: https://www.anandtech.com/show/11367/nvidia-volta-unveiled-gv100-gpu-and-tesla-v100-accelerator-announced

The density improvement from TSMC "12nm" vs "16nm" comes from there being a new smaller 6 vs 7.5 Track cell. What Nvidia likely doing is still using 7.5T (less redesigning) with "12nm", but they still get better efficiency over "16nm" transistors.

TSMC 16nm vs 12nm: https://www.anandtech.com/show/11337/samsung-and-tsmc-roadmaps-12-nm-8-nm-and-6-nm-added/4

Discussion about the 12nm Cells thingy: https://www.semiwiki.com/forum/f293/nvidia-announces-tesla-gv100-gpu-tsmc-12nm-ffn-9285.html

--IonPike (talk) 04:51, 12 October 2018 (UTC)

10-, Volta- and 20-series table re-design/streamlining

Ok, I'm hope I'm not alone in this, but I believe my new "slimmed" tables are just prettier and more useful. But they keep getting reverted by this user. I don't want this to end in some useless edit-war and violate this 3 revert-rule, something they already did. I didn't have to deal with this in the past, so I don't know how to proceed exactly. Should I just report it to the admins? I mean, maybe I'm also just wrong and you guys liked the old tables more even? But eg. see the topic on this talk page here about "who ruined the tables" talking about pre-my-edit look and I even got thanked for my changes on my talk page already, which was a nice surprise. Honestly, this constant fighting is tiring me. And if this is what I have to deal with on here, then I don't think I will continue.

And I'm not the only one affected by this reverting. The user often reverted a table header to "Fab (nm)", as discussed above, and also took back another user's change to move the API support out of the tables. Something that is superfluous there and just wasting space as all models support the same APIs anyway.

IonPike (talk) 02:24, 15 October 2018 (UTC)

I found this users response to a peer review challenge to be quite jarring, rather than simply submit the requested citation they went on a minor tirade where they called me a 'dumbshit'. While normally I would not be perturbed by such displays of ad hominem attack I do feel that in this case the user in question has displayed a pattern of behaviour which does merit further inquiry and oversight.

Aedazan (talk) 03:03, 15 October 2018 (UTC)

• (Non-administrator comment) Just a suggestion, but a less provacative section heading (perhaps one which focuses on the tables and not a specific editor) might help facilitate discussion, and make it easier for the IP editor to participate as well. Editor behavior should be discussed at one of the adminstrator noticeboards; article content should be discussed on article talk pages. Trying to mix the two together is typically not a good idea. You should still want the IP to come to this talk page and discuss things, despite all that has happened so far; the current section heading, however,is likely to only further exacerbate things and be seen by the IP as threatening. -- Marchjuly (talk) 07:34, 17 October 2018 (UTC)

If you have any problems with the re-design, please comment here and do no just revert it. IonPike (talk) 16:40, 17 October 2018 (UTC)

I have temporarily semi-protected the page. This is not an endorsement of the current state, but an encouragement to discuss the matter here and come to a consensus with reasoned discussion. Note that "freedom of expression", which I've seen in edit summaries, is irrelevant; see WP:FREE. --Yamla (talk) 11:17, 18 October 2018 (UTC)

Proposal: split Quadro/Tesla/GRID and GeForce into seperate pages

This gigantic article slows down my Firefox considerably when loading and/or editing. How about splitting the consumer and professional lines of cards into seperate articles? --Wasmachineman NL (talk) 08:52, 4 May 2019 (UTC)

The whole point of this article is to put all of Nvidia's graphics cards into just one article, and while I wish we can optimize it so it doesn't slow down when editing that's on Wikipedia's end not ours. TurboSonic (talk) 14:12, 4 May 2019 (UTC)

Tesla card fillrate

Note that while the Tesla cards do not have video outputs, they are perfectly capable of headless operation of a virtualized graphics environment... something like Geforce Now or Google Stadia would be using these kind of cards. They would have the same fillrates and other stats as their Quadro or Geforce counterparts, assuming equal clockrates.

--73.145.176.30 (talk) 04:49, 5 May 2019 (UTC)

The "TDP" on this page is actually the TBP (Thermal Board Power)

You can see here that the "TDP" on this page is actually the TBP (https://www.hardwareluxx.de/index.php/news/hardware/grafikkarten/49921-nvidia-stichelt-mit-tdp-und-tgp-gegen-amd.html)

TheHoax (talk)

Convert 10 series+ tables to TFLOPs

Looking at the absurdly high FLOPs on newer nVidia GPU's, suggestion to change the tables from GeForce 10 series onwards to TFLOPS for processing power. It is very confusing to read the current tables as is with a jumble of numbers in the processing power section when comparing tables; example 2060 vs 1660. — Preceding unsigned comment added by 112.134.32.92 (talk) 15:37, 9 July 2019 (UTC)

GTX 900A series missing

I own a GTX 960A which has substantially differences to the "normal" GTX 960. Even it is a 9xx the chips are from the last (GM) generation (except 920A A). nice list is found here: [] -- 92.117.176.15 (talk) 09:50, 23 July 2019 (UTC)

I created a GPU performance data google sheet, and did some processing

I was interested to see how price-performance for Nvidia GPUs has fallen over time, so I did some calculations and generated metrics such as cost-per-tflop, and cost-per-billion-transistors - perhaps some of you here would be interested: see this spreadsheet.

