Both Samsung & SK Hynix to Supply HBM2 for NVIDIA's Pascal GPU

[HKZeroFive]

According to industry sources on Sept. 20, Samsung Electronics and SK Hynix are planning to mass produce second-gen High Bandwidth Memory (HBM) for Nvidia's Pascal GPU starting in the first quarter of next year, after completing pilot production and reliability tests by the end of 2015.

Earlier this year, SK Hynix was already supplying first-gen HBM to AMD and Nvidia ahead of Samsung. Micron uses Hybrid Memory Cubes (HMC), a technology similar to HBM, but falling behind it in productivity.

Meanwhile, the two dominant players in the global DRAM memory market are keeping a close eye on a change in the dynamics of technology in the memory industry, centered on HBM.

An industry source said, "There are clear signs of a change in the structural design of CPUs, GPUs, DRAM memory, and storage led by Intel and Nvidia, which control standards in the market." The source added, "Samsung and SK Hynix are both expanding their technical bases for various types of next-gen memory, including HBM chips."

Alright...

I wasn't quite expecting both Samsung and SK Hynix to both supply NVIDIA with HBM2 memory. I'm guessing they will share their HBM2 memory over a range of different Pascal GPUs. This prompts the question - which GPUs will have Samsung's HBM2 tech and which will have SK Hynix's?

If patrickjp93's claims (not singling you out buddy) of Samsung possessing advanced HBM2 are indeed true, then I guess they will be the ones dominating the high-end NVIDIA GPUs.

So what will AMD do? So far, it looks like only SK Hynix is manufacturing their HBM2 memory for their incoming Arctic Islands graphics cards. Unless Samsung is willing to lend some HBM2 to them as well

Sauce: http://www.businesskorea.co.kr/article/industry/manufacturing/12151/hynix-and-nvidia-samsung-sk-hynix-supply-high-bandwidth-memory

[Muz]

I'm waiting for Arctic Islands, cause it'll be cool right?

[kwljunky]

I'm waiting for Arctic Islands, cause it'll be cool right?

It will be a cool story, bro.

[blu4]

It will be a cool story, bro.

Will go down in history like a breeze!

 

I hope the HBM2 from Samsung will overclock like a champ I bet @Lays will love that

[kwljunky]

 

 

Well if there one thing I learnt from the history of breezes..

 

[bogus]

Where is Patrick? LOL

[Prysin]

Where is Patrick? LOL

hiding from his partial failure.

soo... lets see.. will this be the new ramgate?

Cards with different types of HBM depending on which HBM modules they had lying about?

[Alurit]

not surprising considering that AMD had supply problems even with their lower market share and with niche cards

[Guest Strangerbob]

Until there are sub 300$ HBM cards and HBM APU`s... all of this HBM talk is worthless. 

 

I bet its gona be end of 2017 when that actually happens. Until then... only Titan Xs and Fury Xs and XXX Xs will get HBM.

 

 

But you have to start somewhere... obviously the expensive end.

[GeathX]

Cards with different types of HBM depending on which HBM modules they had lying about?

Or regulate the slower grade HBM to mid lower cards. Even HBM1 is going to be more than fine for the mid high range of cards, let alone the mid low. Might get a series of cards released that have HBM1,2 and 3 all at once depending on how cost effective it is to keep different variences going.

[Real_PhillBert]

[Prysin]

Or regulate the slower grade HBM to mid lower cards. Even HBM1 is going to be more than fine for the mid high range of cards, let alone the mid low. Might get a series of cards released that have HBM1,2 and 3 all at once depending on how cost effective it is to keep different variences going.

We are talking of Nvidia... their track history shows that when it comes to memory, they will make weird decisions that does not always make sense....

[Notional]

I really doubt there will be much difference between SK Hynix and Samsung's HBM2 in terms of performance. However, even if there is a difference, the average consumer and gamer will never be able to tell, as no games should be able to mx out the bandwidth. A performance difference would only be noticeable in the professional compute world.

 

Either way, SK Hynix has proved they can make HBM, but we have yet to see Samsung do so. However, considering their 3D NAND used in SSD's, we know they have the technology to stack NAND. I've said it before and I'll say it again, 2016 is going to be a very interesting year in terms of hardware (and gaming too, so many awesome titles and DX12).

[patrickjp93]

hiding from his partial failure.

soo... lets see.. will this be the new ramgate?

Cards with different types of HBM depending on which HBM modules they had lying about?

I move from article to article down the line. I hadn't seen this one yet.

