[PCGamesn Rumor]Nvidia’s next-gen Volta GPUs to get a little die-shrink after all

[Det0x]

Read whole article @ http://www.pcgamesn.com/nvidia/nvidia-volta-gpu-specifications
Main point being Volta has been pushed back to 2018

Originally Volta was supposed to be sporting the 10nm process, but the pace of transistor shrinkage has become rather laggardly in recent years. The last rumour had Nvidia sticking with TSMC’s 16nm tech in order to be able to stick to their roadmap and get actual Volta cards on the shelves in 2018.

TSMC have though announced another stop-gap measure which they’re calling 12nm. It’s based on their existing 16nm design, with density, performance and energy efficiency improvements, but whether it will genuinely be packed with 12nm transistors or whether it’s just going to be clever marketing is about as clear as thermal paste.

We were expecting to get a glimpse of Nvidia’s Volta GPU tech at their upcoming Graphics Technology Conference in May. Pascal was unveiled at 2015’s GTC before appearing in professional form later that year with consumer-class GPUs appearing in 2016.

I’m not sure if Nvidia are going to be able to follow that timeline with Volta, however, as the Oak Ridge National Laboratory’s new Summit supercomputer, which was initially designed to launch with Volta GPUs this year, has been pushed back to 2018. The Summit website has recently been partially changed to detail the 2018 release.

More rumors can also be found @ http://www.fudzilla.com/news/42724-volta-might-get-a-die-shrink-after-all

For those who came in late, Volta was supposed to be a 10nm process, but that proved a little too tricky. The last rumour had Nvidia sticking with TSMC’s 16nm tech so it could keep to its roadmap and get Volta cards on the shelves before the sun turns into a red giant.

TSMC has, though announced another stop-gap measure which it's calling 12nm which is its existing 16nm design with wind spoilers and “go faster stripes”. TSMC claims to have pimped up density, performance and energy efficiency but actual 12nm transistors are about as likely as Jeremy Corbyn being British Prime Minister.

If they cant get the professional version of Volta out the door before in 2018, prepare to wait a long time for the consumer version. I would guesstimate mid/late 2018, or even as bad as 2019.

 

[greatnoob]

Is this 20nm all over again?

 

[Det0x]

Is this 20nm all over again?

Seems like it
Not sure how its going over at GlobalFoundries/Samsung tho

 

[Mondozei]

Is this 20nm all over again?

It has been well-known, and well-reported, that the major GPU vendors have been thinking about skipping 10 nm from the beginning and jump straight to 7 nm. The so-called "12 nm" is in fact just a refined 16 nm. So even if this so-called "stop-gap" solution is implemented, it still won't be true 10 nm. Therefore these rumors are just re-iterating the old news and repackaging it for shock value.

If they cant get the professional version of Volta out the door before in 2018, prepare to wait a long time for the consumer version. I would guesstimate mid/late 2018, or even as bad as 2019.

That isn't news. 2017 was never going to be consumer Volta. AMD is so slow these days that summer/autumn of 2018 was my basecase for consumer Volta. And it will only be GV104 for consumers. Bigly GV102 will only come year after, in 2019.

NV still has a lot of room to go with Pascal. Just look at how Kepler was elongated. Kepler Refresh gave us 30-35% extra performance. They can still do plenty of refinements and have plenty of mm^2 to go to push even more performance at the mid-high to top-end of the spectrum with Pascal.

I'd personally be surprised if we won't see Volta until 2019. We will most likely see it next year, but it will end up on 16 nm. Whether on the current 16 nm version or the refined so-called "12" nm, which is still 16 nm. Either way, I don't see much room for panic here.

 

[Qwertilot]

No room for panic at all. NV have managed their annual 20-30% YoY increase for several years now, and it seems very likely to continue for a while to come.

Its even fairly 'easy' to see how - Pascal refresh (maybe)/P102 this year/Super comp Volta, Consumer Medium Volta next, big Volta the year after that and by then TSMC are very likely to have a major die shrink ready for them to use which is two - four more years of improvements.
(Mind boggling difficult to actually do the engineering of course, but they're very good at it by now and have huge resources.).