It doesn't look like the spreadsheet shows any data that would be useful. TurboSonic (talk) 21:18, 10 August 2019 (UTC)

The data is interesting, but I think on this page it is not relevant, on another page like this yes: https://en.wikipedia.org/wiki/FLOPS Bensuperpc (talk) 07:16, 17 January 2021 (UTC)

I did some cleanup formatting on your sheet, and added launch prices of earlier models (G80/G92) and a plot at the bottom. I like it. Hansschulze (talk) — Preceding undated comment added 23:27, 3 August 2022 (UTC)

Enabling Positive Consumer Advocacy and overcoming e.FUD

@IonPike: Just wanted to mention that I too appreciate that this and several related pages are layed out in a logical and fairly functional manner.

I had to keep working with this data for a few weeks while researching how best to upgrade some classic MacPro3,1 (2008) machines. I've had to gather data for many complicated scenarios before, but the intersection of restrictions created by old CPU instruction sets, newer macOS system requirements, and the ongoing rapid development of CPGPU tech is a very challenging scenario to work out.

Thanks in no small part to all the work that you and others such as @TurboSonic: @Jtenorj: have done to refine this presentation. The logical layouts on the whole group of related wikipedia Video Card pages is far superior to the effectively sparce or even broken matrix spec information that many OEMs make available. That is when they are willing to provide even that much, as too often it seems like some part of many corporations would prefer if none of their customers actually understood how their products worked and just took their word for it.

Rather than some long discussion of the Dilbert like conflicts between STEM and Marketing/Sales, I just want to say: I've seen the raw data, and really appreciate all the work that's been done here. Tree4rest (talk) 21:07, 24 August 2019 (UTC)

References

MX 230/250

I think it is necessary to mention that MX 230/250 is not based on Turing architecture. — Preceding unsigned comment added by 92.42.28.230 (talk) 13:00, 15 December 2019 (UTC)

Power consumption issues

Hello, watch out for TDP in watts, it's not power consumption. I was picking the card also by this atribute, and assumed, when card have less TDP, it will draw less power, but it's not true. Just between cards NVidia GTX 260 and GTX Nvidia 260 core216, I was happy that 260 core216 has even less TDP from 200 at around 180/170, and that SP216 it has even less power consumption, I was at the edge of what Power supply can handle with my 150TDP card before, so I was affraid to get GTX 260 with 200W TDP, but was happy to see core216 has 180/170, so I was about to get that card. BUT...

https://www.anandtech.com/show/2619/9

as we can see from grahps, GTX 260 core216 has bigger power consumption, by 2 watts in idle, and 16 Watts in power output. Be carefull when choosing card by this assumption, that some card has less TDP than your previous card and that you can get it. I warn you, that even with 20 less TDP, it can still draw 20% more power. (the graphs shows total system power output, if you would count only graphic card, +16W, means basically, from 100Watts to 116Watts, it can be almost 20% more).

This was lession for me, I am stood shocked, so I want to warn you, because I know, many people are choosing cards also according TDP, but you cannot count power consumption from it. Usually, it means less power, but not always. — Preceding unsigned comment added by Lofr2 (talk • contribs) 07:02, 7 January 2020 (UTC)

810M and 820M have a kepler architecture variant and NVENC

Wiki only lists GF117, but encode matrix and Techpowerup 810M, 820M lists a GK107 version. — Preceding unsigned comment added by 187.87.64.108 (talk) 01:00, 19 March 2020 (UTC)

Font Size

Why is font size at 80%? Wikipedia style guidelines advise to avoid reducing font size inside of tables in general (only for use when required to fit on-screen, which does not appear to be the case on many display sizes in this case, especially for the early tables), and to never use a font size below 85% WP:FONTSIZE 09:10, 16 April 2020 (UTC)

Tesla note

There is an existing note:

Tesla

Further information: Nvidia Tesla

Note: Due to the Teslas' inability to output graphics, figures such as fillrate and graphics API compatibility are not applicable.

It appears to lack useful context (i.e. clarifying what "inability to output graphics" really means [e.g, within Nvidia's product lineup, the physical PCB's lack of output connectors, how the user is expected to interact with the GPU's functionality, etc]) . Tesla cards can render and "output" graphics, as seen by their usage in Nvidia GeForce NOW servers (reported usage from in-game tools, not officially confirmed by Nvidia) and their utilization for Nvidia vGPU and GRID remote desktop acceleration and remote application rendering suites. The graphics are not output by a DP, DVI, VGA, etc connector, but are output over a network connection (elsewhere in the host computer/server). Similarly, fillrate and graphics API are applicable for that use case.

To rectify this, I believe simply removing the note will be sufficient, since it's the primary offender. Whether or not fillrate and graphics API compatibility need to be added, is a different discussion, IMO. --12.189.223.245 (talk) 20:59, 25 June 2020 (UTC)

Missing "Mobile Workstation GPUs"

Can somebody please add the missing products from this link https://www.nvidia.com/content/dam/en-zz/Solutions/design-visualization/documents/nvidia-rtx-line-card-for-mobile-workstation.pdf

GTX 660 Ti 3G(i)B

The GTX 660 Ti also came in versions with 3GiB of memory;

https://tweakers.net/videokaarten/vergelijken/#filter:NYzLCsIwFER_pcw6QmqlafMBggtXWYqL0lzrxb5IglZL_t1UcDXMDOes4PFJLpj7NBvqqQ08jdC3pvck8KD3a3IWGl1YsrKUWWAIzE1Hhj8EnUuZquOWzpywXS7gfxayZqb2ZD30RRUHkVdVfU1nGo_cB3LpWKGKutpy2GAUUu2TfWiWf4kxfgE

https://www.gigabyte.com/eu/Graphics-Card/GV-N66TOC-3GD#ov

NitroX infinity (talk) 18:54, 31 March 2022 (UTC)

Details of field descriptions

Where do flops numbers come from? Are you getting them straight from Nvidia spec sheets? Or is it just cores*clock rate? Or something more complicated? It'd be nice to clarify this in the field descriptions section, which currently doesn't even mention processing power. Similar questions for fill rate. 68.55.179.229 (talk) 11:10, 27 April 2022 (UTC)

Clean up Data Center section

First would be to get rid of all of the old multi-GPU systems, since this article isn’t about systems. And get rid of the redundant “GPU accelerator” in the names in the first column.

