Using hbm should help. A fury gets 60% more performance/watt than a 390 according to techpowerup.
I hear hbm also helps with die size, since you don't need the on die stuff to drive the signal through the pcb as with gddr.
Vega is a pipe dream. If you look at transistor density and die size the RX480 should be a lot closer to the GTX 1070. Unless Vega is a 500mm sized die with a 350W TDP I dont see how its a competitor to GP102. At best itll be as fast or slightly faster than a GTX 1080.
HBM2 will have significant power savings over GDDR5X (maybe 20ish watts) allowing a full chip to clock at the same speed as the current Titan X. That alongside bandwidth gains will probably give around 10-15% more perf. So thats what Nvidia is holding on reserve.
I'm with some of you guys in spirit. Birthday next week, got the ok to drop stupid money, then stupid car repairs. And I'd rather not dish out car repairs + stupid money, GF might really ask for that horse :|
Any OC'ing reviews yet? I want to see if this thing can hit the sky like 980 Tis did.
EDIT: OP why you no edit 1st post with reviews
It *SHOULD* hit similar clock speeds as GP104 and GP106 so long as its not artificially limited, which means a bigger % overclock given the lower base and boost clock speeds. The cooler is going to struggle though, probably comparable to 780 TI reference model.
I last read something that said March 2017. 2 months from now we might have the 1080ti.
http://www.ebay.com/itm/GeForce-Tit...419370?hash=item1a18837a2a:g:vjAAAOSwZVlXoK6L
It begins!
Hoping I can get at least that for mine, and then actually buy one to use when they are readily available in a month. If my actual cost is more like 700-800 I won't feel like as big of a sucker.
LOL $1800 for a card thats 20% faster wtf! I hope AMD releases a card soon that beats this Titan for $500
Why would AMD do that?
We need a dedicated Titan X benchmark thread for those who are getting their GPUs over the coming days.
I'd like to see some OC results as well. Hopefully it rocks a mean OC. Then hopefully we get a 1080ti for $650 that OCs just as well. A 980ti repeat would be pretty sweet.
Using hbm should help. A fury gets 60% more performance/watt than a 390 according to techpowerup.
I hear hbm also helps with die size, since you don't need the on die stuff to drive the signal through the pcb as with gddr.
Looks like NV will have a 400mm+ chip, a 300mm+ chip and a 200mm+ chip for the first round of Pascal.
When it's time for a refresh they'll drop everything down one segment and put a 500mm+ chip on top of it all.
Think what you want about NV, but they have a plan and a vision and they're executing on it like beasts.
2 x 480 (Assuming that's what Vega is) will be 510mm² die on 16nm TSMC FF. That's bigger than GP102, yet unless they manage to scale clocks higher and improve power efficiency, it's going to be somewhere around 20-25% slower, so basically maybe only 10% faster than a GTX 1080.
Also sure, HBM2 will decease die size. But even a die this big would probably only cost $50/chip on TSMC FF. It would be cheaper to use GDDR5X and use a bigger die, than use HBM2 and a smaller die since the die itself isn't exactly that expensive to begin with and HBM2 is way more expensive than GDDR5X. The actual bottlenecks are TDP and how much performance you are aiming for and what your expected profit margins are.
The only reason Vega is using HBM2 is to get more performance (high core clockspeeds) under a 300W envelope, not for cost reasons because a bigger die with GDDR5X is actually the cheaper solution.