NVIDIA Maxwell GM200 Titan II/GTX 980 Ti Slipping To 2016

H

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
According to this report, production of NVIDIA's Titan II and GTX 980 Ti could be slipping all the way to 2016.

TSMC had originally scheduled 16nm FinFET mass production for Q1 2015. Mainly to satisfy Apple’s A9 chip demands. However TSMC fell short of meeting the original schedule and ended up moving volume 16nm production all the way to late Q2 early Q3. The company now seems to be targeting June/July of 2015 for 16nm FinFET. And a quarter or two later for 16nm FinFET+.
Click to expand...
 
After the Maxwell 20nm fiasco I could really care less. It's all about the *REAL* Maxwell aka Pascal aka formally called Volta.


[EDIT]
Also why would it be the Titan 2 (already brand confusion) or the 980 Ti at that point? You're talking over a year from the GTX 980's release on a new node, that would be Nvidia's next-gen name (probably not GTX 1xxx). They have more than enough die-space and power limits to create a GTX 780 Ti equivalent on 28nm thanks to Maxwell's architecture, so why wouldn't they? That's a little silly and would be strange if they didn't do. Funny thing is this TSMC delay has been known for a couple months now.
 
Last edited:
I doubt NV will wait a full year to release another high end card especially if AMD launches a round of top tier cards in 2015 that beat the 970/980.

On the other hand, NV has the money and current market share to lose a year to AMD. Especially if their R&D on the 16nm is worth it.

It is not like the market share will flip in one year. I think it will mainly depend what AMD can do with their next gen due out in 2015.
 
Big maxwell will most likely be 28nm. Take a 980 and stuff it with more toys. But AMD's HBM though...
 
I wouldn't buy that bullshit for a minute. Its just posturing by Nvidia. They are not going to lose that HUGE amount of money sticking to developing 16nm only for Apple. 20nm has been going strong for a while now at TSMC. Thats a huge improvement alone over 28nm. No way is Nvidia just going to sit by idle on 28nm knowing the power Maxwell has and what that means at 20nm. A 980Ti with 2500+ cores and a dual 990GTX (2x2500+ cores) on 20nm will be produced. Now they will hold out with those on the back burner till AMD produces a 300X series card in the first quarter of next year, I can't believe anyone would think Nvidia would just sit idle for a whole year.

http://www.tsmc.com/english/dedicatedFoundry/technology/20nm.htm
 
Liger88 said:
After the Maxwell 20nm fiasco I could really care less. It's all about the *REAL* Maxwell aka Pascal aka formally called Volta.


[EDIT]
Also why would it be the Titan 2 (already brand confusion) or the 980 Ti at that point? You're talking over a year from the GTX 980's release on a new node, that would be Nvidia's next-gen name (probably not GTX 1xxx). They have more than enough die-space and power limits to create a GTX 780 Ti equivalent on 28nm thanks to Maxwell's architecture, so why wouldn't they? That's a little silly and would be strange if they didn't do. Funny thing is this TSMC delay has been known for a couple months now.
Click to expand...

I believe the point is GM200 was scheduled to be on 20nm, which Nvidia scrapped. It was then supposed to go on 16nm, which is now delayed. So either the 980ti and Titan will be delayed to 2016, or they will not be based on GM200.

That's my understanding anyway.
 
seanclayton said:
. But AMD's HBM though...
Click to expand...

If 16nmFF isn't happening till 2016, then HBM isn't happening till then either since that is the process from which it is based.

Honestly don't believe both manufactures are going to by pass 20nm,which in itself is a big die drop, to wait out a year till 16nm FF is ready. You have smaller process ready to go, can make money, and just pass on it?????? Not buying that one bit.

At the very least we will see a GTX 950/960 on 20nm much like AMD use to do with their low end cards trying out a new process.
 
Cr4ckm0nk3y said:
I doubt NV will wait a full year to release another high end card especially if AMD launches a round of top tier cards in 2015 that beat the 970/980.

On the other hand, NV has the money and current market share to lose a year to AMD. Especially if their R&D on the 16nm is worth it.

It is not like the market share will flip in one year. I think it will mainly depend what AMD can do with their next gen due out in 2015.
Click to expand...

If both are using the 16nm process then all next gen cards are going to be delayed. Can't do much about it if TSMC can't produce the cards...
 
