Samsung mass producing HBM memory in 2016

[nvgpu]

http://www.computerbase.de/2015-08/idf-2015-samsung-fertigt-high-bandwidth-memory-ab-2016/

http://www.fudzilla.com/news/graphics/38555-samsung-to-make-hbm-2-memory-in-2016

It was naive to believe reports that AMD will have SK Hynix love since it had a head start with the Fiji codenamed Fury X branded cards. Nvidia has close to 80 percent of the market, memory manufacturers including SK Hynix or Samsung cannot ignore that. They are in the business of making money.

 

[frozentundra123456]

So it looks like all the hopes of AMD fans that nVidia would not get HBM are not going to come true.

 

[Keysplayr]

I don't really know where that hope came from anyway. The sheer amount of difference in purchase volume between NV and AMD alone would likely make NV the bigger client priority. I mean, who doesn't like money?

 

[beginner99]

I don't really know where that hope came from anyway. The sheer amount of difference in purchase volume between NV and AMD alone would likely make NV the bigger client priority. I mean, who doesn't like money?

Still the possibility there was a contract between AMD and Hynix that AMD gets "premium access" because they helped them develop it. That contract probably was done half a decade ago.

Also why is there money to make? With the proclaimed death of dGPU and dGPU having low volume anyway compared to say mobile chips, I don't really see there being that much money in HBM. As far as we know it's expensive so already a no-go for mobile SOCs besides them not needing it anyway.

NV and AMD will be the only ones to use it in the foreseeable future. Intel/micron have their own thing so also not possible to get money from Intel.

 

[Keysplayr]

Why restrict HBM to dGPU only? HBM can be used in every other segment I would imagine. Mobile. Professional. Commercial/Compute. Super computing. Automotive.
Anyway, if we must restrict HBM to dGPU, then what I said refers to Nvidia having massively more potential order volume than AMD. Nearing 80% and climbing? So, the client where the MOST money to be made from HBM sales in the dGPU market between AMD and Nvidia, is Nvidia. Hynix would have been incredibly stupid to paint themselves into a corner by not giving Nvidia at least EQUAL dibs on memory. But wonders never cease.

 

[5150Joker]

I don't really know where that hope came from anyway. The sheer amount of difference in purchase volume between NV and AMD alone would likely make NV the bigger client priority. I mean, who doesn't like money?

Common sense it seems isn't as common as one would think. Pascal will be out on time with HBM 2 and AMD won't be enjoying some exclusivity like some had hoped (why would anyone hope for that anyway?).

 

[thesmokingman]

AMD won teh right to use HBM1, not exclude NV from using HBM.

 

[tential]

Common sense it seems isn't as common as one would think. Pascal will be out on time with HBM 2 and AMD won't be enjoying some exclusivity like some had hoped (why would anyone hope for that anyway?).

If pascal is out to market first next year amd can just give up. The amount of people who will wait past pascal for a gpu upgrade after seeing benches is going to be miniscule.

Amd has killed any good faith they had with this launch. No one is going to expect amd to change uo the game if they launch second, even if they actually doing.

 

[Despoiler]

AMD won teh right to use HBM1, not exclude NV from using HBM.

The rumor is that AMD has a supply contract with SK Hynix that gives them first dibs on HBM2 should there be supply constraints.

I very much doubt Nvidia is going to have a smooth go at HBM2 supply with Samsung. The articles I've read have indicated it's a monumental achievement to get HBM1 production off the ground given all of the parts that need to come together. AMD and their partners have been working on this for years. Remember that HBM1 isn't a spec. It's a partition that SK Hynix made to allow an easier time getting the product to market when it was needed. HBM(2) is the only thing in the JEDEC spec.

 

[PrincessFrosty]

I don't really know where that hope came from anyway. The sheer amount of difference in purchase volume between NV and AMD alone would likely make NV the bigger client priority. I mean, who doesn't like money?

Me!

Just kidding.

Lets face it Nvidias market share is now so large they can afford to pull exclusive contracts and with GPUs really struggling to get speed bumps while stuck on the 28nm node we've got gamers hungry for a tech shift and some real performance gains.

 

[boozzer]

who cares, guys, you really should care when you have concrete info either way. not speculation or some article from fudzilla.

 

[Abwx]

Common sense it seems isn't as common as one would think. Pascal will be out on time with HBM 2 and AMD won't be enjoying some exclusivity like some had hoped (why would anyone hope for that anyway?).

Samsung will produce HBM1 only given the numbers in the slide, and they had to licence the process from Hynix/AMD to be ready that soon.

HBM being a Jedec standard doesnt mean that the process is free, what Jedec cover are the electrical caracteristics and the communication protocol, rest is not covered, otherwise Rambus couldnt had got royalties from DDR/DDR2/DDR3 manufacturers..

HBM2 is not even ready for production and wont be so untill late 2016, but you can always dream that Nvidia will get it in six months...

 

[Keysplayr]

Nvidia would just use DDR5 again. They seem to have that bandwidth thing under control.

 

[AnandThenMan]

Nvidia would just use DDR5 again. They seem to have that bandwidth thing under control.

This guy doesn't think so.

