Talk:Radeon 400 series

Out of date.

"Polaris 10 may feature 2304 stream processors across 36 Compute Units (CUs), and support up to 8GB of GDDR5(X) memory on a 256-bit memory interface. " By now we know most of this, I beleieve the number of stream processors is still not known for sure but the other three specs we know. 145.132.75.218 (talk) 13:45, 2 June 2016 (UTC)

RX 480 Is released

The RX 480 has been released as of today and this article needs updating — Preceding unsigned comment added by 174.116.43.58 (talk) 14:18, 29 June 2016 (UTC)

Empty TBA fields?

Could someone do a bit of research to figure out some of the TBA fields? Great job with the article, it's comming along well. — Preceding unsigned comment added by Millzie95 (talk • contribs) 15:03, 4 July 2016 (UTC)

Non-free image removal

I have had a conversation with Jesse about the branding images for the Radeon 400 series. https://en.wikipedia.org/wiki/User_talk:Jesse_Viviano#Non-free_image_removal

If I may summarize: The non-free image of the Radeon branding on the AMD Radeon 400 series page was removed as invalid. However, the image that appeared was correct and current on the official AMD Logos and Images Brand Assets site.

The argument against this logo: New logos for the Radeon series in general and for the RX 480 can be found at http://radeon.com/ and http://www.amd.com/en-us/products/graphics/radeon-rx-series . The logo does not apply to just the 400 series because AMD redesigned the Radeon logo for this series. Logos for the series are permissible on the series' pages, while the logo for the Radeon brand belongs on Radeon and AMD (because it is one of AMD's main brands), but not on the series' pages.

I would argue that the most recent officially sanctioned Radeon branding. I agree that it appears AMD has altered their branding on the product page. However the currently radeon brand logo on wikimedia remains the latest officially sanctioned logo and should be used until an update is officially released by AMD. Without verification through an official channel, the assertion that the branding has changed is conjecture. Dbsseven (talk) 16:52, 6 July 2016 (UTC)

It looks like the logos on the mars.amd.com link are as current as of "30.11.2015 12:42" as copied from that page and therefore will never be updated. Someone will have to execute an agreement with AMD about its branding to get a link to a message with the current logos as of now. Jesse Viviano (talk) 21:56, 6 July 2016 (UTC)

AMD Radeon history?

AMD Radeon 400 series#Features. What is this in here for? The full Radeon history in a 400 series article? --Diblidabliduu (talk) 15:55, 12 July 2016 (UTC)

I understand this is a leftover from since [1]. No need for it anymore, right? --Diblidabliduu (talk) 13:43, 13 July 2016 (UTC)

I agree this is unnecessary on a generation specific page Dbsseven (talk) 17:18, 13 July 2016 (UTC)

where can i edit the Chipset table???

Looking HERE and on this page but nowhere the table... Could somebody please insert the new facts from AMD Radeon RX 470 and Radeon RX 460 official specs and performance -- 195.189.93.113 (talk) 08:17, 28 July 2016 (UTC)

you can edit the RX 400 series on the template page. however, the facts you want to insert are still unverified and not a reliable source. Therefore these specs should not be included yet, until they are verified (tonight, based on the NDA note on the purported slides) Dbsseven (talk) 15:00, 28 July 2016 (UTC)

thanks, where can i find this page? And the headline speaks of "official specs" so it seems for me reliable... -- 84.160.126.57 (talk) 10:13, 6 August 2016 (UTC)

Lead length

The lead to this article is getting pretty long. What about moving everything but the first section into the body and breaking it into sections for Polaris and Vega? This would help as details of the 470 and 460 (Polaris 10) are coming out. Dbsseven (talk) 13:49, 29 July 2016 (UTC)

Fill Rates

(I posted this on the template page without response, reposting here.) There appears to be a discrepancy in fill rates between the citations and the formula defined and used here. The formula here uses the base clock rate to calculate fill rates, essentially giving a minimum. But equally valid are the published fill rates, which appear to be calculated using the boost clock. It seems to me that as it is against Wikipedia's policy to include Original Research, we should use the published fill rates. Dbsseven (talk) 13:37, 16 August 2016 (UTC)

All AMD GPU tables seem to use calculations based on standard clock so this should be retained. Boost is nice but is highly variable depending on load.--Denniss (talk) 21:41, 16 August 2016 (UTC)

That is the usual way of doing things, but is it correct? Boost is variable depending upon load, but so it the processing power which we do not derive. As I understand it, it would be more appropriate to use published values, per Wikipedia's No Original Research policy. Perhaps unless there is not published value? Dbsseven (talk) 23:13, 16 August 2016 (UTC)