GM200 tapped out 2 months after GM204 both are 28nm. Expect big maxwell 1st qtr 2015.:D
 
xBanzai89 said:
If both are using the 16nm process then all next gen cards are going to be delayed. Can't do much about it if TSMC can't produce the cards...
Click to expand...

Amds 300 series will be on 20nm not 16nm. I think nvidia planned on the 980 being 20nm but with the delay they changed it to 28nm.

If they plan on using 16nm I don't see them doing a full rework of their gpus for 20nm just to get something out one quarter sooner. I'm not sure if the cost ratio works out for them.
 
spugm1r3 said:
I believe the point is GM200 was scheduled to be on 20nm, which Nvidia scrapped. It was then supposed to go on 16nm, which is now delayed. So either the 980ti and Titan will be delayed to 2016, or they will not be based on GM200.

That's my understanding anyway.
Click to expand...

As others have mentioned GM200 will still be 28nm.

Maxwell was supposed to be 20nm, but TSMC's yields were too horrendous for mass production (the other story I heard was they found the 20nm process offered very little tangible benefit for high performance chips, so nVidia just said fuck it), so nVidia had to turn their team around and re-engineer Maxwell on 28nm, hence the delay in 980/970 launch.
 
Remember that GK110 was primarily for Nvidia's big moneymakers: the business line of Quadros and Teslas. Chips that fail validation end up as consumer parts with extraneous components (e.g. ECC) fused off. Chances are that any sort of 'big Maxwell' will end up in a Quadro or Tesla before coming to a consumer card, particularly on the fairly mature 28nm process where there is likely to be a pretty decent yield.
 
I wasn't expecting 16nm anytime soon anyways. This report is kinda confusing since GM200 was going to be 20nm IIRC, and a 980ti will probably never exist, there is probably going to be a 20nm line ran next year and a 16nm run in 2016 or 2017.
 
Last edited:
Isn't Pascal supposed to hit in 2016? What would be the point of releasing GM200 that late?
 
Do you guys have any idea how expensive it is to pattern at the 14 nm node? Or even the 16 nm or 20 nm half nodes? It's ridiculous. Current practice is still 193 nm light with multiple patterning. They literally pattern the entire wafer multiple times slightly offset resulting in a multiplication of the number of steps resulting in a multiplication of the cost. Intel's latest fabs literally cost more then the GNP of whole countries to build. Since nvidia/amd don't own any fabs they have to rent space from GloFo which is obviously pretty limited. Not surprised there are some delays.
 
If it's slipping that far I can't see any reason to put it out there in the first place, better off just going to Pascal. Not necessarily a bad thing provided they 28nm GM204 parts actually came down in price over time (they won't).
 
lasserith said:
Do you guys have any idea how expensive it is to pattern at the 14 nm node? Or even the 16 nm or 20 nm half nodes? It's ridiculous. Current practice is still 193 nm light with multiple patterning. They literally pattern the entire wafer multiple times slightly offset resulting in a multiplication of the number of steps resulting in a multiplication of the cost. Intel's latest fabs literally cost more then the GNP of whole countries to build. Since nvidia/amd don't own any fabs they have to rent space from GloFo which is obviously pretty limited. Not surprised there are some delays.
Click to expand...

I'm no expert, but I do expect each node to last even longer than it's predecessor, as smaller nodes become more technically and financially difficult to produce.

This is why I think the focus will shift towards architecture efficiency, like what nVidia did with Maxwell. We can no longer rely on smaller transistors to improve things for every generation, we'll have to focus on developing ever more efficient architecture.

The current Maxwell have fewer transistors than the biggest Kepler. So I guess nVidia can release the bigger Maxwell on 28nm next year while waiting for 16nm (if they indeed skip 20nm)
 
Hornet said:
I'm no expert, but I do expect each node to last even longer than it's predecessor, as smaller nodes become more technically and financially difficult to produce.

This is why I think the focus will shift towards architecture efficiency, like what nVidia did with Maxwell. We can no longer rely on smaller transistors to improve things for every generation, we'll have to focus on developing ever more efficient architecture.

The current Maxwell have fewer transistors than the biggest Kepler. So I guess nVidia can release the bigger Maxwell on 28nm next year while waiting for 16nm (if they indeed skip 20nm)
Click to expand...