 

And I take this moment to say this:

[patrickjp93]

I really doubt there will be much difference between SK Hynix and Samsung's HBM2 in terms of performance. However, even if there is a difference, the average consumer and gamer will never be able to tell, as no games should be able to mx out the bandwidth. A performance difference would only be noticeable in the professional compute world.

 

Either way, SK Hynix has proved they can make HBM, but we have yet to see Samsung do so. However, considering their 3D NAND used in SSD's, we know they have the technology to stack NAND. I've said it before and I'll say it again, 2016 is going to be a very interesting year in terms of hardware (and gaming too, so many awesome titles and DX12).

Considering Samsung is already making the TSV's Micron and Intel are using for HMC, I wouldn't doubt Samsung's purported schedule of Q1 2016 ramped up production, meaning supplies would ship to Nvidia late Q1 or early Q2 and Pascal would be out the door by the end of Q2 or early Q3.

[PlayStation 2]

all these shitty puns about arctic islands are giving me cancer

[Notional]

Considering Samsung is already making the TSV's Micron and Intel are using for HMC, I wouldn't doubt Samsung's purported schedule of Q1 2016 ramped up production, meaning supplies would ship to Nvidia late Q1 or early Q2 and Pascal would be out the door by the end of Q2 or early Q3.

 

I doubt HBM supply is the limiting factor or holding back production of Greenland and Pascal. I think the stop gap is the actual GPU due to 16nm FF production difficulties. I fear we won't see either graphics card until Q3-4 next year.

[patrickjp93]

I doubt HBM supply is the limiting factor or holding back production of Greenland and Pascal. I think the stop gap is the actual GPU due to 16nm FF production difficulties. I fear we won't see either graphics card until Q3-4 next year.

16nm FF+ yields are already above 80%. That won't be the issue at all. Not to mention Nvidia and AMD can't afford to be late behind the Knight's Landing launch in Q1 2016. Intel's already scooping up more HPC contracts. Nvidia's grip is slipping, and AMD needs Arctic Islands to be up for serious consideration if it intends to pair them with Opterons and their new HPC APUs.

[Notional]

16nm FF+ yields are already above 80%. That won't be the issue at all. Not to mention Nvidia and AMD can't afford to be late behind the Knight's Landing launch in Q1 2016. Intel's already scooping up more HPC contracts. Nvidia's grip is slipping, and AMD needs Arctic Islands to be up for serious consideration if it intends to pair them with Opterons and their new HPC APUs.

 

Yeah and 14nm FF are already in mass production status. Getting the chips to perform at over 1Ghz at those sizes might be an issue. Seeing the 900 series from NVidia "easily" hit 1.5Ghz, I assume people would like to see these chips matching or exceeding those speeds (even if it's not important at all).

[marldorthegreat]

I'm waiting for Arctic Islands, cause it'll be cool right?

more efificient not more cool. they might run a little hotter since the die is very dense. And they can cram more cores. They probably will have a 8192 core gpu. Using 250 watts with 8gb hbm 2

[Muz]

more efificient not more cool. they might run a little hotter since the die is very dense. And they can cram more cores. They probably will have a 8192 core gpu. Using 250 watts with 8gb hbm 2

How did you not get the joke?

[Bojamijams]

How did you not get the joke?

 

He's british.. they have a different kind of humor there

[Icecube26]

According to industry sources? .... who are they? ... lol... i the funniest rumours are the ones that done have any real source behind them like a person or company.... like this one... this isnt even a rumour... i have a feeling that this is just someones opinion....

[BobbyG2]

Yeah and 14nm FF are already in mass production status. Getting the chips to perform at over 1Ghz at those sizes might be an issue. Seeing the 900 series from NVidia "easily" hit 1.5Ghz, I assume people would like to see these chips matching or exceeding those speeds (even if it's not important at all).

I would imagine with the transition over to asynchronous compute, the increase in shaders and rops from the lower manufacturing process would more than make up for the lack of clock speed.

We're coming to the point where more cores really do mean more performance in gaming with dx12, including with cpus. It's exciting.

[Notional]

I would imagine with the transition over to asynchronous compute, the increase in shaders and rops from the lower manufacturing process would more than make up for the lack of clock speed.

We're coming to the point where more cores really do mean more performance in gaming with dx12, including with cpus. It's exciting.

 

Indeed. But async compute (or compute shaders) do not use ROP's, so we might see hardware just focus on increasing shader count. But generally a lower node means we will get a lot more transistors that more than makes up for the lower hertz. Just look at Skylake versus Haswell.

 

It's very exciting indeed