I see three possibilities:
1) HBM2 controller is backwards compatible with HBM and they'll use HBM for testing the Pascal silicon.
2) NVIDIA is shipping the prototype elsewhere for developing and testing the interposer. HBM2 doesn't need to be available yet to accomplish that.
3) NVIDIA got their hands on HBM2 prototype chips.

That's what I thought too but then wouldn't all companies testing/sampling HBM be included. Just odd to me the report translated to supply ahead of Samsung.

Edit: to make it a little bit clearer:
Incinuating both (SK Hynix & Samsung) were to supply the HBM1 memory to AMD & Nvidia, SK Hynix being able to supply it first before Samsung but both supplying HBM1.

arbiterProblem with report as AMD getting "priority" as claimed in that other guys story, Is AMD even close to being ready with artic islands to make use of that so called priority.

As far as priority is concerned, that whole story, and every reprint of it, hinges on an WCCFtech article based upon an "unnamed source". I have very little faith in WTFtech to start with, and when they go all deep throat, and not a single industry site known for its insider scoops corroborates it, I'm inclined to remain skeptical.

arbiterCould see Pascal in q2 but is AMD gonna have one before q3? if not then screwing nvidia over when they are willing to pay for the chips now and use them would be kinda stupid.

AMD is surely working on Arctic Islands, but I doubt that anything other than the flagship GPU will be using HBM2. Timing of release? Who knows. AMD historically play it close to the chest with GPUs (unlike their megaphone CPU pronouncements). Unless HBM2 in production is as rare as rocking-horse manure (in which case the cost will be prohibitive) I can't see a single vendor bogarting the supply.

XzibitSomething is wrong with the translation or the report. If you go by that report there is a HBM1 Nvidia card or something coming. Nvidia said they would only use HBM2. Unless Nvidia likes buying HBM1 for the fun of it.

As @FordGT90Concept noted, I would think that Nvidia got at least some HBM for evaluation purposes. I doubt SK Hynix would cut itself off from a potential customer knowing Samsung is also going to be a producer. Samsung has roughly twice Hynix's market share, so I would say they can ill afford to pick and choose given the likely limited number of initial customers. HBM2 won't be cheap, and will require a high end price to justify its use. High end graphics are one of the few products that fills the bill.

Then again it might simply be a translation error or typo that changed the intended message. I guess we will find out in due course.

arbiterProblem with report as AMD getting "priority" as claimed in that other guys story, Is AMD even close to being ready with artic islands to make use of that so called priority. Could see Pascal in q2 but is AMD gonna have one before q3? if not then screwing nvidia over when they are willing to pay for the chips now and use them would be kinda stupid.

FordGT90ConceptI see three possibilities:
1) HBM2 controller is backwards compatible with HBM and they'll use HBM for testing the Pascal silicon.
2) NVIDIA is shipping the prototype elsewhere for developing and testing the interposer. HBM2 doesn't need to be available yet to accomplish that.
3) NVIDIA got their hands on HBM2 prototype chips.

Development samples tend to be more relaxed in terms of exclusivity, so you can still develop your product with the timed-exclusive stuff but not get volume quantities.

HumanSmokeAMD is surely working on Arctic Islands, but I doubt that anything other than the flagship GPU will be using HBM2. Timing of release? Who knows. AMD historically play it close to the chest with GPUs (unlike their megaphone CPU pronouncements). Unless HBM2 in production is as rare as rocking-horse manure (in which case the cost will be prohibitive) I can't see a single vendor bogarting the supply.

Using current nvidia gpu line up as reference, gtx970 and higher end would use it where as gtx960 and lower would use GDDR5 yet. HBM is still to expensive to make a cheap gpu with. and will be for least another 1-2 gen.
I know AMD is working on it but if production of HBM2 does start in q1, would that chip be ready in ~6-7months in a product to ship?

HumanSmokeAs @FordGT90Concept noted, I would think that Nvidia got at least some HBM for evaluation purposes. I doubt SK Hynix would cut itself off from a potential customer knowing Samsung is also going to be a producer.