Second would be to add FP16/TF32 stats, PCIe or NVLINK speeds, etc.

Third would be process nodes, transistor counts, die size.

Fourth, remove the MUL+MAD column. This info can be put in parentheses for old GPUs, or deleted since very few scientific routines could take advantage of it, or most likely simply listed as the true theoretical FLOPs. After all, it wasn’t the first time that theoretical FLOPs were hard or impossible to hit. If someone cares they could research the history. A readable table is more important.

An additional nice-to-have would be to add columns to the CONSUMER cards for FP16 multiply+FP32 accumulate which is important for some AI training but is gimped (it’s half of the advertised FP16 number), to clarify their true capabilities. 46.211.1.63 (talk) 07:33, 12 October 2022 (UTC)

Rework the page: Simplify

This page is 421,393 bytes as I post this. This is over four times the size at which it is suggested an article be split (WP:SIZESPLIT). It is also unreadable due to the length, with information overload it is hard to parse through the information provided. The tables also just don't load correctly on smaller screens such as mobile phones and anyone who uses vector 2022 (Which is about to become the wikipedia default). It also goes against WP:NOTDATABASE. This page is simply a data dump of all GPU's from nvidia. I suggest that where the article exists, the table for a specific gpu is moved to the architecture page and a simplified list of "model" and "launch date" is provided instead. I am going to WP:BOLDLY do this work after completing it in my sandbox unless there is any discussion beforehand. Thanks, Terasail^[✉️] 04:13, 14 January 2023 (UTC)

I have made the change to List of Nvidia graphics processing units#Desktop GPUs. All information removed can be found on the respective articles of the GPU cards. Afterall this is a "List of Nvidia graphics processing units" not a list of "List of Nvidia graphics processing units specifications"... Terasail^[✉️] 19:39, 25 January 2023 (UTC)

As long as all the detail spec tables can be found on the individual generation pages like you mentioned, then yes, it's redundant to have all that information in this list. I'm good with the desktop GPU changes you made, where you highlight new features and changes for each generation in a few bullet points or sentences. --Vossanova ^o< 20:31, 25 January 2023 (UTC)

You can still see all the tables, for example at GeForce 30 series or GeForce 20 series shows the relevant specification tables. Terasail^[✉️] 21:09, 25 January 2023 (UTC)

I need a reality check here.. Have I gone nuts? This page is ridiculous... It either needs multiple splits or to be heavily simplified. The data is just somewhat unreadable. Take for example the table at List of Nvidia graphics processing units#GeForce 30 series Not only does it just not fit on the screen, Its somewhat hard to follow the data in the table and most of it is incredibly niche data that almost nobody would need to know... (Core config, ect..) It makes the page unreadable, the load time is too long for the page, and all the infomation is available on the relevant cards article anyway... I keep getting reverted by User:Lofr2 and I assume their IP, who keeps telling me essentially that I am wrong and that its my issue and they have yet to open any line of discussion outside of edit summaries. I was never claiming that the changes I made were perfect and if more information needed to be added it could be, but the only thing that this list is right now is an oversized data dump. Terasail^[✉️] 21:59, 25 January 2023 (UTC)

I do not write here often, but this got my attention because of how important it is to the communities I walk in.

There is a very active hardware community that cares about this for how specific it is hardware.

Precisely because it is not merely a casual listing for the layman.

Specifically these pages have been praised for the technical thoroughness and correctness of these listings. It is also valued by buyers of modern hardware that use this as a body of knowledge to distinguish models and make good buying decisions and educate others about some of the deceptive practices of Nvidia when it comes to Mobile GPU clones/refreshes and downsizing or different chips masquerading under the same model. It also served as the corroborating foundation for the far harder to search Techpowerup and videocardz GPU databases. As those do not provide an easy to search single listing/table of all models and often lack some technical specifics that are only found here. It also helps me as a technical writer to keep track of models when doing reviews or testing new hardware releases. The removal of this page which is also amongst the easiest to find on the net will also make it more difficult for new hobbyists and DIY'ers to get into this space. So please do not remove or drastically oversimplify this page.

The hardware and especially retro hardware communities I am part of have already been soured by prior Wikipedia editors actions for ruthlessly downsizing similar such detailed works of technical literature that took their authors much work to collect. And further cutbacks will likely burn most of the trust for Wikipedia that is left in these communities. Rigred-5 (talk) 13:40, 30 January 2023 (UTC)

I have a feeling that this change is to adapt the content to the new (worse) wikipedia skin. These CNET style comparison tables worked fine on the older wikipedia theme. Rigred-5 (talk) 13:52, 30 January 2023 (UTC)

However if this page were to be downsized, care should be taken to at the very least still adhere to the outlined useful criteria I listed. Especially to usefully distinguish the characteristics of certain models that exist with multiple mixed chips (of often vastly different performance). Omission of such would be tantamount to aligning with the misleading marketing information from Nvidia. Rigred-5 (talk) 13:57, 30 January 2023 (UTC)

@Rigred-5: Arguments about this need to be made based on Wikipedia policy, guidelines, the manual of style, etc., and not based on whether people like this page or it is useful. Wikipedia is not a central place for creating useful Wikis of any description at all, it is an encyclopedia. Would you care to review the discussions of relevant wikipedia practices on this talk page and address them? Because this what you've said about people like the page or whatever is, not to be rude, completely irrelevant. The Wiki software is available for making hypertext catalogs and you can host this exact page, and ones of much great technical detail, on another website. —DIYeditor (talk) 22:43, 30 January 2023 (UTC)

We have initiated steps to provide the contents elsewhere.