Some time round 2025-2030 things may well start breaking down.

 

[tviceman]

Pascal refresh with lower prices in the meantime. Nothing wrong with that.

 

[Sweepr]

Main point being Volta has been pushed back to 2018

Consumer Volta has been 2018 for quite some time now, but maybe you didn't keep up with the rumors. Only GV100 was supposed to launch in 2017, even Xavier is 2018.

If they cant get the professional version of Volta out the door before in 2018, prepare to wait a long time for the consumer version. I would guesstimate mid/late 2018, or even as bad as 2019.

The source implies GV100 has been pushed back to next year based off the fact that Oak Ridge is now 2018. Even if that's the case, may I remind you that GP104 was available earlier than GP100 last year - late May vs August (first DGX-1 delivery). Saying consumer Volta is now 2019 is FUD.

By the way, Ishimura from Baidu (known NVIDIA leaker) says GV104 should deliver 13.5 TFLOPs, which is >50% more than GTX 1080. He also says it's a significant architecture change, and GV102 and GV110 are in the works.

 

[tviceman]

By the way, Ishimura from Baidu (known NVIDIA leaker) says GV104 should deliver 13.5 TFLOPs, which is 50% more than GTX 1080. He also says it's a significant architecture change, and GV102 and GV110 are in the works.

That is a fairly easy (and probably accurate) prediction. Didn't Maxwell and Pascal both increase their GFLOPs over their predecessors by approximately the same percentage?

 

[jpiniero]

By the way, Ishimura from Baidu (known NVIDIA leaker) says GV104 should deliver 13.5 TFLOPs, which is >50% more than GTX 1080. He also says it's a significant architecture change, and GV102 and GV110 are in the works.

Sounds about right; although at the same time the 1080 Ti @ 2 Ghz will be able to deliver that much sooner, albeit for 300-400 bucks more. Guess we will have to see how good the clocks end up being on Volta and how the architecture improvements play out.

 

[guskline]

These companies are working at break neck speed to reduce the size of the gpus. Good grief we have been on 28nm how long and some seem to be critical because Nvidia/AMD or RTG is having difficulty going to a smaller size?

I for one really appreciate the effort of both companies and marvel at the technology and reduction in size of both the gpus and cpus all the while getting performance boasts along the way.

PS. 2018 is only next year.

 

[zinfamous]

Seems like it
Not sure how its going over at GlobalFoundries/Samsung tho

IIRC, GloFo is skipping 10nm altogether, aren't they? Just longer duration of 14nm silicon with a possibly slight advantage in release schedule for 7nm compared to Intel and TSMC? Or was that just some crazy thing that I read somewhere?

 

[Qwertilot]

Sounds about right; although at the same time the 1080 Ti @ 2 Ghz will be able to deliver that much sooner, albeit for 300-400 bucks more. Guess we will have to see how good the clocks end up being on Volta and how the architecture improvements play out.

Well, don't forget one big thing Volta is 99% likely to get - with 16nm being so much more mature by 2018 there's every chance than GV104 won't be much smaller than GP102, and GV102 up to the same sort of size that GP100 is but with a pure gaming focus. That'll get them a fair bit of the performance uplift for 'free'.

The architecture stuff will likely come into play as much in terms of taming its power draw as anything else.

 

[jpiniero]

with 16nm being so much more mature by 2018 there's every chance than GV104 won't be much smaller than GP102

Except they are doing it on this so called 12 nm. Obviously die size would depend on how different Volta is but I wouldn't expect GV104 to be anywhere near GP102's size.

I think it does confirm that the 2017 GPUs will be Pascal Refresh though; simply faster clocks and memory and that's it.

 

[Qwertilot]

I doubt if they're going to get a smaller die from this process refresh, more energy efficiency etc is more likely.

Don't forget we're also getting GP102 at sane prices this year. The biggest conformation about 2017 is that they still haven't done that yet

 

[jpiniero]

Don't forget we're also getting GP102 at sane prices this year.

