-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
Nvidia Volta is 16nm, expected in May 2017
- Thread starter dr_rus
- Start date
Huh... I mean with that line of thinking one can use a 680 and go back to 720p.
Pixel/instruction level preemption is a very significant hardware change.
Well yeah, that's the decision you have to make. Get the performance of the xx80Ti right now, or wait another half a year or more and get that performance for a better price.
That's why I say it's best to buy the Ti cards at launch if you are willing to spend that money, because the longer you wait the more it makes sense (money-wise) to just wait some more for the next gen cards. And we all know that waiting for a new GPU is awful.
I upgraded from a 1080 to a 1080Ti because I wanted to jump over to the Ti-upgrade path and now I'm just gonna ignore the 104 cards from here on out. It hurt a bit paying another 400 after selling my 1080 for the performance boost I got now, but I still think it was worth it in the end.
To me the 1080 Ti made little sense because there was nothing out that actually needed that kind of power. Monitor tech is behind GPUs at the moment and I have no incentive to upgrade my 980 Ti until high refresh rate 4K displays come out. I think I'd rather be on the xx80 upgrades because they usually improve on the previous Ti about 15-20% which at the moment would mean solid 4K @ 60 fps whereas the 1080 Ti is just on the fence for some games. Plus the slightly lower cost.
Fredrik
Member
So I made som extra cash from the sudden nvidia stock rise yesterday. If I have a 980ti which is struggling to perform well in a triple screen setup, should I get a 1080ti or wait for a consumer version of Volta? Waiting is usually boring but how far off is that consumer Volta? I don't want to buy a 1080ti now if Volta is out at the end of the year.
FutureProtagonist
Neo Member
16nm, 12nm, they're all just numbers. The different processes from TSMC, Samsung and Intel are far more than just the number of nanometers. From what I've heard 12nm is basically still 16nm, but with perhaps better power characteristics or voltage scaling. Density seems to be very similar. There exists no clear definition for what qualifies a process to be x nanometers. It seems they're almost more of a branding than something technical.
FutureProtagonist
Neo Member
I can't agree with this. From my experience, the 1080 Ti is a perfect chip for 1440p144 or for 4K60, but 4K144? No way.
neurosisxeno
Member
Most foundries that offer 12nm offer pseudo-12nm. It has a few features that are at 12nm but a majority of the chip is 16nm. The exception--I believe--is Intel who legitimately has 12nm down, with every circuit being entirely at that process size. It's incredibly complicated and I'm sure there are better people than I to explain it, but a lot of the 12/16/20/22nm labels are kind of intertwined.
It is a really good chip for that, but still sometimes falls a bit short of 4K 60 fps in some games. Not a big issue with a G-Sync display for example. But at the same time I am perfectly happy with the performance I get at 1440p with my 980 Ti. I want my next upgrade to be 4K @ 144 Hz, with a GPU capable of solid 60 fps at that resolution. By the time the first consumer Voltas are out there should be at least some displays capable of that.
Nvidia will release GV104 when it's ready. At this point it doesn't matter what Vega is like.
Nvidia's release schedule is not influenced that much by what RTG do. You can't work that way.
Sitting on tech that's ready to go in this market is bad for you. Nvidia gets 50% of its revenue from the games segment and there is plenty of demand for faster GPUs.
It's entirely possible that GV104 will arrive late 2017, in line with the 18-20 months gap between releases we've seen in recent years.
Nvidia's release schedule is not influenced that much by what RTG do. You can't work that way.
Sitting on tech that's ready to go in this market is bad for you. Nvidia gets 50% of its revenue from the games segment and there is plenty of demand for faster GPUs.
It's entirely possible that GV104 will arrive late 2017, in line with the 18-20 months gap between releases we've seen in recent years.
Fredrik
Member
When and where do they usually announce new cards? CES? GDC? Their own events? Different time of the year each time?
You won't know until a few weeks beforehand. They'll likely hold their own press event if there is no suitable show or conference at the time.
In the past we've got some great leaks from the manufacturing companies that give hints a few months out, but those seem to have been closed down hard in the last year. 2016 and 2017 Titan X just came out of nowhere.
neurosisxeno
Member
I think you mean Titan Xp.