You are as such free to render this page as stylish and devoid of knowledge as you wish.

That will be all. Rigred-5 (talk) 23:36, 30 January 2023 (UTC)

Maybe this page could be made into just a simple bulleted list, get rid of all the diverse tables, or a simple table with uniform columns, or multiple tables and sections but without all the noise and disjointed ways of listing and discussing the GPUs. —DIYeditor (talk) 22:38, 25 January 2023 (UTC)

For example from Wikipedia:Manual of Style/Lists:

• Use a bulleted (unordered) list by default, especially for long lists. Use a numbered (ordered) list only if there is a need to refer to items by number, the sequence of items is important, or the numbering exists in the real world (e.g., tracks on an album).

And from MOS:LISTFORMAT:

• List items should be formatted consistently in a list. Unless there is a good reason to use different list types in the same page, consistency throughout an article is also desirable.

So I tend to agree this page needs to be cleaned up and mostly trimmed away. —DIYeditor (talk) 22:47, 25 January 2023 (UTC)

I'm using it almost daily. I definitely use also core name, introduction date, and core config. Also memory and core speed. I use also lots of old , pre 2010 entries, as I'm working with retro hardware. In this case, compatibility is more, than speed. As you find all in one page, which is great in comparing graphic cards. Not everything has to be simplified for mobile users, there's lots of people working with retro hardware, cause modern hardware is boring and non-exciting for them.

The one entry I would sacrifice from old tables, is die size for example. And Transistors(in millions) is too wide for value, that is often 2 or 3 digit. That name of the column should be simplified and reducing width of table this way. But I never used number of transisitors anyway, so I would sacrifice that too, that information would be enough to be included in specific page belonging to the generation.

The "MOperations/s MPixels/s MTexels/s MVertices/s" should be reduced in width too, as it holds only simple number value, and it's too wide for the information value.

The page is big, but in this case, big is good I would say. The many people is using it this way for ages. I would wait for it even 3 seconds to load. It is worth. If you want simplified version, why you don't make new page, instead of screwing it for many people, that like it this way, and are used to it for years. It's also disrespectful to remove by such barbaric way so many hunreds hours of work of other users, that made corrections to the tables over the years. Lofr2 (talk) 08:56, 26 January 2023 (UTC)

These diverse tables should be on individual pages for each separate section. This is a plain mess and too large. I think Terasail is correct. This should be trimmed to a simple list or series of lists all with the same few parameters, not a bunch of random looking tables.

If each section doesn't list the same information why does it all need to be on the same page? This is aesthetically bereft and confusing.

Can you cite any policy-based reasons this should be kept all on one page rather than your personal wants and feelings? Wikipedia is an Encyclopedia not a reference book. There are lots of sites on the internet that can host technical catalogs for retro hardware. Why is it easier for you to scroll through hundreds of lines on one page rather than click a relevant page for the GPU of a particular generation? I think it is less easy. —DIYeditor (talk) 09:56, 26 January 2023 (UTC)

WP:NOTPAPER. "Aesthetics" or vague references to data load are not sufficient reasons to impoverish the Wikipedia. WP:NOTREFERENCE is not a real thing, I don't think. "Your personal wants and feelings" is a curt and I think inappropriate judgment against stakeholders who actually use the article; I think it is not productive to dismiss their needs like this. As far as those use cases, cross-referencing details from across rows and eras is a reasonable and common use case of Wikipedians and visitors from across the Internet at large, who will be harmed by overzealous attempts to prune this list. Splitting up the list would make cross-referencing detail across generations more cumbersome. Should we investigate ways to make the list more concise? Of course! But I strongly resist mischaracterizing the specialized user's usage of Wikipedia as being insufficiently up to standards - common sense and charity are, after all, also Wikipedia policies.

Speaking of WP:NOT, WP:LISTCRITERIA specifically states that even the list "list of shades of colors of apple sauce" could be considered for inclusion if the argument that it is worthy of inclusion was accepted. Lo and behold, quite a few users active on this page have signaled just such a view, and brandishing vague references to policies doesn't seem productive. And I think it would be good if roving Wikipedians wielding policies would defer occasionally to subject matter enthusiasts on what, per LISTCRITERIA, "contributes to the state of human knowledge." Edwin Herdman (talk) 13:10, 10 April 2023 (UTC)

Edwin Herdman,

• WP:NOT
  • Keeping articles to a reasonable size is important for Wikipedia's accessibility, especially for readers with low-bandwidth connections and on mobile platforms, since it directly affects page download time (see Wikipedia:Article size). Splitting long articles and leaving adequate summaries is a natural part of growth for a topic (see Wikipedia:Summary style).
  • Listings to be avoided include, but are not limited to: business alliances, clients, competitors, employees (except CEOs, supervisory directors and similar top functionaries), equipment, estates, offices, store locations, contact information, patent filings, products and services, sponsors, subdivisions and tourist attractions.
• Wikipedia:Summary style (all of this is relevant)
• MOS:LIST
  • List items should be formatted consistently in a list. Unless there is a good reason to use different list types in the same page, consistency throughout an article is also desirable.
  • Keep lists and tables as short as feasible for their purpose and scope: material within a list should relate to the article topic without going into unnecessary detail; and statistical data kept to a minimum per policy. Some material may not be appropriate for reducing or summarizing using the summary style method. An embedded list may need to be split off entirely into a list article, leaving a See template

—DIYeditor (talk) 14:31, 13 April 2023 (UTC)

This is a technical article list, and all our discussions have to keep that in mind. WP:NOT and accessibility concerns are guidelines - and to my way of thinking, guidelines should not be used to talk over people who have stated, clearly and repeatedly, what their needs are from articles like this and the value it provides. I'm not calling concerns about bandwidth or accessibility totally hypothetical, but there are a number of possible compromises (and simple features like collapsible lists can be looked at) that should be researched before making a radical destructive change - like happened over on the less-visited EXMOR article which has essentially become a stub after overly ambitious paring down of the article.