I'd prefer to see the vendors get their hands dirty in working with developers on software efficiency over just throwing out a less capable GM200/390x. The current crop of high end cards are ludicrously fast as it is, developers have just sort of gotten used to having massive overheads to make up for less than optimized code. I mean look at the most recent example in Alien, that game runs at a bajillion FPS on PC but actually struggles to maintain a solid framerate on the consoles. A little elbow grease can do wonders.
 
Serves them console peasants right ;):p

But yeah considering that GM204 with 5.2 billion transistors can edge out GK110 with 7.08 billion transistors, I'm really looking forward to what GM200 with god knows how many transistors can pull off.
 
What really sucks for me at least, I am in the process of building a very high end build with tri-SLI. I am buying the cards last, hopefully the 8gb 980's will be out in 3 weeks or so when im ready for them, but this seriously sucks that in 3-4 months something considerably better is coming out. Waiting 4-5 months with no computer is also bad...
 
Behyadehto said:
What really sucks for me at least, I am in the process of building a very high end build with tri-SLI. I am buying the cards last, hopefully the 8gb 980's will be out in 3 weeks or so when im ready for them, but this seriously sucks that in 3-4 months something considerably better is coming out. Waiting 4-5 months with no computer is also bad...
Click to expand...

Unless you're talking 20nm amd parts (3-4 months), nvidia 16nm parts are more like 7-8 months away.
 
. That would further validate my Tri-sli titans if true.

I've already had them almost two years with nothing remotely worth upgrading them to, which is an eternity in graphics cards. GTX 980 is not much faster (10-20% usually) and handicapped on VRAM. To have that as status quo and have my Titans near the top of the heap for three years total when it's all said and done would be amazing.

Remember the 8800GTX was a beloved longevity champion and even it got wrecked by the GTX 280 to the tune of 40% a mere 1.5 years after it came out
 
Hornet said:
I'm no expert, but I do expect each node to last even longer than it's predecessor, as smaller nodes become more technically and financially difficult to produce.

This is why I think the focus will shift towards architecture efficiency, like what nVidia did with Maxwell. We can no longer rely on smaller transistors to improve things for every generation, we'll have to focus on developing ever more efficient architecture.

The current Maxwell have fewer transistors than the biggest Kepler. So I guess nVidia can release the bigger Maxwell on 28nm next year while waiting for 16nm (if they indeed skip 20nm)
Click to expand...


It's true. Until we come up with a viable silicon replacement we're going to start tapping out transistor wise around 16nm. Right now 10nm is just fantasy as everyone is struggling just to get to 14nm.

However, Intel is kind of a piss poor example for three reasons:
1) They've run out of IPC/architecture improvements in the past 3-4 years resulting in minimal performance gains.
2) They could add more cores/transistors, but they're pushing the fail whale APU instead.
3) Any small gains like that of the CPU world would kill GPU manufacturers. Intel can get away with people doing 3-4 year upgrade cycles. Nvidia/AMD could not. I just see too many negatives.

When the shrinkage freeze happens it's going to suck and it's coming. It'll be interesting to see the day when even enthusiasts are running the same GPU for years because the upgrades are no longer worth it.
 
Steve said:
According to this report, production of NVIDIA's Titan II and GTX 980 Ti could be slipping all the way to 2016.
Click to expand...

The amount of bullshit info is amazing. AMD and Nvidia are at 28 nm currently and the previous many news articles stated that the next die shrink was 20 nm. WTF is 16 doing here? 20 is fine and would advance GPU's signifigantly. Think about it? 28 to 20 so where did 16 come from? First I ever heard of it and it's obviously FUD. :rolleyes:
 
Dildonose said:
The amount of bullshit info is amazing. AMD and Nvidia are at 28 nm currently and the previous many news articles stated that the next die shrink was 20 nm. WTF is 16 doing here? 20 is fine and would advance GPU's signifigantly. Think about it? 28 to 20 so where did 16 come from? First I ever heard of it and it's obviously FUD. :rolleyes:
Click to expand...

The possibility of them skipping 20nm and going straight to 16nm FinFet has been kicking around for months, it's not new information. TSMC have taken so much time to get 20nm functional and at this point it's only really setup for low power mobile parts (all that Apple silicon).
 
