AMD Working On An Entire Range of HBM GPUs - Has Priority To HBM2 Capacity

Source :- Wccf.

 

That is pretty fast.

 

Yeah Wccf.

 

Once that happens, game devs and Nvidia will move the goal posts further (with a gameworks 2 or whatever) and cripple 2016 cards .

 

Wccf..journalism at it finest.

 

lol they made the nv summer sales seem like they were going out of business


but good for amd if true since they have to work on it....or do something
so the next cards I get don't cost 2000usd

 

This is what I have felt about Fiji all along. AMD would be well ahead of NV in the Pascal vs Arctic Islands. They have already cracked HBM where as it would be NV's first attempt.
Fury may not sell as well hoped(I plan on buying one tho)But the preemptive strike for the next gen has happened

 

What do you think is going to happen exactly? Nvidia selling engineering samples of an architecture that they haven't prepared for HBM?

 

Well seeing as AMD developed this tech exclusively in conjunction with Hynix,. I really don't think either Hynix or AMD just gonna pass on all their R6D or tech to NV do you? So that means NV has to come up with their own take on it...From scratch. So as I said AMD is already one step ahead of NV.

HBM didn't exist until AMD and Hynix came up with it.

 

HBM is a JEDEC standard. It's not proprietary. It's no different then what AMD did with GDDR. Nvidia was going with a different tech, developed by Intel and Micron, called HMC, but it got delayed so they are switching to HBM (Which is the better standard for GPU's anyway).

Regardless, Nvidia doesn't have to develop it's own take, it's a standard. TSMC/Samsung/GF will/have all developed approved methods for manufacturing it. The hard part for Nvidia will be to build an architecture that communicates/works with it. I'm sure they'll manage. But PASCAL is definitely going to be a major launch for them. A lot of new technologies all coming together.

 

Mid range segment could become more interesting. If the past is any indicator, memory bandwidth was often something the mid range segment really ... well sucked at.

 

+1 to HBM APUs, especially if they are 8CPU cores Zen + 16~32 graphics cores.
Idea of having this kind of SoC at size of Fury X + to just plug in PSU + HDD and it would just work. It could steal sales from mATX and maybe even ATX too.

HMC not only got delayed, but is not exactly best suited for GPUs due to its memory controller. While it is great idea and very universal, I think it is better for clean CPU.

 

It will be interesting to see how this actually pans out.

 

4k or not, i'm looking for a card that can play all games 100% maxed out, at least 120fps on 1440p, single card...

Never happened never will, i guess i'm looking for too much

 

New GPU and CPU architectures, new manufacturing processes... Well, 2016 should be interesting

 

Adding to Denial's replies, Nvidia is going straight for HBM2 with Pascal.
That's probably the only reason AMD is already ramping up to it for 490s since Pascal'd be out in 2016.

 

HBM is a JEDEC standard so AMD cannot deny Nvidia the right to use the technology.
And HBM will become pretty standard soon and other companies like Samsung will probably begin to manufacture them chips.

And how is AMD one step ahead of Nvidia? We can clearly see from Fury X's performance that HBM did very little. I don't know why some of you guys see HBM as the second coming of jesus christ, magically making cards twice as fast. You need a powerful GPU to actually utilize that fast RAM. If you slap HBM on a 610m it won't magically run 4k at max details.
All HBM does is remove the potential bottlenecks which are semi-present in 4k and which will be present soon with further IQ advancements coupled with 4k.

Not to mention Pascal has been in the works for quite a while now. You can't speculate on what you do not know. I know what AMD presented this generation. And it was disappointing.
I have no reason to believe the next gen will be better considering how much money AMD is bleeding, you can say that I've lost faith in them.

 

Yeah, the company that failed to beat 980Ti reference cards with their hyped "Fury" cards now apparently has exclusive access to future tech.

Derp.

Even IF this is true, the fact remains that with Gameworks, nVidia has managed to integrate themselves into the games (unlike the useless TressFX crapware). So if I really want to experience the bells and whistles integrated into a game like Witcher 3, I need nVidia card.

THAT and I am myself a game developer who uses Unreal engine, which is basically designed with nVidia tech integration in mind. Chances of me abandoning my work for a GPU that won't run nVidia tech integrated into Unreal 4, are none.

 

On the first slide I read it as 8Gb per die up from 2Gb.

The current HBM stack is 4 high x 2Gb per die = 8Gb = 1GB per stack. 4 stacks = 4GB total vram

The new stacks with be either 4 or 8 high at 8Gb per die (4GB/8GB per stack) assuming 4 stacks this = 16GB at 4 high stack and 32GB at 8 high stack.

Perhaps my math is wrong somewhere but I don't see HBM2 allowing for only 8GB vram as per the article.