You will notice, I hope, that the text you quote me includes the phrases "good reason," "also desirable," and "as short as feasible" and "unnecessary detail." Opinion comes into this as a matter of course, but it seems to me that testimonials from people who use these types of articles routinely cite thoroughness (which is not the same as unnecessary detail) as a virtue.

Consensus is important and I do not think that the movement against specialized technical lists have provided evidence of a consensus that these articles are too long except amongst page-length-concerned editors. --Edwin Herdman (talk) 05:13, 17 April 2023 (UTC)

Looking at again, I think these complex tables should be left for the GPU series articles (e.g. GeForce 40 series) and that this list, and probably all the microarchitecture articles, should be in the format found in Turing (microarchitecture) which is just a simple bulleted list. These tables are unwieldy and don't serve the function of a "list" which is what this article is supposed to be. —DIYeditor (talk) 18:24, 27 January 2023 (UTC)

If it's a discussion of simplification of this page, then this would have to trickle out to other pages like it. On one hand, I get the sentiment that it's too large, but on the other hand, grossly simplifying this when say I like to do a comparison of multiple generations of video cards wouldn't work. And I'm sure I'm not the only person who uses this page for that reason. Some parts I agree on, like maybe separating all the workstation GPUs from the consumer GPUs.

But if trimming data is a concern to lessen how much raw data is on the page, yes, I think this needs to happen. I'd say lets take some baby steps to trim what's here before people can decide on where to take the content further. Some issues that I have with the page are (in no particular order):

• For the GeForce 30 and 40 series, we need to get rid of the columns in the ray-tracing performance section that have nothing in it. NVIDIA hasn't given anything, nobody's came up with a value for them, so let's just get rid of it. (I'll probably do that after this post)
• If a column is completely merged, then why have it? Just put it up on the header of common features
• I don't know if it's a limitation of the wiki system, but could we have footnotes be that: a foot note? Like on the bottom of the page? It takes up unnecessary amounts of room when there's 30+ copies of the core config when we only need like 2-3 unique versions of it. Now that hovering over the footnote brings it up in a tool tip, this won't cause a disruption by having to go down to the footnote to see what it is
• Getting rid of the name "GeForce" in every model name with some exceptions (like the standalone "GeForce 3" or "GeForce 2"). I think it's redundant since there's a big heading that says "GeForce N series." It creates clutter that doesn't really add to the value of the name.
• Some foot notes shouldn't really be in a page that lists specifications, because they don't really add value to it. For example this one from the GeForce GTX 960 "Some manufacturers produced 4 GB versions of GTX 960. These were often criticized as useless move, as titles that would use so much VRAM and actually gain advantage over 2 GB versions, would already run too slow on those resolutions and settings, as GTX960 didn't have enough compute power and memory bandwidth to handle it." Just put a citation on the name or the 4GB part and if someone's really curious about it, they can look at it.
• Some footnotes sound more like trivia, e.g., "The GeForce GT 730 (DDR3, 64-bit) is a rebranded GeForce GT 630 (Rev. 2)" This I think should be better suited for the series page itself than this one.
• For the common feature header, could we make that a table instead? On that note, what items really need to be kept?

ShuffyIosys (talk) 03:11, 9 February 2023 (UTC)

Yes, I think this is a good approach and should be the model going forward. Edwin Herdman (talk) 12:42, 10 April 2023 (UTC)

I strongly oppose the 'Getting rid of the name "GeForce" in every model name' part, as then it makes it harder to differentiate between GeForce consumer GPUs and Quadro workstation GPUs. One might sit there looking at RTX 4050 next to an RTX 4000 and get stuck thinking what segment of GPU is this?.

It also creates potential confusion as one might look at "RTX 5000" without the "Quadro" part and think, "oh wait a minute, look at that new RTX 5000 gaming GPU! It's out!!!".

By having the GeForce and Quadro names, it makes the distinction between the two very clear.

Another problem of removing GeForce is non-GTX and non-RTX branded GPUs. Oh, why is there a stray "210". What's a "940MX", is that the name of some chipset or, some decoder chip, ...

—  AP 499D25  (talk) 13:39, 13 April 2023 (UTC)

The list is already divided into consumer and workstation lines so the idea at least makes sense. However the brand names are part of the proper names of the units and shouldn't be removed as "710M" isn't the proper name of a Nvidia product - "Geforce 710M" is. That would indeed make the list less readable, accurate, and be bad for web search as well. --Edwin Herdman (talk) 05:25, 17 April 2023 (UTC)

I'll retract my suggestion for removing "GeForce". In the mean time, I made some tables in User:ShuffyIosys/sandbox on what I was thinking. Note it's a very rough draft, I just wanted to get some thoughts down while I had them. ShuffyIosys (talk) 08:23, 30 May 2023 (UTC)

Moving all the tables to templates

I am considering moving pretty much all the tables in this list article, into templates, like how it's done on List of AMD graphics processing units and Intel Arc.

The primary reason for doing this is to de-duplicate the tables between multiple articles that may have the same table, e.g. between this article and GeForce 40 series.

So if a new SKU is to be added to the GeForce 40 table for example, instead of having to make the same change to the 2 or 3 various articles that have the same table, you only need to make the change to the template once, and it will go "live" on all the articles that have that template, simultaneously.