A valid comparison can only be made at standard clock rates. It's not OR to calculate this from the given hardware units.--Denniss (talk) 00:38, 17 August 2016 (UTC)

But it is to assert the current wikipedia definition is the "correct" fill/texture rate. This is substituting that definition with a citable source's. Dbsseven (talk) 22:29, 22 August 2016 (UTC)

Put another way: where is the citation for the current definition of the fill/texture rate? Dbsseven (talk) 23:48, 22 August 2016 (UTC)

Calculation base is shown at each table + also confirmed by sites using the incorrect boost clock for this. A valid comparison across manufacturers is only possible at guaranteed/base clock rates, not at variable boost (AKA overclock) rates as the latter is never stable under all work loads.--Denniss (talk) 09:29, 23 August 2016 (UTC)

I believe my question was misunderstood, where is the citation validating that the base rather then boost clock should be used? Otherwise, the use of base clock is in conflict with the source material where boost is used. I know the boost clock would give an unstable maximum rather than a stable minimum. And yes, this does make comparison more difficult, but wikipedia is not a benchmark. Is there an alternative citable source for the Base-clock derived values? Dbsseven (talk) 15:33, 23 August 2016 (UTC)

Series Codename

Before the codenames Polaris and Vega were released, the 400 series was known as the Arctic Islands (following the trend of the Volcanic Islands, the Southern Islands, etc) and is referred to as such in articles such as Radeon and Graphics Core Next. However, it is unjustified to name the series the Arctic Islands since AMD themselves do not call it that, nor do the names Polaris and Vega reflect the title like previous series and chip codenames did so. Therefore, shouldn't the series be addressed as either simply the 400 series or as Polaris and Vega in other articles? It is unfortunate that AMD is not calling the series the constellations or something similar. Pasclaris (talk) 14:27, 19 August 2016 (UTC)

Mobile 4xx Data Source techpowerup

https://www.techpowerup.com/gpudb/2854/radeon-r7-m465

https://www.techpowerup.com/gpudb/2855/radeon-r7-m465x

https://www.techpowerup.com/gpudb/2856/radeon-r9-m470

https://www.techpowerup.com/gpudb/2857/radeon-r9-m470x

https://www.techpowerup.com/gpudb/2858/radeon-r9-m485x

two things. Citations should go in the article, not here. Also, I don't believe techpowerup to be a reliable citation. Dbsseven (talk) 16:40, 22 November 2016 (UTC)

Source AMD is small or not Actual. — Preceding unsigned comment added by 2A02:810B:C53F:B9E8:A4D9:2425:3402:C2F7 (talk) 14:49, 23 November 2016 (UTC) See many unknowns — Preceding unsigned comment added by 2A02:810B:C53F:B9E8:A4D9:2425:3402:C2F7 (talk) 14:51, 23 November 2016 (UTC)

OpenCL 1.x and 2.x conformance

https://www.khronos.org/conformance/adopters/conformant-products#opencl

AMD is not Up to Date like Intel and Nvidia with their new products. — Preceding unsigned comment added by 95.90.229.220 (talk) 16:26, 22 November 2016 (UTC)

New Chipset table layout proposal

Hey, I designed a slightly modified and more straightforward (IMHO) version of the Chipset table here: Template:GPU Chipset Table
which I would like to use instead of the current one and also for future GPU chipset tables.
Please check-it out, read the talk page for the rationale and the steps taken and leave comments. Thanks! --Wikiinger (talk) 19:55, 6 February 2017 (UTC)

Vega is not a part of the 400 series

The 400 series only contains Polaris 10 and 11, as well as older Graphics Core Next chips from older generations. Vega 10 and 11 are not expected to join the 400 series. Instead, Polaris 21, 22 and 12(new low-end chip) are thought to make up the 500 series starting April. Those may be based on a modified 14nm (LPP instead of LPE) process: https://benchlife.info/radeon-rx-500-with-14nm-finfet-lpp-process-03192017/ (Now, those are not actual verified facts I expect to be added anywhere, but I thought those were interesting to note anyway)

Vega (10 and 11) will be marketed as "Radeon RX Vega", as announced on the Capsaicin & Cream Event. http://www.anandtech.com/show/11164/amd-gdc-2017-asynchronous-reprojection-for-vr-vega-gets-a-server-customer-more

So, Vega should probably separated from this article and be moved to a new Radeon 500 article by release of the new GPUs. 149.172.134.182 (talk) 00:21, 21 March 2017 (UTC)