But Nvidia is kind of all over the place. Usually it's some trade show 1-2 months before they plan on launching the actual product. I think AMD is actually hurting themselves by dragging out Vega this long. If it's not a price/performance beast by the time it launches people are not going to be happy.
But Nvidia is kind of all over the place. Usually it's some trade show 1-2 months before they plan on launching the actual product. I think AMD is actually hurting themselves by dragging out Vega this long. If it's not a price/performance beast by the time it launches people are not going to be happy.
Titan X 2016 was also a surprise to most, with almost no leaks and revealed two weeks before it went on sale.
Fredrik
Member
Ah okay, seems like you really can't plan ahead much then. :/
The 1080ti hasn't been out for long though so I guess they won't announce the consumer Volta model in a few months at least. Maybe I should just take the plunge after all, hmm
Finaika
Member
Wait for Volta, it will be more powerful than the 1080 Ti.
Frankfurter
Member
Nope. ~15 TF FP32, possibly ~30 TF FP16, but I think Nvidia didn't mention FP16, so it's not quite clear if it actually offers 2x FP32 performance in FP16.
ethomaz
Banned
They only capped FP64 and FP16 perf... plus the memory register were smaller... everything else was the same Arch.
E-Cat
Member
12nm is still part of the 16nm FinFET+ family - don't be fooled by the semantics!
Next stop: 7nm (this will be a real node transition).
ethomaz
Banned
Because this 12nm is basically a 16nm improved. It is not a new node.
slightly off-topic, but is this a complete list of Nvidia Titan graphics cards?
GTX Titan (GK110, Kepler) - February 2013
GTX Titan Black (GK110, Kepler) - February 2014
GTX Titan Z (dual GK110s, Kepler) - March 2014
GTX Titan X (GM200, Maxwell) - March 2015
GTX Titan X (GP102, Pascal) - August 2016
GTX Titan Xp (GP102, Pascal) - April 2017
Also, Telsa V100 is an absolute MONSTER.
I'm going to assume that Titan Volta is going to be again a different chip, a GV102, without most of the fp64 and Tensor cores. and also use GDDR6 instead of HBM2. This would also be used as the basis for GTX xx80 Ti, and a GTX xx80 will use the mid sized GV104 GPU.
Edit: I'm going to make a guess that Nvidia will NOT unveil yet another new architecture beyond Volta at next year's GTC 2018.
Process technology is coming too slow. So rather. GTC 2019 is probably where Nvidia will reveal Volta's successor on one of TSMC's 7nm FinFET process.
GTX Titan (GK110, Kepler) - February 2013
GTX Titan Black (GK110, Kepler) - February 2014
GTX Titan Z (dual GK110s, Kepler) - March 2014
GTX Titan X (GM200, Maxwell) - March 2015
GTX Titan X (GP102, Pascal) - August 2016
GTX Titan Xp (GP102, Pascal) - April 2017
Also, Telsa V100 is an absolute MONSTER.
I'm going to assume that Titan Volta is going to be again a different chip, a GV102, without most of the fp64 and Tensor cores. and also use GDDR6 instead of HBM2. This would also be used as the basis for GTX xx80 Ti, and a GTX xx80 will use the mid sized GV104 GPU.
Edit: I'm going to make a guess that Nvidia will NOT unveil yet another new architecture beyond Volta at next year's GTC 2018.
Process technology is coming too slow. So rather. GTC 2019 is probably where Nvidia will reveal Volta's successor on one of TSMC's 7nm FinFET process.
Dictator93
Member
It is kinda amazing how NV just leap frogged AMD for the 3rd or 2nd time now and are just competing amongst its own products. Bravo for such dominance I guess.
Typically the number relates to the smallest feature they can pattern. However because transistors (and everything else) are more complex than a single element and it will often be other constraints that limit density.
One such (simplistic) example would be if the metalization layers cannot support the same resolution as polysilicon or N or P layers. That obviously means interconnects have to have to conform to spacing constraints thus limiting the overall number of components you can fit in an area.