The use of templates also help with keeping a consistent, uniform look between the various articles and lists, as there will be no longer cases where for example GeForce 30 table has 14 columns on one article but 15 on another.

Any thoughts or objections to this? — AP 499D25 (talk) 01:49, 1 June 2023 (UTC)

The only issue I see with this is the ongoing problem between what specs this page should contain and the individual series pages. If we use a simplified template to make this page less technical or whatever, that makes the table in the individual pages less useful. But if we use a detailed template for the individual pages, we're back to the problem we want to solve here.

The only compromise I can think of is have the simplified table with an additional one in the individual pages for anything else. ShuffyIosys (talk) 15:56, 1 June 2023 (UTC)

Missing product

The HPC card "NVIDIA A800" is missing in the article. Specs and further informations can currently (08.02.2024) be seen via https://www.nvidia.com/en-us/design-visualization/a800/. Clash crafter (talk) 00:45, 8 February 2024 (UTC)

There is no point adding anything. List of product articles are now banned on Wikipedia and soon all will be deleted. Listing is only allowed as Category now, so each product should have own article. Elk Salmon (talk) 23:13, 29 February 2024 (UTC)

Remove/Clarify FLOPs for fixed function GPUs

I see a number of modifications were made by 67.84.203.109 which added FLOPS32 to the fixed function GPUs.

What does this even mean? It appears that the clock speeds were just multiplied by 2. The fixed function GPUs were not programmable. They unlikely used standard floating point, so FLOPs doesnt make sense. Also you would need more than 2 FLOPs to emulate the operations performed by a fixed function GPU per clock.

Suggest remove these columns? Squater (talk) 14:38, 18 April 2024 (UTC)

@Squater: Also, between Rankine and Curie there's a huge drop in FLOPS. Remove - seems to be WP:OR. --Zac67 (talk) 12:03, 16 November 2024 (UTC)

@Zac67 I Removed up to and including the GFLOPs on Rankine. Curie potentially should also have GFLOPs removed since, AFAIK, Nvidia only started publishing GFLOPs from Tesla+ Squater (talk) 09:43, 17 December 2025 (UTC)

Amount of detail in Nvidia, AMD and Intel GPU lists

Some editors want to remove details. How and what should we remove?

Excessive technical details over a very large range of products; reads like a product catalog or line sheet rather than an encyclopedia article. Poorly referenced -- many sections have no references at all
— User:JWiki2151 Nvidia 05:33, 7 June 2024 AMD 05:31, 7 June 2024

{{Overly detailed|date=May 2023}}
— User:Freedom4U Intel 15:17, 4 May 2023

I oppose removing info. It's a list not article. I'm [also] using it ... core name, introduction date, and core config. Also memory and core speed. #Rework the page: Simplify was inconclusive. #Moving all the tables to templates may also complicate removal. 142.113.140.146 (talk) 07:39, 1 September 2024 (UTC)

50 series core count

Hi all, the core count on 50 series gpus seems off - tensor cores and ray-tracing cores (SMs) are swapped. Tensor cores come in 100s, whereas ray-tracing cores come in 10s, thus different orders of magnitude. https://www.nvidia.com/en-us/geforce/news/rtx-50-series-graphics-cards-gpu-laptop-announcements/ states explicitly "680 5th Generation Tensor, and 170 4th Generation Ray Tracing Cores" for the 5090. 2A02:8428:4FE9:9201:2040:6762:7995:3419 (talk) 10:25, 15 January 2025 (UTC)

You seem to be right. I have just fixed it. Thank you. -Geek3 (talk) 13:30, 15 January 2025 (UTC)

Thank you for fixing it 85.170.142.51 (talk) 14:54, 27 January 2025 (UTC)

Combine Workstation and mobile

The two sections Workstation GPUs and Mobile Workstation GPUs are highly parallel, and the both have tables for each new generation of GPUs. The workstation and mobile versions often use the same chips and get released at similar times. In the current arrangement it is rather confusing that these similar versions are so far apart in the article, and one is easily confused to scroll to the wrong section, if one was actually looking for the other corresponding table. Therefore I propose to restructure the two sections such, that the two tables of workstation and mobile versions will each be placed right after one another. This also makes it easier to compare workstation and mobile versions of one generation to each other. Any thoughts? --Geek3 (talk) 14:37, 8 May 2025 (UTC)

Since nobody had any objections, I just combined the two sections. --Geek3 (talk) 19:33, 11 June 2025 (UTC)

I would even suggest to separate Mobiles GPUs to separate entry.

So keep it only for desktop PC graphics as it getting too long. And loads quite a long time.

And rename it to List of Nvidia desktop graphic processing units. Lofr2 (talk) 18:18, 12 March 2026 (UTC)

NVIDIA L40 Double Precision Performance

The Datasheet by NVIDIA does not mention double precision performance for the L40 GPU. Some sources (https://vast.ai/article/comparing-nvidia-l40-vs-l40s-and-more) claim that the L40 does not have FP64 support. Where does the number in the table come from? Curious Caiman (talk) 13:43, 13 May 2025 (UTC)

I confirmed the data now on cloud servers (got roughly the same results), but I would still be curious where the data in the table came from. Curious Caiman (talk) 10:40, 16 May 2025 (UTC)

Get rid of the mourning borders

By default this page shall open using the standard width appearance w/o mourning borders, e.g. using the whole width available.

If you want to read the content of this page on a smart phone or other device with limited viewing capacities, please create your own representation having just a fraction of the information of this site and matching your presentation taste. 2001:16B8:E1F5:F600:B75E:FBAA:160A:E71E (talk) 11:26, 1 November 2025 (UTC)

Anyone, who doesn't like Defapianoplayer's "overhaul" ?