• I forgot to mention: Vega is definitely a newer architecture, not the big chip of GCN4. It is GCN5 (AMD internally part of the GFX9 family, compared to Polaris' GFX81). http://www.anandtech.com/show/11002/the-amd-vega-gpu-architecture-teaser (The information in the article is confirmed and could already be written up in a future Radeon 500 article, if someone wants to start one) 149.172.134.182 (talk) 00:25, 21 March 2017 (UTC)

I generally agree that these are the rumors, but I have not seen a citable source discussing the RX 500 series or if Vega will be included in this product line. I would be hesitant to remove this content until we have more firm evidence it is true. When we know, it should be logical where the information should go. Dbsseven (talk) 16:49, 21 March 2017 (UTC)

Radeon 500 series comming

1. I miss the link to the successor 2. Four graphic cards have know by now, so please add them in the new article Thanks in advance -- 195.189.94.14 (talk) 10:29, 23 March 2017 (UTC)

RX 460 PCIe lane width

I have an RX 460 and the PCIe lane width is reported as x8, not x16 as listed here. — Preceding unsigned comment added by Lostdistance (talk • contribs) 19:04, 12 December 2017 (UTC)

If that's not a connection issue or a problem with the card itself it may be some nonstandard or OEM card. Please check via GPU-Z with Render Test running. --Denniss (talk) 21:58, 12 December 2017 (UTC)

It's a retail card - XFX RX-460P4HFG5. I use Linux, and the lspci command reports "LnkCap: Port #0, Speed 8GT/s, Width x8, ...". Given that the similarly specified Radeon RX 560D is also x8, is there any reason to think that the RX 460 should be x16?— Preceding unsigned comment added by Lostdistance (talk • contribs) 00:08, 13 December 2017 (UTC)

Smells like a revision of the 460 using a PCB (and chip?) from the 560, this combination is known to have only x8 connection. EDIT: several german retailers state x8 connection for this card, both 2GiB and 4GiB version). May be limited to the special passive edition. --Denniss (talk) 08:36, 13 December 2017 (UTC)

The XFX RX-460P4HFG5 was on sale in 2016; the RX 500 series was announced in April 2017. Reviewing the websites of the major RX 460 manufacturers (Asus, Gigabyte, MSI, Sapphire, PowerColor, XFX), only Gigabyte and MSI state a PCIe lane width: x8. None of the manufacturers state a PCIe lane width of x16. Where is the authoritative source for x16? — Preceding unsigned comment added by Lostdistance (talk • contribs) 10:03, 13 December 2017 (UTC)

If one searches the net with rx460 and x8 then a lot of info is found. THG had a test of two cards of which one was x8 and one x16. I have not seen any info from AMD regarding this during launch but they also had no reference cards issued but let board partners do the cards on their own. It is possible that some x16 cards were available but as of now most are likely x8 (still sufficient though). --Denniss (talk) 13:16, 13 December 2017 (UTC)

Vulkan 1.1 only for GCN 2nd gen. or higher?

Vulkan 1.1 only for GCN 1.2 (2nd gen.) or higher See Khronos Vulkan product Table. No Vulkan 1.1 for 1st GCN gen.

See https://www.khronos.org/conformance/adopters/conformant-products/vulkan — Preceding unsigned comment added by 95.90.228.17 (talk) 23:37, 19 April 2018 (UTC)

GCN 1.1 or 1st gen in Khronos are not conformant to Vulkan 1.1. They are only compatible to the Driver with Vulkan 1.0 as a Subset. it is the Same for OpenCL. Only OpenCL 2.x Cards can Do Vulkan 1.1 and that is GCN 1.2 and higher.

See also OpenCL in Khronos See

Nvidia is here better. Kepler and higher can Do Vulkan 1.1. All products of the Last years can Do it and not only half. In OpenCL they are less good with OpenCL 1.2+.

References

The above comment was written April 2018 (back when Vulkan 1.1 was released) but in March 2020 someone anonymously added that Vulkan 1.1 is now supported with newer drivers, and also added it at Radeon HD 7000 series#Architecture. I couldn't verify it though it seems plausible, so I added a cite-needed tag. Meanwhile Radeon HD 7000_series#Radeon Feature Matrix still says only Vulkan 1.0 is supported on GCN 1.0. --TeeEmCee (talk) 12:58, 10 September 2020 (UTC)

Vulkan 1.2 available for GCN 1.2 Equal 2nd Gen and higher in Windows and Linux Mesa 20.0

See sheet 457+ in https://www.khronos.org/conformance/adopters/conformant-products/vulkan

See https://www.amd.com/en/support/kb/release-notes/rn-rad-win-20-1-2

See new Features in https://mesa3d.org/relnotes/20.0.0.html — Preceding unsigned comment added by 2A02:6D40:3485:2101:7045:2013:7210:742C (talk) 00:04, 1 March 2020 (UTC)