Nvidia drops Samsung and uses TSMC for Pascal
Click here to post a comment for Nvidia drops Samsung and uses TSMC for Pascal on our message forum
Kaarme
Q1? Hmm... I thought the new GPUs from a smaller process node would start to appear towards the end of the next year. If AMD fails to catch up now, like it did when Nvidia released 900 series, it's surely all over for them. Although I don't know why they would considering they already have experience with HBM, unlike Nvidia. So, logic dictates AMD will also release starting from the beginning of the year. I'm beginning to wonder if upgrading a video card now is sheer folly.
Ryu5uzaku
Well if you always wait for the next gpu you will wait forever. But I am planning on getting either amd or nvidia on smaller node next year. And then going for amd zen or kaby lake for my next build. Will be interesting if AMD goes for glofo/samsung or TSMC.
HonoredShadow
"Nvidia's new GPU is highly likely to mark a milestone in the next-gen graphic market."
What does that mean?
mR Yellow
Q1 might be when the first chips are produced but the cards will only come out toward Q3/Q4.
southamptonfc
It says a "release in Q1 2016" but I would think it'll be in the workstation/Tesla or whatever it's called form. Not Geforce until later in the year and even then it'll be the cut down chip.
Matt26LFC
Don't see why we can't have a q1 release, the 970 & 980 launched approx 1 year ago. So if we get q1 launch that's like 15 to 18 months after the 970 & 980.
Really looking forward to these cards, my Tahiti's are getting old lol
Kaarme
I see. That would certainly match with the earlier estimations.
Noisiv
First chips produced few months ago. Target release Q1 2016.
https://forum.beyond3d.com/posts/1848329/
Nvidia is on ~12 months release cycle. So we'll definitely have something in Q1/Q2.
Titan X was released in March, 2015, so I am guessing Titan (and pro versions) in Q1
and consumer grade Pascal in Q2.
fantaskarsef
Probably marketing catch phrase.
Demrock
No way we will see a Geforce Pascal in Q1 2016....
Willing to bet my left hand for it 🙂
Denial
It's usually about 9-10 months from tape out till launch (or at least it was with Fermi/Kepler and Maxwell). According to all the rumors it was taped out back in June so launch should be expected around March 2016, which fits with Nvidia's annual release cycle. It's definitely possible it will get pushed to April, which I guess is technically out of Q1 but yeah, it's most definitely coming around March/April.
Fox2232
Likely one of Huang's marketing spins along lines of 4 x 6 = 10 for pascal in terms of performance.
(iirc it was 4x memory bandwidth, 6x performance of mixed precision compute, and that should have resulted in 10 times higher total performance...)
Denial
6x memory 4x mixed precision.
Listen man, don't be mad Huang best at math and marketing. His leather jacket knows no bounds.
Andrew LB
This isn't AMD we're talking about. nVidia cards are usually very close to marketing claims. Especially when you realize that "Up to..." means "in very select circumstances, usually it's not quite that fast".
Denial
Idk man, Nvidia's own slide has a disclaimer at the bottom that said "VERY ROUGH ESTIMATE" and while at the conference he definitely said he was talking about neural networks, half the "news" sites that reported it used misleading as **** titles like "PASCAL 10X FASTER THAN MAXWELL" .. yeah at one very specific thing that no one here is going to use. Not that the second part is Nvidia's fault -- but yeah. Both sides have retarded marketing.
waltc3
Your first couple of sentences don't make much sense...;) But by the last couple of sentences you seem to regain your logic footing...Yes, it's nVidia very much in catch-up mode at the moment...not AMD. AMD's first-gen experience with HBM will be invaluable with its second-gen HBM products--nVidia gets 0 points for skipping the first generation--probably more like a -10 points, I'd say.
More importantly for nVidia is catching up to where AMD currently is in hardware D3d12 support...This will become far more evident as D3d12 benchmarks and games appear...At that point, Brian Burke will have some explaining to do--but when does he not?...;) Ah, competition can be the spice of life!
Kaarme
Yeah, I wasn't actually thinking about their capabilities as much as the schedules. It has been a while since the Maxwells first appeared, whereas AMD's Fiji only appeared this summer, and we are still getting new cards like Nano and possibly 380X. I was just thinking it would be strange if they were rendered obsolete half a year from now.
But then again, if you consider the current Fury cards only curiosities and proofs of concept, then it wouldn't matter, as the 300 series itself is more or less old stuff in new clothes. This extra 28nm generation really was a pity.
Solfaur
Even though it's still a long wait ahead, I'm looking forward to Pascal, my 970s will have to do til then (even though I had the itch to upgrade for a while lol, but the cost-performance ratio is just too bad atm).
gUNN1993
I don't really think the catch up is going to be particularly relevant, to my understanding HBM defeats a bottleneck along with a few other little things ... but we haven't hit that bottleneck yet. So unless the bottleneck is hit in the next line of new manufacture nodes (or whatever) AMD have the upper hand in a technology that only has minimal use.
Denial
The biggest gain from HBM was the power savings. If you took a 980Ti and put HBM on it, it would be a 225w card. You could then potentially clock it higher given the power headroom. The second biggest gain would probably be form factor honestly. As you said memory bandwidth isn't really that much of a bottleneck at the moment -- unless you're running 4K with AA.
HBM2 is a JEDEC standard now, so any memory company could develop HBM2 modules. AMD is paired with SK Hynix so they have an advantage there. I don't know who Nvidia is pairing with for HBM. There is no exclusivity deal despite what some people say. The advantage is in the fact that AMD/SK essentially wrote the rulebook on HBM -- every other company including Nvidia still needs to figure it out and build it. Building it seems to be the hardest part. Who would have known that growing 10,000+ nano crystals per stack would be difficult? Then you need to fuse it to an interposer.
The other advantage AMD has is that it may possibly produce it's next generation GPU's on Samsung/GF's 14nm process. Samsung's 14nm does have a slight density advantage over TSMC's 16FF+.