He made it more chaotic. Tables are less readable and more chaotic. Lots of unnecessary things, like power connectors (which are often even not true, because they are card manufacturer dependant not "core" dependant (i.e like 2/3 of GTX 750 ti has 6-pin Power Connector anyway), this column takes also lots of space and it's not important as anyone needs to go check card itself anyway, how manufacturer made it.

It's also not nice, he decided on his own, such changes of this scale like overhaul of whole list should be first discussed here. He made it basically for himself.

For example me, I have now problems to orient in the table, I've used it for decade almost daily. If it was me, I would probably discard it whole, for me, but I don't want to decide for others... well, it's what he did basically. For me, it didn't add anything improtant, just made it difficult to orient in and looks worse now. Not sure how for others. Lofr2 (talk) 13:55, 28 March 2026 (UTC)

In total, Defapianoplayer's edits seem to improve the article. To me it seems more congruent in general, and more compact and readable. For instance, he took out some columns, where every row had the same value, and inserted columns with missing cache memory values. Many numbers have a more reasonable precision of around 4 digits now. Sure, the power connector columns where most entries are "None" can be disputed. But the interest in discussions here seems to be rather low, so I can understand that he went ahead and implemented some changes.
So I am against discarding the changes. However, if you feel that certain changes are for the worse, try to list them here, and then later revert these specifically. --Geek3 (talk) 18:52, 30 March 2026 (UTC)

The fact that he removed almost all notes and references in the tables is something that shouldn't have happend.

Also because of that most of the data regarding pixel rate is wrong, he removed the GPCs data thats important, one exaple of this would be the GTX 1060 despite having 48 ROPs it can only rasterize 32 pixel as it only have 2 GPCs, and the same happens to the 1070 despite having 64 ROPs, it has disabled one of it's GPCs, having only the ability to rasterize 48 pixel/clock, this also extends backwards to Fermi, nvidia usually had more ROPs that they could properly setup their pixel/clock due to not having enough Rasterizers. And now that it's not disclosed like it use to be.

Other thing that is wrong is the cache sizes particulary L1/texture, Kepler has configurable sizes for example. In regards to Kepler he added that it functions as 2 Flops/clock instead of 1 of Fermi, when in reality both has support for FMA so obviously both of them have 2 Flops/cycle for single precision, i assume that he wanted to point that Kepler had dual issue.

Also removing references pointed at Fermi controversial TDP.

The changes are either myopic or deliberate to obfuscate the history of the page. ~2026-22328-02 (talk) 04:18, 11 April 2026 (UTC)

I agree. Also, he added so many unnecessary columns, like TurboCache in 6000-7000 generation? Cmn... such unnecessary wide column, this made tables too wide. At the same time he removed comments, which was often more useful column. Lofr2 (talk) 18:44, 15 April 2026 (UTC)

Some of it is perhaps better, but a good chunk is worse. The power connector stuff shouldn't be there, since as you mentioned, it's card related and also OEM dependent. Eddmanx (talk) 16:20, 11 April 2026 (UTC)

Seems, he's just copying data from Tech Power Up database, which is full of errors.

He reintroduced many mistakes or bad interpretations, which are stored there. Some of them I've reported over years, but they don't change anything, it's not maintained, particulary on old GPUs.

Like GTS 450 rev.2, I've got one with GF116 core, and it still has DDR3, not GDDR5.

to the GTS 450 rev.2 rev.3 names... Whole internet search doesn't reveal anything except Tech Power Up record. This is where I've realized, he is just copying that database here. Yeah, GTS 450 were updated with GF116 core later, but he just copyied whatever he found on Tech Power Up database, it's not anything officially released, or named.

Another problem I've realized... because of these various records, introducing new tables/rows, he made the page not only more chaotic, less simple, but it loads much longer. Which was the point of this page, making it simple enough, so it loads fast enough. It wasn't necessary to store every freaking OEM variant of graphic card, separate everything into special lines and records. Before, it was united (like example on GTS 450, where were just named both cores GF106,GF116 and GDDR5,DDR3 under ,in one box. It was enough imho.

So, this overhaul making the table too complicated, big, overloaded, which is problem also to future as new graphic cores will be introducing. So history of graphic cards should be kept more simple and compact.

imho: Power connectors columns should be completly removed. OEM variants and "rev.2 , rev.3, rev.4" nonsense should be more limited and put into one record if possible. The point of table was not to include every odd variant of card, as this page was not about graphic cards, but more about "cores" introduced by Nvidia (so eighter reference designs or official releases). Of course every manufacturer could do special version, put older core here and there, use other memories, other memory width, undervolt card and use it without power connector with weaker VRM, etc...

These are just thoughs for the future of the table. Right now, I think it's overcomplicated, and Defapianoplayer's change was not positive one. Didn't added any real value. You are finding there what was already there before, just longer. Made table more chaotic, less simple. Usually, people got here to check only some one main stat of graphic card or compare two cards, so they wanted it as quick as possible, with longer loading times, it get worse in this aspect. Lofr2 (talk) 14:29, 14 May 2026 (UTC)

H200 memory erroneous info or is there a refresh of the H200 ?

