Nvidia Post-Volta (Ampere?) Rumor and Speculation Thread

Status
Not open for further replies.
G

Geeforcer

Harmlessly Evil
Veteran
Supposedly, next set of Nvidia cards will be based on Ampere microarchitecture:
https://m.heise.de/newsticker/meldu...eftiger-Gewinnsprung-dank-Gaming-3887633.html

I have to say, IF true, this will actually be the natural next step in the evolution of Gaming-Compute divergence we have been seeing, each subsequent step requiring substantially more resources to execute:

1) Same chip for both purposes (Tesla)
2) Different chips within the family for different purposes with limited crossover in Titan (Kepler)
3) Dedicated chips within a family with no cross-use (Pascal)
4) Entire uarch dedicated to each market

Considering that Volta’s tensor cores were always of very limited utility in consumer market, this is would a perfect inflection point.
 
Last edited:
I don't know what this architecture will bring, but I'm willing to go out on a limb and say that if NVIDIA has just released Volta, and is about to reveal Ampère, then the following generation will probably be called Watt.

Let's hope for NVIDIA's sake that they'll keep making the same kind of progress in power-efficiency, lest they expose themselves to easy jokes.
 
el etro said:
TSMC 7nm HVM is H1 2018. In time for Apple i guess.
Click to expand...

TSMC didn't create a custom Nvidia 12nm FFN (as in FinFet Nvidia) process so Nvidia could proceed to transition to 7nm as soon as humanly possible.

I expect Nvidia will create at least a full lineup (pro + consumer cards) on 12FFN in 2018 before they move to 7nm.

NVIDIA is genuinely building the biggest GPU they can get away with: 21.1 billion transistors, at a massive 815mm2, built on TSMC’s still green 12nm “FFN” process (the ‘n’ stands for NVIDIA; it’s a customized higher perf version of 12nm for NVIDIA).
Click to expand...

https://www.anandtech.com/show/1136...v100-gpu-and-tesla-v100-accelerator-announced
 
Dunno if Volta+12FFN alone will give tangible benefits enough to get a new Geforce line. Nvidia could give up all FP64 and Tensors for the consumer cards like Maxwell, is a nice strategy.
 
el etro said:
Dunno if Volta+12FFN alone will give tangible benefits enough to get a new Geforce line. Nvidia could give up all FP64 and Tensors for the consumer cards like Maxwell, is a nice strategy.
Click to expand...
Oh heck it would, they could use the 610mm die (as a GV102/Titan without DP) with similar power demand to the GP102, massive increase in cuda cores and SM/polymorph engines just with that.
Remember there is also a fundamental redesign of the compiler and also how the cores-operations can be used in number of cycles.
 
el etro said:
Dunno if Volta+12FFN alone will give tangible benefits enough to get a new Geforce line. Nvidia could give up all FP64 and Tensors for the consumer cards like Maxwell, is a nice strategy.
Click to expand...

GP104 is only 314mm2 and GP102 is only 471mm2.

Worst case, Nvidia could continue using 16nm and just increase die size to extract more performance. Plenty of headroom left.

So with 12nm's modest density improvements and Volta's architectural improvements, it's even easier to get a generational bump.

Also, the FP64 and tensor units are effectively gone for the consumer Volta products (assuming there are any). That's basically a guarantee at this point.
 
I’m not sure I would read anything into the public road map. They haven’t updated it a while. In the past, they would share the architecture code names for the next several iteration, but there’s nothing on the charts post Volta.

2 years after Pascal, I expect more than just a refresh.
 
http://m.3dcenter.org/news/nvidia-u...bringt-im-q22018-gleich-die-ampere-generation

google translate:

nVidia skips "Volta" in the gaming sector and will bring the "Ampere" generation in Q2 / 2018


Long story short, due to these circumstances, we expect the ampere generation already in the 10nm production of TSMC. After this (for nVidia) long waiting period between two graphics card generation as well as due to the earlier in the spring / summer of 2018 predicted availability of this manufacturing process would really surprise anything else.

Above all, the use of 12nm production in the spring / summer of 2018 would hardly make sense, as long as the 10nm production would soon be ready to go - nVidia (as explained above) can not achieve as much performance with 12nm production as you really wanted to bring. With the ampere generation in 10nm production, however, the generally usual performance boost of almost twice is possible
 
Yeah, unlikely this is a real new generation. My bet is on Pascal shrink to 12nm. The naming is probably a diversion. Why? Because it saves money, and I bet they think they don't need to put any more effort in the gaming sector. And they are probably right, until they get caught cold by AMD (Navi) or Intel (whatever they are cooking).
 
BoMbY said:
Yeah, unlikely this is a real new generation. My bet is on Pascal shrink to 12nm. The naming is probably a diversion. Why? Because it saves money, and I bet they think they don't need to put any more effort in the gaming sector. And they are probably right, until they get caught cold by AMD (Navi) or Intel (whatever they are cooking).
Click to expand...

What's the point of Pascal 12nm or Volta 12nm? Too large die, too expensive. They could just dump the price of a GP104 to $300 and still make a killing.

The Ampere 10nm story makes sense.
 
Status
Not open for further replies.

Similar threads

A
NVIDIA Morgan Stanley Webcast: Sony, G80 and 80/65nm
2
Replies
25
Views
13K
Nelsieus
N
Back
Top