Talk:GeForce 10 series

Article name

This should be called GeForce 1000 series, since it comes after the GeForce 900 series, not after the GeForce 9 series. --uKER (talk) 04:10, 7 May 2016 (UTC)

I would agree with this, unfortunately Nvidia has decided to confuse everyone and actually name the series the 10 series. We can only hope that everyone else decides to call it the 1000 series regardless and that Nvidia ends up changing their minds about the naming since everyone else calls it the 1000 series anyway, but we can't really do anything for now given: http://www.geforce.com/hardware/10series/geforce-gtx-1080 Erik.Bjareholt (talk) 08:09, 7 May 2016 (UTC)

'10' is the prefix, so it's 10xx. This pronunciation is correct, but the 10xx series is the successor to the 9xx. While the 9xxx already existed, but there was no 10xxx series. Millzie95 (talk) 23:32, 8 May 2016 (UTC)

10-series designation is more user-friendly, since the vast majority of regular users don't remember 9xxx-series anyway. So this is a good idea to call the new series 10-series, since it is more laconic and less confusing. TranslucentCloud (talk) 09:56, 16 May 2016 (UTC)

Yes, it's a logical progression, since the 900 series is nine-hundred, the 10 series is ten-hundred, rather than one-thousand. So 1080 is pronounced ten-eighty, rather than one-thousand-eighty, which makes more sense, as the naming system remains. Millzie95 (talk) 20:39, 16 May 2016 (UTC)

@Millzie95: That makes sense, thanks for explaining! Erik.Bjareholt (talk) 19:55, 21 May 2016 (UTC)

References & syntax

References at the bottom of this page need tidying up, with consistent syntax and dates for each of the links, a caption or brief description would help. Millzie95 (talk) 22:59, 8 May 2016 (UTC)

Also noticed there are multiple references linking to the same webpages, should we link them all to one reference instead of having duplicate references? References 2 and 5 for example. Millzie95 (talk) 07:18, 16 May 2016 (UTC)

Citations needed

The table in the Products section, in particular, seems to be speculating about some of the information. A lot of the information is cited,b ut not all of it. But Wikipedia is not a crystal ball. All the information must be cited with a reliable source (not a speculative blog post). This article is actually pretty good. The authors have worked hard to list the information we know for sure. But there's still room for improvement. Without a reliable citation, information should be removed. Unfortunately, that includes much of the info for the 1070 and some of the info for the 1080. --Yamla (talk) 14:57, 13 May 2016 (UTC)

I have updated the products table with new information and added 2 references, most of the specs for the 1080 should be nailed down by now. Specs in the 1070 row may still change as new information becomes available, I tried to get the syntax as consistent as possible but the links could use some work. Millzie95 (talk) 16:32, 13 May 2016 (UTC)

Thanks. If the specs for the 1070 are subject to change, they should be removed immediately and only reintroduced once we have a reliable source for them. --Yamla (talk) 16:39, 13 May 2016 (UTC)

I have attempted to clean up the table to remove uncited information. --Yamla (talk) 12:42, 15 May 2016 (UTC)

Hi, just to clear something up about my last post. I was referring to cells in the table marked with a "-", all values on my last edit were either confirmed specs, approximate values (~) or calculations based on them (~Ax~B=~C). Could you please revert to the previous table (then possibly remove any ballpark values?) as some columns have gone missing, as it's also tricky to get everything aligned correctly. Millzie95 (talk) 14:17, 15 May 2016 (UTC)

I specifically searched the citations given and could not find the numbers specified. You say these are confirmed, and that's great. That certainly means we can replace the numbers, but only if we have a reliable source (not a speculative source). I understand the embargo lifts in two days, so if we can't find a reliable source today, we should be able to very shortly. --Yamla (talk) 14:30, 15 May 2016 (UTC)

We need better citations, I have made a new table with the columns re-added but cleared a few cells in the top row, the references are from Nvidia and GPU-Z, they should show all the listed specs, tell if if I missed anything. Millzie95 (talk) 14:42, 15 May 2016 (UTC)

As previously noted, neither source seems to cite GP104-200 vs GP104-400. Neither source cites transistor count or die size. Neither source cites GDDR5 (vs 5X) in the 1070. It's critically important we only display information we have a reliable citation (rather than a speculative source) for. Please immediately remove any of the information you added unless backed up by the citations. Note that I did make a mistake removing support for OpenGL 4.5 and Vulkan; both are indeed provided in the citation, so long as you click a link. I still can't find any claim of OpenCL in the cites, but you haven't reintroduced those. --Yamla (talk) 15:05, 15 May 2016 (UTC)

In the first link, click on specs, then view full specs. In the second link, scroll down, then look at the cropped screenshot. Nvidia's own slides said the 1070 will use regular GDDR5 (not 5X) and GP104-400/GP104-200 are the variants of GP104 used in the 1080 and 1070. I haven't reintroduced the specs in the API section, but we can assume APIs supported by the 900 series will continue to be supported unless support is dropped for some reason or we have a source which tells us otherwise. Millzie95 (talk) 16:26, 15 May 2016 (UTC)

Approximate vs confirmed specs

Approximate spec values in the Products table should be marked with a tilda (~) rather than being calculated first then corrected later, this will help to differentiate between confirmed and rough specs. Millzie95 (talk) 16:52, 13 May 2016 (UTC)

No. We don't allow original research. If something is not confirmed, it should be removed from the article. --Yamla (talk) 12:43, 15 May 2016 (UTC)