Nvidia was pretty clear to be skipping HBM1, which looking at how limited stock of fury cards have been was a good idea else stock would been even more limited.

64KWhy expect GP100 to be affordable? It won't be a mainstream GPU. It's a luxury and luxuries put a big dent in your wallet.

$500 ten years ago is now $610 today due to inflation.

data.bls.gov/cgi-bin/cpicalc.pl?cost&=

And the 8800 Ultra which was the single most expensive card in 2007at $830 today comes in at $953 after inflation.

data.bls.gov/cgi-bin/cpicalc.pl?cost&=

i can only handle so much stupidity at once, this thread has taken 3 goes to read.....

hbm brings nothing xD


that is comedy gold, when all cards are hbm i will quote that for you xD

Eh, wot? GDDR5 has been around since 2007. It's old. HBM is past due.

FordGT90ConceptEh, wot? GDDR5 has been around since 2007. It's old. HBM is past due.

I'm not so sure about that. For the next series of cards, yes, probably.
For any current or previous generations of cards? No.
Hawaii with GDDR5 isn't appreciably behind Fiji with HBM as a general rule (and I suspect the difference will be smaller with DX12). The GPUs - especially in consumer (gaming) applications just aren't bandwidth constrained in the vast majority of usage scenarios. With a smaller manufacturing process allowing a jump in GPU performance via increased throughput the wider interface will begin to show its strengths, but you wont be seeing HBM allied with a mainstream card for a while.

The advantage of HBM is not only bandwidth, but latency (right next to GPU), density (1 GB chips--GDDR5 is only just getting that), and power consumption. I'd argue Fiji does exhibit all of those features compared to Tonga.

FordGT90ConceptThe advantage of HBM is not only bandwidth, but latency (right next to GPU), density (1 GB chips--GDDR5 is only just getting that), and power consumption. I'd argue Fiji does exhibit all of those features compared to Tonga.

For sure. But I'm not convinced that those positives outweigh the disadvantages of increased cost and lack of performance conferred insomuch as the current and previous generations are concerned. GDDR5 like its cousin DDR3 is dirt cheap, and the increased size of the PCB required for usage is negligible, so it would be hard to justify the increased cost of HBM+interposer+microbumping for a current/previous architecture where it makes little if any performance impact. I'm going to go out on a limb and say that Fiji with a 512-bit/8GB GDDR5 implementation wouldn't suffer in comparison even if its core count were shaved down to accomodate the GDDR5 I/O and IMC's, so long as the ROP and ACE/HS remained intact.

Speaking in English again, does the HMB guarantees higher performance in games with 2160p resolution and 4xMSAA compared to GDDR5, or not?

Prima.VeraSpeaking in English again, does the HMB guarantees higher performance in games with 2160p resolution and 4xMSAA compared to GDDR5, or not?

As of right now, the GPUs (Fury Lineup) seem to be more processing-limited than memory limited.

ZeDestructorAs of right now, the GPUs (Fury Lineup) seem to be more processing-limited than memory limited.

This has been the case for a while, for some reason people have been sticking their heads in the sand about the fact that GPUs run out of power before they run out of memory. The Fury cards take slightly less of a hit going into 4K than equivalent GDDR5-equipped cards, but it's not that big of a difference considering the limitations of Fury/HBM (goodbye overclocking).

xenocideThis has been the case for a while, for some reason people have been sticking their heads in the sand about the fact that GPUs run out of power before they run out of memory. The Fury cards take slightly less of a hit going into 4K than equivalent GDDR5-equipped cards, but it's not that big of a difference considering the limitations of Fury/HBM (goodbye overclocking).

The Compute/HPC guys can't get enough of it though :P

But yeah, between the constant AMD circlejerk (if you have a long enough memory, you know full well AMD is just as bad as Intel/nVidia when they're in the lead), misinformation, lack of understanding (both technical and marketing, especially in cases like the 970), endless leaks and overhyped everything... the average "modern" gamer is a hyperexcited idiot...