The product brief linked in the website of the listed source 514 claims a memory bandwidth of 4,813 GBps instead of the 3,360 GB/s given in the table. Is this a error in the table, or due to a new version of the card? How should this be handled in the table? — Preceding unsigned comment added by ~2026-22864-51 (talk) 13:26, 14 April 2026 (UTC)

Missing GPC Data

Since Defapianoplayer's begun the changes of the article, now GPC data is missing in every entry since Fermi and onwards and it should be re-added, it is important to have the number of rasterizers each cards has access, since that impacts how they can properly setup their Pixel/clocks, despite having more addressable ROPs ~2026-29609-21 (talk) 04:04, 17 May 2026 (UTC)

Standardize Tensor performance on the latest GPU's

Could we standardize the denomination and add the Tensor FP16/FP8/FP4 perofremance to all the tables (Workstations, Mobile and Server) for the latest GPU's (Ampere and on, unless it makes sense to start even earlier)? It's only on some, and sometimes the precision is not indicated. ArnaudContet (talk) 09:03, 18 May 2026 (UTC)

Another big error on Defapianoplayer's "overhaul"

He is revamping the tables again, now on Fixed pipeline cards, and is butchering their vertex performance, in a clear misguided way he is multiplying the Vertex shaders by the clock of the cards, thats not how MVertices are calculated... and the clueless puts a unsourced reference claiming that's the theoretical maximum but Nvidia only disclose a quarter (that's wrong)

NVIDIA officially advertises 1/4 the maximum GV/s for Vertex Shader's performance compared to the theoretical maximum as it is more realistic for real world applications.

When will a moderator intervine, this is not the first neither the last error Defapianoplayer will make, at this point is downright vandalism. ~2026-30838-61 (talk) 05:24, 23 May 2026 (UTC)

He's definition of "well-meaning but incompetent helper" phrase, in our language we even have more offensive term "užitočný idiot". Not sure if it can be translated as useful idiot. He just copying Tech Power Up database, which is full of errors, reporting is broken, i've tried to report many mistakes, but it is just ignored, and links by reporting don't work, it's outdated.

I've repaired like 30 errors and missing stuff over last 4 years in this table, he basically reintroduced all errors. TPU database is no reference!!! It's just other 3rd party database, built by submitters. This is why wikipedia should not be inferior and just copy all stuff from TPU and refernce to it. Or even have reference in every freaking line to it.

Not to mention, that table loads like double time as before, after his overhaul. Editing is even more painful, like 10-15 second waiting times. The key was to kept table as simple as possible, as it will getting only larger with upcomming GPUs. Should not to introduce every damn variant of card (not even official GPU specificaion, just card type made by manufacturer).

Honestly, this topic should be really moderated. Problem is, who will repair those millions of little errors, moderators usually don't catch them or don't have time for them. Seems wikipedia hit the rock with modern people, who just change it completly according themselves, without any dialogue or discussion, or planning. Like it's their personal table. So selfish... my 20 hours of work on this table in last 4 years, fixing various errors, are done. Can do it again. Lofr2 (talk) 08:08, 28 May 2026 (UTC)

Table for Mobile GeForce RTX 50 series

This is the only "vertical" table , while all others are horizontal. Consistency make the whole article easier to read. ArnaudContet (talk) 12:27, 26 May 2026 (UTC)

another mistake Defapianoplayer screwed Geforce FX data, needs to be redone and repaired lots of errors and nonsense from TPU database

it was good, PCX 5950 was there. And PCX 5900 had correct release date. Check old version: https://en.wikipedia.org/w/index.php?title=List_of_Nvidia_graphics_processing_units&diff=1340622812&oldid=1340620962#GeForce_FX_(5xxx)_series

this date and table also fit specific Geforce FX page https://en.wikipedia.org/wiki/GeForce_FX_series#GeForce_FX_model_information

He basically removed PCX 5950 from the list, because of some TechPowerup nonsense (probably didn't find it there), and again screwed PCX 5900 release date, how can be PCX graphic card released in 2003? When even in specific FX page is written, that Nvidia remade FX series into PCX versions in early 2004. Which is logical when PCI-E slot was introduced during 2004.

Now FX table is full of errors from errors from Tech Power up database. And possible other tables too. For example this is wrong: https://www.techpowerup.com/gpu-specs/geforce-pcx-5900.c56 it was NOT released in 2003, that's false information.

This is not way how to do edits. Copy some non-reference table and overwrite old (correct) values. Often, over years, it was corrected here. Lofr2 (talk) 16:07, 3 June 2026 (UTC)

Defapianoplayer still didn't address a single error that has been pointed out

Defapianoplayer still didn't address a single error that has been pointed out in his edits, and not only that he seems fixated in using TPU as THE ONLY SOURCE, despite clear errors on their database, he is also vandalizing other people edit if they don't use TPU as their source

And funnily enough, he has the nerve to say to not put personal interpretation, when he butchered the whole vertex rate from the Geforce3 and onwards, apart for other errors that were pointed earlier in this thread

He is either Vandalizing the article for his own sake, or working for TPU ~2026-34743-79 (talk) 01:06, 13 June 2026 (UTC)

Splitting proposal

I propose a split of this article into the following:

• List of Nvidia graphics processing units (266+ kB)
• List of Nvidia desktop graphics processing units (304 kB)

In this split, the desktop section will be condensed into a simplified table. If there are no major objections, I will perform this split around late June. zsteve21 (talk) 15:01, 13 June 2026 (UTC)

The whole desktop section needs reworking, since now is it full of errors and riddled with unnecessary data, but it would be better if all entries remained under the same page ~2026-34848-66 (talk) 00:41, 14 June 2026 (UTC)

I suggest:

1. Simplify tables, remove unnecessary columns (e.g. power connectors), citations in every GPU type, less lines in tables (more as it was before march 2026, not every odd variation of GPU)

2. Detach only mobile GPUs (in first stage). Keep desktop and workstation GPU in one article.

Should be enough for years as I don't expect new generations of GPUs to be too complicated and add too many lines & tables. Lofr2 (talk) 03:46, 14 June 2026 (UTC)

@~2026-34848-66 @Lofr2 Which data or columns should be removed? We can see that the tables are beyond the page width limits. zsteve21 (talk) 20:37, 22 June 2026 (UTC)