I wasn't referring to original research, rather confirmed rough values we don't know exactly. ~6500 could be 6382 or 6616, but we know it's around 6500. Millzie95 (talk) 14:32, 15 May 2016 (UTC)

Founders Edition & MSRP

Should Founders Edition pricing be included alongside MSRP? Millzie95 (talk) 17:04, 14 May 2016 (UTC)

Sure it should. Done. TranslucentCloud (talk) 10:16, 16 May 2016 (UTC)

Thanks for your contribution TranslucentCloud. Millzie95 (talk) 20:30, 16 May 2016 (UTC)

Article improvements

The article is comming along nicely, more specific detail added and products table is updated with the most recent information we have, references have been improved and see also section has been expanded. Millzie95 (talk) 08:06, 15 May 2016 (UTC)

Single-precision performance of 1070

@210.187.221.194: is claiming [1] cites his/her claim that the 1070 has a single-precision processing power of 6500 GFLOPS. However, the article states, " Single precision performance is calculated as 2 times the number of shaders multiplied by the boost core clock speed." The article claims the 1070 has 1920 shader processors and the citation shows a boost clock of 1.6 GHz. 2 * 1920 * 1.6 is 6144, not the 6500 that the user is claiming. Perhaps the article is explaining how to do this calculation incorrectly? I'm concerned that this user is adding uncited and incorrect information and then launching personal attacks when challenged. --Yamla (talk) 16:22, 17 May 2016 (UTC)

http://images.nvidia.com/geforce-com/international/images/050616-geforce-dot-com/nvidia-geforce-gtx-1070/nvidia-geforce-gtx-1070-introducing-the-geforce-gtx-1070.png

Are you stupid or really unable to use your brain to find info? Nvidia says 6.5 TFLOPs right there on their website. — Preceding unsigned comment added by 210.187.221.194 (talk) 16:27, 17 May 2016 (UTC)

Enough with the personal attacks. Any more and you'll have to take a time-out. For other editors, our claim and the citation no longer matches up. Which is correct? Our claim on how you calculate the single-precision performance, or the nvidia website? --Yamla (talk) 16:30, 17 May 2016 (UTC)

I think the same can be said of GTX1080. 8223 is calculated by base core clock (1.607 * 2 * 2560 = 8227.84). but it could be 8873 calculated by boost core (1.733 * 2 * 2560 = 8872.96) — Preceding unsigned comment added by 124.35.68.250 (talk) 09:43, 18 May 2016 (UTC)

The GTX 1070 does have 1920 CUDA cores, 120 TMUs and 64 ROPs, the difference comes in the way single precision performance is calculated. Now, we have a number from Nvidia of 6500 GFLOPs, but this is an approximate figure. GFLOPs can be calculated to the nearest whole number, this can be done by multiplying the CUDA core count by 2, then multiplying that by the base clock or boost clock, both results are now listed in the products table but the boost clock figure is in brackets. Millzie95 (talk) 12:12, 19 May 2016 (UTC)

Reliable sources

210.187.223.163 (talk · contribs) removed external links in this edit, saying "Stop adding links that are not official Nvidia links". Actually, we prefer journalistic sources over vendor sources. That is, journalistic sources are better and we should use them in preference to vendor sources. See WP:RS, specifically the "Vendor and e-commerce sources" section. I have not reverted the removal because I have not verified whether or not the removed sources are of sufficient journalistic integrity, or whether they just serve as spam in this article. Their removal may be entirely appropriate. --Yamla (talk) 19:26, 18 May 2016 (UTC)

I agree, we need a mix of official and journalistic sources, claims from Nvidia's site might be comming straight from the horse's mouth, but we need third party sources to verify the flow of information from them. Now I see that journalistic sources have been removed from the products table, new journalistic sources should be added now we have the full specifications confirmed, as not all specs are listed on Nvidia's website so we can't 100% verify any of their claims, some independent journalism would be appreciated. Millzie95 (talk) 12:19, 19 May 2016 (UTC)

Floating point performance calculations

I've noticed there's a bit of back and forth from editors and users as new information comes out on the official clock speeds of the known 10 series parts. Some are editing floating point performance numbers to reflect (shaders * 2 * boost clock), some are instead following the proper (shaders * 2 * base clock) calculation. In some ways, they're both valid. With parts like these having a minimum boost clock that is much higher than their base clock, their peak FLOPS can be significantly higher than their rated base FLOPS (both the GTX 1070 and GTX 1080 are gaining somewhere in the region of 600-700 GFLOPS at minimum turbo). However, the technically correct calculation is (shaders * 2 * base clock), and no previous articles have displayed calculations using the boost clock of any particular card, such as the article for the GeForce 900 series. I personally feel that the boost clock calculation is unnecessary, but also inaccurate as Nvidia's Turbo Boost technology will allow cards to turbo well beyond the minimum rated turbo boost. --Arbabender (talk) 01:41, 19 May 2016 (UTC)

I guess you're right about how to calculate performance. but, first, you said, "Single precision performance is calculated as 2 times the number of shaders multiplied by the boost core clock speed." and "by the boost core " is clearly wrong. I think you should have written "by base core" so good. — Preceding unsigned comment added by Byunsangho (talk • contribs) 09:28, 19 May 2016 (UTC)

Thanks for your comments, both base and boost clocks for FP32, FP64 and FP16 are now included with the boost clocks in brackets. Millzie95 (talk) 12:26, 19 May 2016 (UTC)

Whitepaper

Thanks whoever added the whitepaper link, it's a good source and provides plenty of useful info. Millzie95 (talk) 14:45, 23 May 2016 (UTC)

i removed a broken link

there is no volta article so the link just went right for the geforce article.