Don't bet on it...

 

[Qwertilot]

Saner seems fairly guaranteed Sane is likely to be a matter of perspective.

 

[jpiniero]

Saner seems fairly guaranteed Sane is likely to be a matter of perspective.

Well, the rumor for the (1080 Ti?) was $899, and that chip would be cut down further from the Titan XP. Maybe nVidia will do a new Titan XP that's full or maybe slightly less cut?

 

[RussianSensation]

Well, the rumor for the (1080 Ti?) was $899, and that chip would be cut down further from the Titan XP. Maybe nVidia will do a new Titan XP that's full or maybe slightly less cut?

As far as I recall, no reputable site ever claimed that the price of 1080Ti would be $899. People expect that due to lack of competition but I don't recall a reputable leak detailing this. Even as of now, we have no idea where 1080Ti will land in terms of price other than it will be between $600 and $1000.

Wrt to Volta, it was always on NV's consumer roadmap for 2018. Sounds like 12nm may be mostly marketing as the underlying transistor specs will most likely be based on slightly improved 16nm.

In 2015-2016, certain members on this forum continued to propagate the idea that Volta was launching in 2017, but this always contradicted NV's own roadmap. There is no official consumer delay for Volta since 2018 launch can mean anytime between Jan 1-Dec 31, 2018.

What's missing from this big picture are true next generation PC games. There are literally NONE on the PC. All this extra GPU power means very little when PCs have little to show for it. The trailers for The Last of Us 2 and Uncharted 4: The Lost Legacy have better graphics than 99% of all PC games. At the same time, there is still nothing better looking than SW:BF/BF1. The current software landscape on the PC is the biggest snoozefest in decades. Besides EA, almost every AAA game us made for consoles in mind. EA is the only major AAA studio who admitted to making AAA games on the PC and scaling them down to consoles [no wonder SW:BF/BF1 are easily the best looking and optimized PC games in all of 2015/2016]. Ironically. These games run maxed out on $200 RX480/1060 cards -- Volta not needed.

My excitement meter for next gen PC GPUs is very low. It seems new AAA games run worse and worse on modern CPU/GPU hardware, but graphics are going nowhere. To this day, we are yet to see a game that shows serious benefits from 6-8 core CPUs.

Extreme/Ultra settings in games have become a complete joke where performance hit is 25-50% and literally requires still screenshots and 200% zoom to notice the difference. MSAA performance hit is at the worst optimized level in decades.

Therefore, whether or not Volta makes it to early 2018 or late 2018 is almost irrelevant at this point. It seems until we get PS5/XB2, it's just going to be poorly optimized "demanding" AAA console ports one after the other.

The main catalysts that keeps next gen GPUs relevant at this point are high resolution gaming (3440x1440, 4K, multi-monitor) and high refresh rate gaming (1080/1440/4K 240Hz). The vast majority of Steam users are on "1080p 60Hz peasant" and below resolutions.

At this point even 2X faster than Titan XP is meh (i.e., it's exciting from a hardware perspective but not from a point-of-view that PC software will actually efficiently harness that power) since AAA developers are still going to be targetting 80M+ of PS4/XB1 users. Years have passed since UE4/Kite/Inflitrator demos and not 1 AAA PC game based on UE4 looks anything remotely close to those!!!

The question is will Star Citizen come out before GV104?

 

[tamz_msc]

As far as I recall, no reputable site ever claimed that the price of 1080Ti would be $899. People expect that due to lack of competition but I don't recall a reputable leak detailing this. Even as of now, we have no idea where 1080Ti will land in terms of price other than it will be between $600 and $1000.

Wrt to Volta, it was always on NV's consumer roadmap for 2018. Sounds like 12nm may be mostly marketing as the underlying transistor specs will most likely be based on slightly improved 16nm.

In 2015-2016, certain members on this forum continued to propagate the idea that Volta was launching in 2017, but this always contradicted NV's own roadmap. There is no official consumer delay for Volta since 2018 launch can mean anytime between Jan 1-Dec 31, 2018.

