Talk:GeForce 900 series

Untitled
This page is unjustified and speculation. There is no talk anywhere of this being the Nvidia 900 series. This page should be deleted as it provides false information. — Preceding unsigned comment added by Oatman13 (talk • contribs) 22:53, 19 March 2013‎ (UTC)

Double precision problem
The information about the double precision seem to be partially contradictory to what was stated on the NVidia CUDA webpage: http://blogs.nvidia.com/blog/2013/03/13/geforce-gtx-titan-cuda/ — Preceding unsigned comment added by WikimatCS (talk • contribs) 15:54, 4 November 2015 (UTC)

Wikipedia is time traveling now?
" GeForce 900 series cards were first released in September 2014, starting with the release of the GeForce GTX 980 and the GeForce GTX 970 on September 19, 2014." considering its only Sept. 3rd this seems like speculation to me — Preceding unsigned comment added by 170.202.22.2 (talk) 15:39, 3 September 2014 (UTC)

I agree with statement for the most part. Though there is a report that the NDA will be lifted on the 19th of September: http://videocardz.com/51426/nvidia-to-skip-geforce-800-series-geforce-gtx-980-and-gtx-970-mid-september. But still everything here is speculation until Nvida does lift the NDA. — Preceding unsigned comment added by Danjw1 (talk • contribs) 17:41, 10 September 2014 (UTC)

Pixel fill rate calculations
I just redid the pixel fill rate calculations because http://techreport.com/blog/27143/here-another-reason-the-geforce-gtx-970-is-slower-than-the-gtx-980 has proven that our method is wrong. We might need to rearchitect the table to account for the amount of active rasterizers, the number of active streaming multiprocessors, and the fragments both of those can process per cycle. Since the pixels are only generated by the ROP from one or more fragments depending on the antialiasing mode, the rasterizers and streaming multiprocessors can force some ROPs to go idle. Jesse Viviano (talk) 18:23, 13 January 2015 (UTC)

Paragraph I removed from the GM2xx section
Having written a new section about it, I removed this stuff about the 970 specs controversy. It has some good info, but I think it's probably too technical for the article, so I'll just leave it here.

--uKER (talk) 15:40, 28 January 2015 (UTC)
 * Nvidia revealed that it is able to disable individual units each containing 256KB of L2 cache and 8 ROPs without disabling whole memory controllers. This comes at the costs of dividing the memory bus into high speed and low speed segments that cannot be accessed at the same time for reads because the L2/ROP unit managing both of the GDDR5 controllers shares the read return channel between which makes simultaneous reads between the two GDDR5 controllers impossible. This is used in the GeForce GTX 970, which therefore can be described as having 3.5 GB in its high speed segment on a 224-bit bus and 512 MB in a low speed segment on a 32-bit bus.
 * It is needed to explain why Nvidia is able to claim the peak speeds it is claiming. However, Nvidia is getting in trouble for false advertising due to disabling one L2/ROP unit without telling people about it. I have restored the paragraph. Jesse Viviano (talk) 23:00, 28 January 2015 (UTC)

Is this true? PC Perspective claims:


 * UPDATE 1/27/15 @ 5:36pm ET: I wanted to clarify a point on the GTX 970's ability to access both the 3.5GB and 0.5GB pools of data at the same. Despite some other outlets reporting that the GPU cannot do that, Alben confirmed to me that because the L2 has multiple request busses, the 7th L2 can indeed access both memories that are attached to it at the same time.

Is Anandtech the other outlet? I don't see how it can be true, since GPU performance would not be where it is above 3.5GB, if both pool couldn't be read at the same time. 82.0.122.93 (talk) 21:09, 2 February 2015 (UTC)

Both the Anandtech and PCperspective articles state that while it isn't possible to read from both memory segments at the same time it is possible to utilize the full 224GB/s bandwidth by reading from one segment while writing to the other. read/read = NO write/write = NO read/write = YES  — Preceding unsigned comment added by 124.148.125.233 (talk) 01:44, 6 February 2015 (UTC)

Where does the PCperspective article state that? 82.0.122.93 (talk) 20:03, 21 February 2015 (UTC)

Memory bandwidth error for 970
It's currently listed as 192 + 28, when it should be 196 + 28. I created an account to try to change it, but I don't have an edit tab when I'm on the page. After all, those two numbers still need to add up to 224 GB/s. RazrLeaf (talk) 14:42, 10 March 2015 (UTC)

✅ --uKER (talk) 14:48, 10 March 2015 (UTC)

