It's not going to be 17B transistors. Not the consumer version anyway. Maybe GP110 (Full Pascal) will be in 2017, but not the initial release. 16nmFF+ costs more per transistor than 28nm does currently. So any chip on it, even if it's the same number of transistors, will be more expensive. Couple that with HBM2 yield and a chip that large just isn't manufacturable. I predict a 1080GTX (GP104) only being about 25% faster than 980Ti and 1070GTX (Cut Down GD104) being about 5-10% faster than a 980Ti. They will launch at $550/450 respectively. About a year later they will launch 1180GTX (GP110) that will add an additional 25-30% on top of the 1080GTX and refresh the rest of the line up throughout 2017 Obviously some Titan card will also launch along both of these and be overpriced but I doubt even that will have 17B transistors, at least the initial one. I expect the GP104 chips to be announced around March for May/June release.
Also what about the GDDR5X information we got a while back? Are they doing anything with that or is that more unknown information just like HBM2? Because both Rumors cant be true especially with the higher end cards.
love the bandwidth is has , and 16 gbs?? i thinking it might be more like 8gb but time will tell, along with how much
I still don't think the initial GP104 based Titan will have that many transistors. With current prices, manufacturing a 17B transistor chip on 16nmFF would cost $272. That's without the interposer and HBM, both of which add significant costs. By comparison a 8B transistor Titan X on 28nm FF only costs $112 per chip -- and they don't need to worry about interposer and HBM. A theoretical 17B transistor Titan (Pascal) would probably be priced at like $3000 knowing Nvidia. I just don't see them doing that. There is no reason for it -- Volta isn't launching till 2018 so they're going to need a refresh in 2017. They will hold off on maxing 600mm2 on 16nmFF until then.
There have been no G*104 or 204 titans to date. Big kepler gk110, big maxwell gm200 had titans. I do not see a $1000 titan based on performance gp104 performance Pascal.
Ya I know, I'm just speaking theoretically. I don't think they could build a chip that big, even if they marketed it as premium.
I'm still wondering on how HBM will behave... if the bandwidth is enough, or if we actually would want to overclock that VRAM (and if it would be possible).
Then Thanks for joining forums and pushing that sense into him. Spoiler j/k He is right, and those who see themselves getting 17B transistor 16/14nm GPU next year as upgrade from gtx 970 are more than naive. But I would like nVidia to make such GPU from the bat. Then we would see a lot of sad faces looking at "empty" wallets prepared for upgrade. Because That GPU alone would cost more than their entire current system did.