What's missing from this big picture are true next generation PC games. There are literally NONE on the PC. All this extra GPU power means very little when PCs have little to show for it. The trailers for The Last of Us 2 and Uncharted 4: The Lost Legacy have better graphics than 99% of all PC games. At the same time, there is still nothing better looking than SW:BF/BF1. The current software landscape on the PC is the biggest snoozefest in decades. Besides EA, almost every AAA game us made for consoles in mind. EA is the only major AAA studio who admitted to making AAA games on the PC and scaling them down to consoles [no wonder SW:BF/BF1 are easily the best looking and optimized PC games in all of 2015/2016]. Ironically. These games run maxed out on $200 RX480/1060 cards -- Volta not needed.

My excitement meter for next gen PC GPUs is very low. It seems new AAA games run worse and worse on modern CPU/GPU hardware, but graphics are going nowhere. To this day, we are yet to see a game that shows serious benefits from 6-8 core CPUs.

Extreme/Ultra settings in games have become a complete joke where performance hit is 25-50% and literally requires still screenshots and 200% zoom to notice the difference. MSAA performance hit is at the worst optimized level in decades.

Therefore, whether or not Volta makes it to early 2018 or late 2018 is almost irrelevant at this point. It seems until we get PS5/XB2, it's just going to be poorly optimized "demanding" AAA console ports one after the other.

The main catalysts that keeps next gen GPUs relevant at this point are high resolution gaming (3440x1440, 4K, multi-monitor) and high refresh rate gaming (1080/1440/4K 240Hz). The vast majority of Steam users are on "1080p 60Hz peasant" and below resolutions.

At this point even 2X faster than Titan XP is meh (i.e., it's exciting from a hardware perspective but not from a point-of-view that PC software will actually efficiently harness that power) since AAA developers are still going to be targetting 80M+ of PS4/XB1 users. Years have passed since UE4/Kite/Inflitrator demos and not 1 AAA PC game based on UE4 looks anything remotely close to those!!!

The question is will Star Citizen come out before GV104?

As much I would hate to admit it, this is true. There have been no games in the past few years that make you go wow over the graphics alone. We get our hopes up only to be served with a downgrade from what is shown to what is in there in the final product. On the other hand, it is impressive what devs like Naughty Dog can do with limited hardware.

Ten years ago, Crysis made you *WANT* that 8800 GTX. There hasn't been anything like that for a long time.

 

[Aristotelian]

My excitement meter for next gen PC GPUs is very low. It seems new AAA games run worse and worse on modern CPU/GPU hardware, but graphics are going nowhere. To this day, we are yet to see a game that shows serious benefits from 6-8 core CPUs.

Extreme/Ultra settings in games have become a complete joke where performance hit is 25-50% and literally requires still screenshots and 200% zoom to notice the difference. MSAA performance hit is at the worst optimized level in decades.

Therefore, whether or not Volta makes it to early 2018 or late 2018 is almost irrelevant at this point. It seems until we get PS5/XB2, it's just going to be poorly optimized "demanding" AAA console ports one after the other.

The main catalysts that keeps next gen GPUs relevant at this point are high resolution gaming (3440x1440, 4K, multi-monitor) and high refresh rate gaming (1080/1440/4K 240Hz). The vast majority of Steam users are on "1080p 60Hz peasant" and below resolutions.

At this point even 2X faster than Titan XP is meh (i.e., it's exciting from a hardware perspective but not from a point-of-view that PC software will actually efficiently harness that power) since AAA developers are still going to be targetting 80M+ of PS4/XB1 users. Years have passed since UE4/Kite/Inflitrator demos and not 1 AAA PC game based on UE4 looks anything remotely close to those!!!

The question is will Star Citizen come out before GV104?

As usual, you raise so many good points in your posts like this one. I bought a 1440p monitor in 2010-2011 and have been using it since. Sometime last August I was ready to drop almost 10K on a new build (including the Dell OLED 4k monitor that appears to be cancelled) with custom watercooling out the wazoo. Part of that was of course the 'build excitement', and the other part - the performance. But with the monitors I want not even being released yet; the CPUs not producing a vast improvement over my overclocked 2600K from 2011, and more - well, my wife rightly convinced me to put that 10K towards something else.

