[wccftech]Nvidia Volta Allegedly Launching In 2017 On 12nm FinFET Technology

[happy medium]

More like early 2018 . Looks like Nvidia is not slowing down.
AMD need to launch Vega now, with volta coming so soon.

http://wccftech.com/nvidia-volta-12nm-finfet/

 

[Flapdrol1337]

wtfftech though

 

[Krteq]

WCCFTech - take it with mountain of salt

 

[Qwertilot]

Well yes, but it *is* an awfully plausible time scale. iirc the contracts that NV have out more or less dictate that there will be some - maybe super computer only - Volta production in 2017? There's that Soc using it that they showed at CES too (Xavier). Definitely less than no reason to expect a major delay on it. Both TSMC (Apple's money!) and NV are executing very nicely past few years.

Best reason to expect Spring 2018 for consumer stuff is probably that they haven't shipped the 1080ti yet and that'll surely get given a fair run as the top chip?

 

[Sweepr]

Better source: GV100 will be announced @ GTC 2017 (May 8-11th)

http://news.mynavi.jp/articles/2017/01/19/nvidia_volta

 

[CakeMonster]

Still on the yearly schedule for the consumer part. Not like there wasn't a 90% chance of that before reading this rumor anyway..

 

[moonbogg]

We have another 2 to 3 years before they release the card people actually want to see. We'll get the $800 mid rangers coming in first.

 

[VirtualLarry]

Which foundries offer a 12nm node? I haven't heard about that one at all. Half-shrink 14nm?

 

[ConsoleLover]

12nm? Now I know this is redacted. 12nm is set for 2018, this is assuming everything goes PERFECT. I hardly doubt they are creating it on a new in development process.

I do think they are probably not that further out, I do think it will probably launch in 2017, I mean by summer 2017 they would have had 1 years after their last major release to work on it, and of course its likely been in development for at least 1 year prior to that, so I expect a late 2017 release, probably early Q4.

AMD needs to hurry up though and release Vega ASAP. I thought they would have at least had more substantial info now and maybe even announcement of when its going to launch in terms of a month. I feel like if Vega is basically May or later it would be worthless. It just won't achieve anything, even if its say 15% faster than Nvidia's offerings right now, it would be too late and by that time talk ov Volta would pause many purchases of Vega.



No profanity allowed on the technical forums.


esquared
Anandtech Forum Director

 

[happy medium]

I feel like if Vega is basically May or later it would be worthless. It just won't achieve anything, even if its say 15% faster than Nvidia's offerings right now, it would be too late and by that time talk of Volta would pause many purchases of Vega.

I must agree.....Why release Vega when in 8 months something much faster and feature packed will be out?
I think Nvidia will refresh Pascal and make maximum money before they talk Volta.
If there is no Pascal refresh, I think Volta will be out quicker.

 

[tviceman]

All of it sounds too good to be true. Nvidia is rarely ever on the cutting edge of nodes and they managed to get several different generations of chips out of each of the last several nodes (they had GT 240 and a couple other chips from that family on 40nm before Fermi, they had Kepler on 28nm before Maxwell), I just don't see them migrating to 12nm with an entire new lineup when there is plenty of die space left to improve single-precision performance on GP102, GP104, GP106, and GP107. GP100 hasn't even been out for a year, GF110 lasted 3+ years on the market and wasn't even as aggressive in it's HPC design as GP100 is.

 

[krumme]

More like early 2018 . Looks like Nvidia is not slowing down.
AMD need to launch Vega now, with volta coming so soon.

http://wccftech.com/nvidia-volta-12nm-finfet/

Man i actually clicked that link !!! Lol.

Glad i did. I can read Amd will launch 7nm gpu in 2018. Happy times is comming.
/sarcasm

 

[Arachnotronic]

All of it sounds too good to be true. Nvidia is rarely ever on the cutting edge of nodes and they managed to get several different generations of chips out of each of the last several nodes (they had GT 240 and a couple other chips from that family on 40nm before Fermi, they had Kepler on 28nm before Maxwell), I just don't see them migrating to 12nm with an entire new lineup when there is plenty of die space left to improve single-precision performance on GP102, GP104, GP106, and GP107. GP100 hasn't even been out for a year, GF110 lasted 3+ years on the market and wasn't even as aggressive in it's HPC design as GP100 is.

TSMC "12nm" is an enhanced version of 16nm.

 

[happy medium]

I just don't see them migrating to 12nm

TSMC "12nm" is an enhanced version of 16nm.

Exactly, and then you milk the Volta line with a 7nm refresh.

 

[xpea]

All of it sounds too good to be true. Nvidia is rarely ever on the cutting edge of nodes and they managed to get several different generations of chips out of each of the last several nodes (they had GT 240 and a couple other chips from that family on 40nm before Fermi, they had Kepler on 28nm before Maxwell), I just don't see them migrating to 12nm with an entire new lineup when there is plenty of die space left to improve single-precision performance on GP102, GP104, GP106, and GP107. GP100 hasn't even been out for a year, GF110 lasted 3+ years on the market and wasn't even as aggressive in it's HPC design as GP100 is.

more than 10k GV100 must be shipped this year for SUMMIT super computer alone. GV100 reference can be found in drivers for a year. Xavier will also be sampled by end of the year.
HPC and SoC parts always ship after gaming silicon.
Draw your own conclusion but Volta schedule is very well know...

 

[tviceman]

more than 10k GV100 must be shipped this year for SUMMIT super computer alone. GV100 reference can be found in drivers for a year. Xavier will also be sampled by end of the year.
HPC and SoC parts always ship after gaming silicon.
Draw your own conclusion but Volta schedule is very well know...