Dildonose said:
The amount of bullshit info is amazing. AMD and Nvidia are at 28 nm currently and the previous many news articles stated that the next die shrink was 20 nm. WTF is 16 doing here? 20 is fine and would advance GPU's signifigantly. Think about it? 28 to 20 so where did 16 come from? First I ever heard of it and it's obviously FUD. :rolleyes:
Click to expand...



These rumors sparked during this past summer leading up to Maxwell's release where a lot of speculation was going on. People speculated 20nm might be skipped because it was co-developed at the same time as 16nm and TSMC was making some generalized statements about how the improvements in 20nm only benefited low-power designs.

People took that to mean there are no such thing as 20nm capable high-end parts and that every TSMC partner was skipping 20nm for said devices in favor of 16nm. Perhaps it might have been possible, except for a little unforeseen (or obvious if you're familiar with TSMC) delay with 16nm a couple months ago. The report from TSMC is not new. 16nm would be delayed 6 months do to manufacturing issues.

Needless to say I think it would be in both AMD/Nvidia's best interest to look past TSMC the same way they did UMC back in the day. They are apparently no longer up to the challenge as shrinking nodes get more complex.


tl;dr people made assumptions based on rumors of Maxwell and things spiraled into truth.
 
Behyadehto said:
What really sucks for me at least, I am in the process of building a very high end build with tri-SLI. I am buying the cards last, hopefully the 8gb 980's will be out in 3 weeks or so when im ready for them, but this seriously sucks that in 3-4 months something considerably better is coming out. Waiting 4-5 months with no computer is also bad...
Click to expand...

Are you going to Tri SLI $1000 Titan 2s?..
 
EDIT: Gonna start a new more cheerful thread instead :D
 
Last edited:
Nice way to get people who are sitting on the fence and clinging on to the last bit of patience they have to jump on the 970/980 bandwagon.

I think that plain and simple, you can expect to see Bigger Maxwell flagship whenever AMD shows their hand. No sooner and not much later at all.
 
Dildonose said:
The amount of bullshit info is amazing. AMD and Nvidia are at 28 nm currently and the previous many news articles stated that the next die shrink was 20 nm. WTF is 16 doing here? 20 is fine and would advance GPU's signifigantly. Think about it? 28 to 20 so where did 16 come from? First I ever heard of it and it's obviously FUD. :rolleyes:
Click to expand...

you don't know what you're talking about, but if you want to learn, you can read all the TSMC news from the past year with google
 
Behyadehto said:
What really sucks for me at least, I am in the process of building a very high end build with tri-SLI. I am buying the cards last, hopefully the 8gb 980's will be out in 3 weeks or so when im ready for them, but this seriously sucks that in 3-4 months something considerably better is coming out. Waiting 4-5 months with no computer is also bad...
Click to expand...

Do what I did. Get three gtx 970's. Tri sli for less than 1K. When the 980ti is released the 970's will be swapped out for those. Probably sell them and won't lose to much money on them. If you get 980's instead you'll lose a lot more when the TI shows up.
 
Stoly said:
Or it could come out as early as next month

http://www.eteknix.com/nvidias-next-gen-geforce-gtx-titan-ii-teased-gm200-coming/
Click to expand...


So quickly does the rumor get even more confused by the report on the rumor. Why are these clowns thinking Jen-Hsun Huang's words "next generation" refer to Maxwell not an early Pascal jump given the 20nm skip as already claimed? To think he's referring to GM200, the high-end current gen Maxwell chip, as being a next-gen product is not like him to confuse. Yes I've read the WCCFTech explanation and it contradicts itself. He already mentions going full throttle on Maxwell (i.e. more releases coming *hint* higher-end card) and then states his interest in Pascal (next-gen) given it'll probably not only be on a new node, but contain everything of Volta and Maxwell in one.

Don't even get me started on the rumored specs. With 50% more cores it'll need to have a die-size of 600mm2 or cutting down on a ton of L2 cache to make room for 550mm2 die. Lets not even hit on the fact 16nm has been pushed 6 months to mid-2015 with Fall 2015 as the likely mass product availability Now they're trying to hype up 16nm in a couple months? Wha..........where do they come up with this shit? No sources, just on-the-fly journalism if you even want to consider these guys like Videocardz, Fudzilla, WCCFTech, etc. as journalists.
 
You must log in or register to reply here.
Back
Top