Nvidia's upcoming Pascal GPU pictured with HBM 2.0

If AMD and Nvidia are neck and neck, AMD still looses. Nvidia will keep banging it's "The way it's meant to be played" and GameWorks drum. Even if the cards were equal on a performance standpoint, Nvidia would win out in actual games. On top of that, no one is going to switch over to AMD because they are equal value to performance compared to Nvidia. AMD absolutely has to beat Nvidia or find another rich donor or else it'll go bankrupt.
Who knows maybe the optimization for the AMD Gpus in the consoles will help AMD on the PC stay competitive against gameworks.
 
16GB sounds extremely expensive.

The titan X had 12Gb and was around £800 (probably around $1000 cause they are cheaper than UK). I just got a 980ti and its list price is £550 so expect the new pascal cards to be between £500-1000 (maybe 700-1500$)


Also around June ish there were rumours that Microsoft plan to buy AMD not sure of any more details though
 
I was speaking about the scenario in which Nvidia Pascal is awesome and does crush current offerings, as it's forecasted.
I wouldn't worry too much about Pascal vs current offerings. If history is any indicator, Pascal and AMD's Arctic Islands should be close in performance. That will determine the market segment lineup.
AFAIK, GP100 will (at least initially) be exclusively deployed as Tesla products in HPC/ Virtualization/ Visualization/ Neural Net environments (the latter being a potential high dollar goldmine technology going forward). By the time a GeForce variant arrives I would expect both AMD and Nvidia to be fairly competitive (as usual).
Arctic Islands will be produced on 14nm process, Pascal will be made by TSMC on a 16nm process.
Source? Latest rumour has TSMC's 16nmFF+ as the frontrunner (with GloFo tapped for 14nm LPP for Zen in late Q4 2016 and entry/mainstream GPUs). Probably because they are already shipping large Pascal silicon, where Samsung 14nm LPP has yet to deliver anything tangible, and the current 14nm LPE seems to be running into some problems. Samsung were originally tasked with the bulk of Apple's A9 production, but even though Samsung's 14nm LPE A9 is slightly smaller than TSMC's 16nm FF (non "+" the bulk of deliveries seem to favour TSMC. It doesn't auger well for 14nm LPP vs TSMC 16nm FF+/FFC. At least Samsung's HBM2 is made on a mature (20nm) process.

Close in performance is one thing. The main thing is release date. If AMD is going to be 8 months behind again, their cards really need to be better or much cheaper for same performance.
 
Also around June ish there were rumours that Microsoft plan to buy AMD not sure of any more details though
That rumor just sort of disappeared, and many analysts found it to be very unlikely as there wasn't much to gain of it. Even though nothing was said either way, I'd place all my money on it wouldn't happen.
 
That rumor just sort of disappeared, and many analysts found it to be very unlikely as there wasn't much to gain of it. Even though nothing was said either way, I'd place all my money on it wouldn't happen.
The buyout rumours always appear just before the announcement of a particularly bad quarter for AMD. It stabilizes their stock. Likewise AMD tend to jam in a few feelgood announcements at the same time. You can expect at least a couple more ( besides the mobile FirePro W7170M/W7150M launch and the Dell/Oculus deal already announced) between now and October 15, when AMD announce their third-quarter financial statement.
Close in performance is one thing. The main thing is release date. If AMD is going to be 8 months behind again, their cards really need to be better or much cheaper for same performance.
8 months behind? How do you get that? GM 200 launched in March, Fiji in June
 
Last edited:
Maybe they are referring to HBM 2.0, where AMD is currently at the first version. Or am I misunderstanding HBM versions?
I haven't heard anything too much in the way of definitive information.
AMD are using the first iteration of HBM, but that seems to be more a proof of concept design. SK Hynix needed a launch customer, and AMD as a co-developer of the specification, needed to fulfil that role. I think you can consider the Fury line as pipecleaners for the technology - hence the slow ramp of the interposer package and AMD's near total lockdown on official voltage adjustment.
As for HBM2, we know that SK Hynix and Samsung are both gearing up production, and Nvidia (and probably AMD) will be sourcing from both companies. The memory itself is just DDR - very much a known quantity, but the microbumping (the extremely small solder connections between layer of both the DDR, base die, and substrate) requires a large measure of precision - and some specialized test and verification equipment that is only just now coming on line. Something I suspect is at the heart of the lack of Fury cards.

I think the issue you are referring to revolves around some rumours that AMD has priority on HBM from SK Hynix, and that Hynix is supposedly ramping HBM2 at a slower rate than Samsung (with Nvidia reportedly sourcing from both). It doesn't necessarily follow that AMD cannot source from both companies, or that the HBM2 production ramp at Samsung/Hynix is the limiting factor in when the next generation of cards arrive - I'd think that viable GPU production (wafers produced and yield) would be the prime mover in when we see the next gen.

The first Pascal parts will undoubtedly be earmarked for professional deployments, so even if GP 100 arrives (test samples from a TSMC 16nm FF+ hot lot are already being tested) first I doubt it will impact the gaming community - forum wars aside, until production is great enough to supply the consumer channel. The other marker will be how AMD's big die is shaping up. Neither side will want to tip their hand too early - the penalty for not getting the clockspeed/heat/power/price calculation right and missing the mark against your competitor is severe, as AMD found out to its cost when Nvidia ruined the Fury X party with the pre-emptive GTX 980 Ti launch.
 
Between this and Zen (yes I know GPUs to CPUs is like apples to oranges), 2016 will be a very interesting year. Then again it may be as much a letdown as Bulldozer and Maxwell all over again.
Your comment makes no sense. First of all what is wrong with Maxwell? And second, how can you even compare it to the disaster that was Bulldozer?!
 
Back