• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Nvidia GTC16 Webcast Live Stream@5pm/9am PDT *Jen looks great in leather.*

Associate
Joined
24 Nov 2010
Posts
2,314
The chip - 4 stacks of HBM2 takes up what 367.97mm2? of the area of the interposer alone. Which wouldn't leave enough space for the rest of the spec if the whole package was 600mm2.

600mm2 is the GPU chip. Just as 597mm2 was Fiji's GPU (not interposer size with HBM and GPU).

IMO they likely have some serious density issues. TSMC 16nmFF is supposed to be quite a bit less dense than Samsung 14nmFF, and they may have had problems re-adding all the stripped hardware features of Kepler / Maxwell. It really shouldn't be that big.

Also, new process chip of that size ... it's going to cost them an absolute fortune, and will probably have gigantic defect rates.
 
Caporegime
Joined
18 Oct 2002
Posts
33,188
So does that mean AMD Vega with HBM2 is fighting big Pascal with HBM2 next year??

No no, GP100 is launching in April this year.... everyone said.

yes, 600mm^2 chips on a new process in the first year was a no, always a no. Yields will be terrible and... lets just say volume production sounds like BS if they are only actually launching boxes in Q1 next year. They will do some extremely expensive, extremely low yield production(which can count as volume) and get samples out there, a few boxes to get certain companies started but volume boxes are clearly only being sold next year.

Keep in mind that iirc GK110 launched as professional products in the November, and as Titan in February.

There is actually nothing physically stopping you making a 600mm^2 gpu on a new process on day one, it's just the financial viability of it. Over time yields improve on a process so day one you'd be looking at like 1% yields and $8000 wafer prices. 6 months later yields might be 15%, a year later yields might be 60%. So the choice for launch is simply where the chip becomes properly viable.

From the dates today I'm reading that Nvidia don't expect yields to produce really financially viable chips till 2017, for professional cards. I wouldn't be surprised to see consumer versions again 4+ months later than that.
 
Soldato
Joined
9 Nov 2009
Posts
24,988
Location
Planet Earth
Details specs of P100 on Nvidia website:

https://devblogs.nvidia.com/parallelforall/inside-pascal/

300W TDP!!!!!!!!!!!!!!!!!!!!!!!!!!!!


It also looks like this timeline leaked early in the year might be true:

http://www.bitsandchips.it/english/52-english-news/6620-rumor-nvidia-pascal-timeline

April 2016 – GTC: Tesla (GP100)
June 2016 – Computex: GTX 1080 (GP104) & GTX 1070 (GP104)
4Q 2016 – GP106 & GP107
4Q 2016 or 1Q 2017 – Titan (GP100)
1Q/2Q 2017 – GP108 (GM108 rebrand?)

During the COMPUTEX NVIDIA will show the first GTX 1080 and GTX 1070 video cards, but the commercialization will begin just during the 3Q 2016 (As we said earlier). Also, NVIDIA will use the first GP100 samples to produce the high profitable Tesla cards (Oak Ridge National Laboratory and Lawrence Livermore National Laboratory will use these cards).
 
Associate
Joined
26 Aug 2010
Posts
554
Could this GPU be the new Titan? As I recall the first Titan was sold to the server market.

Looks like it would easily be more than double the powaa of the titan x :D
 
Associate
Joined
24 Nov 2010
Posts
2,314
It also looks like this timeline leaked early in the year might be true:

http://www.bitsandchips.it/english/52-english-news/6620-rumor-nvidia-pascal-timeline

He was vague about 'showing' / announcing and availability. They haven't shown any Pascal cards today. They haven't talked about availability either. Lots of waffle re: enterprise functionality which they've talked about many times in past 2 years.

It wouldn't surprise me if GP104 slips into late Q4 and GP100 (as opposed to TP100) still isn't on the market a year from now.
 
Back
Top Bottom