 

[Keysplayr]

This guy doesn't think so.

Well since you apparently have all the answers, what's it going to be?
Is it going to be Nvidia using HBM for Pascal limited to 4GB?

Or wait to release Pascal at the time HBM2 is available at the end of 2016 according to folks in here?

Or stick with GDDR5?

Well, to the first option I say Har dee har harrrrrr! I don't know of any company that would shoot themselves in the fo......

Second option? I seriously doubt it.

Third, if need be.

Then there is also the option that Pascal will release on schedule with HBM2 properly in place long before the end of 2016.

http://www.fudzilla.com/news/graphics/38555-samsung-to-make-hbm-2-memory-in-2016

http://linustechtips.com/main/topic...sung-is-in-and-beat-hynixs-initial-offerings/

http://wccftech.com/samsung-enters-hbm-market-1h-2016-hpc-gpu-ready-hbm-15-tbs-bandwidth-48-gb-vram/

 

[sandorski]

The presumption that Nvidia has to be courted to sell HBM because Nvidia has 80% of the Market is not a good presumption to make. Market share could change very quickly in Graphics. Especially if AMD released with HBM 2 and Nvidia GDDR5. The potential technical advantage could change perceptions very quickly.

 

[jpiniero]

GP100 is definitely coming out before HBM2 is available. It has to; they will lose HPC to Intel if they wait.

The presumption that Nvidia has to be courted to sell HBM because Nvidia has 80% of the Market is not a good presumption to make. Market share could change very quickly in Graphics. Especially if AMD released with HBM 2 and Nvidia GDDR5. The potential technical advantage could change perceptions very quickly.

16 nm GPU performance increases (in games) are going to be tiny. It's not like GDDR5 is going to suddenly become a huge bottleneck. The power savings of HBM are important but as long as nVidia's GPUs are much more efficient they can live with it.

 

[sandorski]

GP100 is definitely coming out before HBM2 is available. It has to; they will lose HPC to Intel if they wait.



16 nm GPU performance increases (in games) are going to be tiny. It's not like GDDR5 is going to suddenly become a huge bottleneck. The power savings of HBM are important but as long as nVidia's GPUs are much more efficient they can live with it.

That remains to be seen. However, so much of what succeeds in this Market is Customer perception. There is no guarantee that Nvidia will maintain its' current Market dominance.

 

[positivedoppler]

There's a lot of silly speculation here, but lets put our common sense hat on. AMD funded the development of HBM. AMD is still a billion dollar corporation that is probably smarter as a whole with their fleet of lawyers smarter than any poster here. They did the earlier "Heavy lifting" for Hynix. It would be incredibly incompetent of them if there is nothing in the contract of Hynix that give them a competitive advantage.

 

[positivedoppler]

This guy doesn't think so.

Oh for F' sake, can he please wear an shirt underneath so we don't have to see those saggy nipples....

 

[Keysplayr]

There's a lot of silly speculation here, but lets put our common sense hat on. AMD funded the development of HBM. AMD is still a billion dollar corporation that is probably smarter as a whole with their fleet of lawyers smarter than any poster here. They did the earlier "Heavy lifting" for Hynix. It would be incredibly incompetent of them if there is nothing in the contract of Hynix that give them a competitive advantage.

Well that's great, if Hynix was always going to be the only HBM game in town.

And by the way, Hynix is many times the worth of AMD. Something over 20Billion US market cap.

 

[Arachnotronic]

AMD funded the development of HBM. AMD is still a billion dollar corporation that is probably smarter as a whole with their fleet of lawyers smarter than any poster here. They did the earlier "Heavy lifting" for Hynix. It would be incredibly incompetent of them if there is nothing in the contract of Hynix that give them a competitive advantage.

[citation needed]

 

[Abwx]

[citation needed]

No need of citation, the whole process is patented...

One cant just do an analysis of the methodolgy by scrapping the chip and then state :

"hey, there s nothing that can be patented, everything is logical and straigtforward, there s no other mean to manufacture it, so we had forcibly to use the very same solutions".....

Apple almost managed to patent a square and you think that Hynix/AMD are that stupid to give their work for free...??.

 

[Keysplayr]

No need of citation, the whole process is patented...

One cant just do an analysis of the methodolgy by scrapping the chip and then state :

"hey, there s nothing that can be patented, everything is logical and straigtforward, there s no other mean to manufacture it, so we had forcibly to use the very same solutions".....

Apple almost managed to patent a square and you think that Hynix/AMD are that stupid to give their work for free...??.

So, so far, there isn't a need to educate yourself, and no need for citations now. If they are asked for, Abwx, that generally means they are wanted and needed.

 

[Abwx]

So, so far, there isn't a need to educate yourself, and no need for citations now. If they are asked for, Abwx, that generally means they are wanted and needed.

Much simpler manufacturing processes than this are patented, what you fail to understand is that it s the methodology that is patented.

Of course you can dispute this patent in courts, but you ll have to prove that there was functional prior art, and there s no prior art that is functional, otherwise one could patent anything that has a potential to be manufactured without even proving that it can be manufactured, that is, to patent a possibility...

Now, do you think that it s a guess less educated than what you posted..?.