TSMC "12nm" is an enhanced version of 16nm.

Assuming this is all true and Volta makes an appearance this year, that would make Pascal one of the most short-lived GPUs ever, which is surprising given it's gigantic sales success (revenue and profit).

 

[Arachnotronic]

Assuming this is all true and Volta makes an appearance this year, that would make Pascal one of the most short-lived GPUs ever, which is surprising given it's gigantic sales success (revenue and profit).

Pascal is going to be ancient soon, this thing debuted in May 2016.

In four months, it's going to be a year old. Product cycles should ideally be one year anyway, but I don't think we see Volta until early 2018.

 

[xpea]

Pascal is going to be ancient soon, this thing debuted in May 2016.

In four months, it's going to be a year old. Product cycles should ideally be one year anyway, but I don't think we see Volta until early 2018.

HPC Volta will be introduced at GTC 2017 in May for sure (confirmed by Dally in Japanese media) for shipment starting in Q3.
On the other hand, GV104 announcement is still a mystery even if the chip design is probably finished...

 

[RussianSensation]

Assuming this is all true and Volta makes an appearance this year, that would make Pascal one of the most short-lived GPUs ever, which is surprising given it's gigantic sales success (revenue and profit).

Not really. Pascal is just Maxwell+ on a new node. The underlying architecture traces its roots to September 2014. Besides, the article discusses Volta for professional applications not consumer. For consumer, NV had Volta for 2018 for the last 2 years and it's still 2018.

NV's bifurcating a generation strategy is clear:

- Kepler mid-range GTX680 March 2012
- Kepler high-end GTX780Ti Nov 2013

- Maxwell mid-range GTX980 Sept 2014
- Maxwell high-end GTX980Ti May 2015

- Pascal mid-range GTX1080 May 2016
- Pascal high-end GTX1080Ti/2080Ti March-April 2017 (rumoured March 10, 2017)

Using last 3 generations, we should see Volta GV104 mid-range launched as a GTX2080/3080 in Q1-2 2018, followed up by Volta GV100/102 high-end in Q1-2 2019.

Vega looks to underdeliver and underperform which may result in higher prices of GV104, or more cut-down x70 marketed as an x80, or even GV106 becoming a GTX2080/3080 $599+ card. If NV does launch Volta for consumer, I could see it in the lower segments as a repeat of GTX750/750Ti strategy. It wouldn't make sense for NV to cannibailze what is likely to be an expensive GP102 1080Ti with a chip that's faster and uses 30-50% less power for a cheaper price when AMD is nowhere in the map. NV is going for > 60% gross margins which means keeping prices as high as possible and manufacturing costs as low as possible. Volta doesn't seem to fit that target for 2017 mid-range and high-end consumer segments.

 

[Mondozei]

Don't really get the people saying that "Pascal will be shortlived". If you had actually paid attention, you'd noted that the NV pattern has stayed the same since Kepler, as RS pointed out.

The Kepler refresh added around 30% of extra performance compared to the original Kepler line. I doubt they'd need to add as much this time around to stay competitive with Vega.

Volta will only be out for HPC this year. Even the car-centric Xavier SoC will only sample this year. Full-scale production is likely next year.

When GV104 comes out, it will most likely be a full 24 months after Pascal was introduced. In 2019 we'll see high-end Volta.

Where I differ with RS is that I don't think we'll see high-end Volta(GV102) until H2 of 2019. Just as I don't think we'll see GV104 until mid-summer of next year. The reason for both predictions are the same: AMD is very slow these days.

 

[Qwertilot]

I dunno, NV really do seem to like their annual update cycle each Spring, so I don't think they'll miss that if it is sensibly possible to do so. Its simply good for business to get things predictable.

 

[happy medium]

gtx680 series = gtx1080 last year
gtx780 series = gtx1180 this year
gtx750 ti Maxwell v1 = Volta low end at the end of this year?
gtx980 Maxwell v2 (new architecture) = Volta 2080 early next year

I see the pattern.

 

[Mopetar]

Don't really get the people saying that "Pascal will be shortlived". If you had actually paid attention, you'd noted that the NV pattern has stayed the same since Kepler, as RS pointed out.

I think it depends on how people are using the term. I don't suspect NV is in a hurry to release new cards as AMD doesn't have any competing products in a lot of market segments so there's little need for a refresh. In that regard, I don't think Pascal is going away soon. We may see a refresh though as part of the reason Pascal isn't a radical new architecture itself is because NVidia wanted to avoid building a new architecture on a new process, which just increases the risk factor. The thinking may be the same for Volta and the refresh could be to help them understand the process characteristics with some known good hardware.

What I think some people mean by the term is that Volta is going to be a major architectural shift and that Pascal is not going to age particularly well, even by more recent NVidia standards. I don't think that's really much of an issue for forum users though as most will upgrade pretty quickly as evidenced by the most recent hardware cycle.

 

[Ajay]

TSMC "12nm" is an enhanced version of 16nm.

Lols, now foundries are rebranding process nodes. Too bad, 16FF++ would have been more geek friendly.

 

[Teizo]

Vega looks to underdeliver and underperform which may result in higher prices of GV104.

I sincerely hope that is not the case. AMD is in good position on the CPU side to finally gain some market share back. It is definitely a missed opportunity for Vega to not launch along with Ryzen. If they could have managed their timelines better, they would have really made bank on total system update sales. Now, people who buy Ryzen and looking to build new computers are going to left wondering when Vega will release and will be tempted to pair it with an Nvidia GPU.