This is why Intel can say their process is 14nm and Samsung can also say their process is 14nm, both correct but Samsung ICs are nowhere near as dense as Intel ones.
E-Cat
Member
7nm for HPC is very late 2018 at the earliest, most likely 2019.
I can see them revealing the next couple generations of architectures (not an actual chip, mind you). It has been a long time since we haven't known the next NVIDIA architecture to immediately succeed the current one.
TatteredHat
Member
Well I'm glad that nvidia, unlike intel, isn't sitting on their laurels despite their dominant position in the market.
They acquired IBMs microelectronics business, the 7nm process is the first one they're doing with their expertise. So hopefully that means it will be successful.
llien
Member
In the beginning neither was Intel.
E-Cat
Member
I blame that bean counter Krzanich .
eqqsquizitine
Member
Because NVIDIA is aggressively trying to establish itself in a number of new markets. Intel no longer has to try doing that.
Yeah, that's true. Nvidia hasn't hinted on further out GPU architectures in couple of years. I guess for one reason, Pascal was added into the mix between Maxwell and Volta.
Hopefully next year Nvidia will at least name the next two GPU architectures.
My shortlist of scientist names are Einstein, Galileo Galilei, Newton and Hawking.
dr_rus
Member
Source. 1710 means that the GPU was produced on a second week of March 2017.
Nvidia's Mammoth Volta GPU Aims High for AI, HPC
Well, this is a straight up lie.
While there were more improvements in Polaris compared to GCN3 than in Pascal compared to Maxwell it's only because GCN3 was significantly behind Maxwell in features when it was launched. Should I remind you that it didn't even have HDMI 2.0 outputs, for example?
I'm also still puzzled by the "again" wording. When was the last time there were no architecture improvements between different NV architectures? Even Pascal brought a lot of changes compared to Maxwell (already described above) and this is the closest NV got to what you're suggesting. I feel like you're slipping lately.
It's "FFNvidia". Not a joke: "They are building GV100 to the limit of the TSMC process, the die literally cannot get any bigger on the 12nm FFN ("FinFet Nvidia") process."
It's also a further optimization 16FF+ process hence why there's basically no transistor density change between GP100 and GV100.
icecold1983
Member
the again refers to pascal of course
Coulomb_Barrier
Member
As opposed to being completely taken in by Nvidia marketing for this unveil, where silver-tongued exaggeration is par for the course. Whether it be Nvidia or AMD what they say should be treated with some degree of skepticism at least.
Fredrik
Member
I know, waiting is always the smartest thing to do, but... but it's no fun!! ;P
And once once Volta is out I could wait for "Volta Ti", which will be even more powerful, etc. :/
A 1080ti would let me play at high or even ultra at ~6000x1080 (3xscreen) with decent framerates, it would be a nice rig until 4K 144Hz screens come down in price.
D
Deleted member 465307
Unconfirmed Member
Was no new roadmap given? Is Volta still the only future architecture we know about?
E-Cat
Member
Yeah, no new roadmap.
michaelius
Banned
E-Cat
Member
Frankly, it doesn't matter what it should have been called. Look for the area and power reduction vs the preceding node. If those fall somewhere around 50%, you're good.And that 7nm will probably be what should have been called 16nm
DonMigs85
Member
Wonder if they'll continue to iterate on Volta for a while like AMD is with GCN.
dr_rus
Member
Not to the public but the next architecture name was disclosed under NDA afaik so I'd expect it to leak at some point.
Unknown Soldier
Member
Nvidia consistently​ delivers on their hype
Wow this suggests Nvidia is approaching Apple's level of customer priority for TSMC.
dr_rus
Member
What's even more important, there is no hype on Volta in this thread.
E-Cat
Member
Jensen and Morris Chang (TSMC Chairman) are known to be chums. That and TSMC is betting big on AI.
Pretty much, my only problem with Nvidia these days is their pricing, and sometimes drivers. 1080ti was decently priced though so maybe a sign of things to come.
Electret
Member
I bet they're gym buddies, and Morris is tired of spotting Jensen at the bench press.