The thing is, I have defended the price doesn't matter mantra for a while but the point I always tried to convey is that it needs to serve a function as a performance part. I started to play Overwatch recently on my recently minor upgraded PC (overclocked 2600K, 1070 ROG Strix OC) and I'm never wanting for FPS (even though my monitor is 60Hz and, well, when I was younger I was a Quake 3 Arena champion and my deaths or inability to carry my team in Overwatch has nothing to do with my FPS...).

A buyer like me wants to pay cash for a paradigm shift or something close to it; not huge sums for what appear to be incremental improvements in practical terms.

Are we going to have to wait for VR to really take off for the huge performance boosts to come? Or PS5? Or PS6?

 

[[DHT]Osiris]

As far as I recall, no reputable site ever claimed that the price of 1080Ti would be $899. People expect that due to lack of competition but I don't recall a reputable leak detailing this. Even as of now, we have no idea where 1080Ti will land in terms of price other than it will be between $600 and $1000.

Wrt to Volta, it was always on NV's consumer roadmap for 2018. Sounds like 12nm may be mostly marketing as the underlying transistor specs will most likely be based on slightly improved 16nm.

In 2015-2016, certain members on this forum continued to propagate the idea that Volta was launching in 2017, but this always contradicted NV's own roadmap. There is no official consumer delay for Volta since 2018 launch can mean anytime between Jan 1-Dec 31, 2018.

What's missing from this big picture are true next generation PC games. There are literally NONE on the PC. All this extra GPU power means very little when PCs have little to show for it. The trailers for The Last of Us 2 and Uncharted 4: The Lost Legacy have better graphics than 99% of all PC games. At the same time, there is still nothing better looking than SW:BF/BF1. The current software landscape on the PC is the biggest snoozefest in decades. Besides EA, almost every AAA game us made for consoles in mind. EA is the only major AAA studio who admitted to making AAA games on the PC and scaling them down to consoles [no wonder SW:BF/BF1 are easily the best looking and optimized PC games in all of 2015/2016]. Ironically. These games run maxed out on $200 RX480/1060 cards -- Volta not needed.

My excitement meter for next gen PC GPUs is very low. It seems new AAA games run worse and worse on modern CPU/GPU hardware, but graphics are going nowhere. To this day, we are yet to see a game that shows serious benefits from 6-8 core CPUs.

Extreme/Ultra settings in games have become a complete joke where performance hit is 25-50% and literally requires still screenshots and 200% zoom to notice the difference. MSAA performance hit is at the worst optimized level in decades.

Therefore, whether or not Volta makes it to early 2018 or late 2018 is almost irrelevant at this point. It seems until we get PS5/XB2, it's just going to be poorly optimized "demanding" AAA console ports one after the other.

The main catalysts that keeps next gen GPUs relevant at this point are high resolution gaming (3440x1440, 4K, multi-monitor) and high refresh rate gaming (1080/1440/4K 240Hz). The vast majority of Steam users are on "1080p 60Hz peasant" and below resolutions.

At this point even 2X faster than Titan XP is meh (i.e., it's exciting from a hardware perspective but not from a point-of-view that PC software will actually efficiently harness that power) since AAA developers are still going to be targetting 80M+ of PS4/XB1 users. Years have passed since UE4/Kite/Inflitrator demos and not 1 AAA PC game based on UE4 looks anything remotely close to those!!!

The question is will Star Citizen come out before GV104?

This, right now the real reason to upgrade VG is for higher res. 1440p @120hz+ in no-kidding-everything requires a 1080GTX, and unless you're running $1200+ in GPUs you aren't getting pushing over 60hz in 4k, much less 120hz (which is on the very-near-horizon for PC gaming whales out there). That *could* be the push but if the games are still graphically snoozy, nobody's gonna really care about 90fps 4k, if they can spend a quarter the price and get 1080p/1440p @120fps with AA jacked up.