"False advertising" category
I see that there has been an edit war over whether or not to include this article in the category "False advertising". I temporarily protected the article to stop the edit war. But I believe the people deleting this category have a point. The category refers to the "controversy" section, about a particular item that did not perform as the manufacturer had claimed. But I did not see the term "false advertising" anywhere in that paragraph. And although the paragraph has many references, I did not see any Reliable Source using the term "false advertising" for this controversy. If it wasn't called "false advertising" by a Reliable Source, cited in the article, then it should not be categorized as such. Does anyone have such a source? If so, can they please add it to the article? And if not, we should delete that category. Calling the incident "false advertising" without a Reliable Source is editorializing and/or original research. --MelanieN (talk) 21:42, 26 March 2015 (UTC)
 * Ok, well, I did some searching on my own and found the class-action lawsuit alleging false advertising. So I added it to the article, and the category now does appear to be appropriate. --MelanieN (talk) 17:05, 27 March 2015 (UTC)
 * I just added the section and put the 970 controversy in there. also added the new false advertisement claim by the developer Oxide, about lacking DX12 compliance. We could expand the section about direct compute as well, as the cards dont seem to be able to do it natively, though im not sure if that is part of the any "promise" made by nvidia about those cards (is it part of DX compliance or not?)... Ceremony64 (talk) 18:04, 31 August 2015 (UTC)

Incorrect /outdated information in "Limited DirectX 12 support".
Unfortunately this section has a few outdated sources, mostly from an AMD (Nvidia's competitor) affiliated game developer (Kollock, from Oxide), who made claims about lack of ASync support, only to be corrected later on by Nvidia.

Several of the lines are contradictory, at first implying that indeed Maxwell is absolutely incapable of Async compute, then suggesting that it merely hasn't been implemented yet and Nvidia+Oxide are already working on a full implementation, which is the latest update in the situation. 

For the third paragraph, the claim in the article is this: "Oxide claims to have been pressured by Nvidia not to include the asynchronous compute feature in their benchmark at all" (i.e. total disabling of the setting) Whereas if you actually look at the source given (as well as #43), the quote says this: "NVIDIA wanted the Asynchronous Compute Shaders feature level disabled by the dev (Oxide) for their hardware as it ran worse." (disabling ONLY for their own hardware, not for any/all hardware - And no quote from the developer supporting the claim in the article. In fact, to quote the relevant person, "the only 'vendor' specific code is for Nvidia where we had to shutdown async compute."

And in any case - ASync compute isn't even part of the DX12 specification, so I don't know why it's discussed under the DX12 subheading. You can see this both in Wikipedia's own Direct3D Feature level article, and also in this article which also features quotes from the same developer from Oxide "It’s also worth noting, as Kollock does, that since asynchronous compute isn’t part of the DX12 specification, its presence or absence on any GPU has no bearing on DX12 compatibility.": PS: Apologies for inevitable formatting mistakes. — Preceding unsigned comment added by 110.175.244.197 (talk) 15:10, 23 November 2015 (UTC)


 * AFAIR Asnyc Compute is part of the DX12 specification but as optional feature, not as requirement. And this feature seems not fully supported in Hardware.--Denniss (talk) 15:26, 23 November 2015 (UTC)


 * And yet you're contradicted by a professional DX12 developer and the author of the tech-news article quoting him, with no strong reference to say otherwise. I could go and fetch a whole bunch of different articles which explain DX12's specifications and you would find a complete lack of any mention of Async compute. As I said it's also not part of the 'Feature levels in Direct3D' wiki page (nor any others I could find), which of course DOES cover optional features as a separate category, which is where ASync would fall if it were such a feature. Considering this sort of thing is the prime purpose of that page, you'd think it'd be more believable than a small footnote elsewhere. And in any case regardless of whether the bulk of it is handled in hardware and minor parts are handled in software, the support is FULL according to both Nvidia and the very same devs that earlier suggested that support was lacking (it's intellectually dishonest to claim a feature isn't supported just because minor parts of it are handled in driver software). As a result both the (sub)title is misleading, as well as the vast bulk of the content which strongly implies that Maxwell is not fully DX12 and/or ASync compute compatible using outdated sources, which is then contradicted by a single tiny sentence using updated info from those same sources (Oxide and various websites). There's also the other issues I brought up. PS: Again apologies for my unfamiliarity with this formatting style etc. 110.175.244.197 (talk) 10:11, 1 December 2015 (UTC)

How are the "Memory (MT/s)" calculated?
I can't figure out how the Megatransfers are calculated. Can someone explain it to me? Also maybe it is worth to also add a footnote for it, likewise to the pixel and texel fillrate. Wikiinger (talk) 07:01, 21 June 2016 (UTC)

(adaptive) V-SYNC removed in 970M and 980M?
Unfortunately I cannot find any information on the Nvidia website, so does GM204 still support it? Or has it been replaced with G-SYNC?

If not it should be added to the according section as a removed feature? — Preceding unsigned comment added by 212.202.253.130 (talk) 09:59, 15 December 2016 (UTC)

External links modified
Hello fellow Wikipedians,

I have just modified one external link on GeForce 900 series. Please take a moment to review my edit. If you have any questions, or need the bot to ignore the links, or the page altogether, please visit this simple FaQ for additional information. I made the following changes:
 * Added archive https://web.archive.org/web/20130629025423/http://www.xbitlabs.com/news/graphics/display/20130412175120_Nvidia_Next_Generation_Maxwell_Architecture_Will_Break_New_Grounds.html to http://www.xbitlabs.com/news/graphics/display/20130412175120_Nvidia_Next_Generation_Maxwell_Architecture_Will_Break_New_Grounds.html

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

Cheers.— InternetArchiveBot  (Report bug) 03:44, 12 October 2017 (UTC)