Samsung mass producing HBM memory in 2016

[Ma_Deuce]

How happen that the so called underdog is the most innovative.?.

Because they have to be IMO. The underdog has to do something great to set themselves apart. They have to risk more. Once you are able to sit on the throne it's easier to take the safer bets and rely on reputation or your other competitive advantages that you have.

 

[tential]

^bold. This is really the entire "problem" we've come across in here. Because AMD does so poorly, people are hoping that they can get one over on the one that is doing so well, as in Nvidia. Underdog support is all well and good, but there does come a time where the dog has to be put down because it can't stand up anymore and can't eat.

I don't believe AMD should be put down. I believe the x86 license should be transferable, and that AMD should be bought out so someone else can have a chance at competing.

The way the market is structured now, like MANY markets, is not true capitalism.

 

[Keysplayr]

You mention all of the other markets that HBM could be useful for, but then quote Nvidia of having 80% more potential order volume. Is that 80% you are throwing out just the small discrete gpu market or is it taking into consideration other markets as well?

If Hynix is selling all they can make now and getting contracts in place to ramp up production and sell all the future units they can make, they aren't painting themselves into anything by excluding Nvidia. (If they are, all it is is speculation from both sides that I can tell).

Do you think what I said was wrong? How does saying that Nvidia has about 80% market share over AMD change all the potential uses for HBM? I'm not following, sorry.

 

[Keysplayr]

Because you know something about it..?..

Tell us, we are all ears...



So i guess that Nvidia does even more poorly since they are constantly copying what their competitor do first, they may make more money but technologicaly they are not up to AMD...

How happen that the so called underdog is the most innovative.?.

Exactly my point, Abwx. I'm smart enough to know, that I actually don't know, and can admit that, while you sit there and spout off about things you don't know at all. Just guessing by your own admission but attempt to pass off as more than just your guess or opinion, but fact.

So why don't you tell US all about it. About the language in the said contracts between AMD and Hynix. We need dates, details, specifics. NOT GUESSES. They are worthless when it comes to this. Guessing what is inside a contract is absolutely hilarious.

 

[Headfoot]

I don't believe AMD should be put down. I believe the x86 license should be transferable, and that AMD should be bought out so someone else can have a chance at competing.

It's pretty irrelevant. And the licenses might be transferable to a purchaser, we don't know. There are plenty of merger and acquisition models that leave the acquired entity as the surviving entity so AMD would still exist as a holding company/operating company, etc. Unless there's a "here's how it works when you get acquired" (m&a) clause they can work around it pretty easy. Most likely there is a m&a clause in the license given how big a deal it is

But even that isnt dispositive. Anyone who bought AMD would have absolutely no trouble getting licensed for x86 tech from Intel because the reason Intel did that in the first place was as a cross licensing deal in order to get access to AMD's patents on microprocessor tech. And AMD did it for the same reasons, to get access to Intel tech. All the same market forces which compelled the original cross license still exist and still would exist if they got bought. That license is in reality a total non issue. Not to mention Intel is 100% sure to be gun shy regarding anti trust action after they had to pay out more than a billion. Big tech operating companies cross license with each other so they can do business. It's small, non producing licensing companies like (formerly) Rambus that you have to watch out for lawsuits.

tl;dr; there are a multitude of reasons why an amd sale would result in a full licensing agreement with amd's purchaser, and very few reasons why it wouldnt.

An AMD purchaser very well may not be interested in the x86 market however, and that is another issue entirely.

The market may not be true capitalism, but it sure isn't because of Intel's IP licensing strategy.

 

[Abwx]

Exactly my point, Abwx. I'm smart enough to know, that I actually don't know, and can admit that, while you sit there and spout off about things you don't know at all. Just guessing by your own admission but attempt to pass off as more than just your guess or opinion, but fact.

So why don't you tell US all about it. About the language in the said contracts between AMD and Hynix. We need dates, details, specifics. NOT GUESSES. They are worthless when it comes to this. Guessing what is inside a contract is absolutely hilarious.

I m talking of the fact that it is patented, the contract between Hynix and AMD is another matter, that said it s not difficult to guess that AMD will have the priority...

That said i guess that you are worried that AMD could get a definitive lead, and indeed that would be fair that they get back some marketshare, at some point their innovation will allow the GPUs to keep evoluting and the gamers will benefit from it, including you because if it was Nvidia you would see nothing at all other than moves to rak in more profits at the expense of the consumer..

Of course this doesn not apply for whom manage to get free gear in exchange of his support in various forums...

 

[tential]

It's pretty irrelevant. And the licenses might be transferable to a purchaser, we don't know. There are plenty of merger and acquisition models that leave the acquired entity as the surviving entity so AMD would still exist as a holding company/operating company, etc. Unless there's a "here's how it works when you get acquired" (m&a) clause they can work around it pretty easy. Most likely there is a m&a clause in the license given how big a deal it is

But even that isnt dispositive. Anyone who bought AMD would have absolutely no trouble getting licensed for x86 tech from Intel because the reason Intel did that in the first place was as a cross licensing deal in order to get access to AMD's patents on microprocessor tech. And AMD did it for the same reasons, to get access to Intel tech. All the same market forces which compelled the original cross license still exist and still would exist if they got bought. That license is in reality a total non issue. Not to mention Intel is 100% sure to be gun shy regarding anti trust action after they had to pay out more than a billion. Big tech operating companies cross license with each other so they can do business. It's small, non producing licensing companies like (formerly) Rambus that you have to watch out for lawsuits.

tl;dr; there are a multitude of reasons why an amd sale would result in a full licensing agreement with amd's purchaser, and very few reasons why it wouldnt.

An AMD purchaser very well may not be interested in the x86 market however, and that is another issue entirely.

The market may not be true capitalism, but it sure isn't because of Intel's IP licensing strategy.

In that case, if it can be transfer then I really don't care even more about what happens to amd. If they make idiotic moves then I hope they suffer the consequences. Let their stock plummet if they can't run a business and let someone with actual intelligence purchase a controlling number of shares.

 

[Sabrewings]

Everyone and his dog knows AMD engineered HBM and holds several patents for the memory. A 0.2 second Google search turned up this.

http://www.google.com/patents/US20140089609
http://www.google.com/patents/US8922243
http://www.google.com/patents/US20140181387
http://www.google.com/patents/WO2014025676A1

That doesn't matter. What does matter is that AMD/SK Hynix submitted HBM to JEDEC as a standard. They can hold all the patents they want, but since they're also a part of JEDEC (required as they use patents held by many other companies within JEDEC, including Nvidia) they have to license it fairly.

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..

That's not how that went. JEDEC invited Rambus to participate, but since they didn't agree with JEDEC's patent policy they left. Rambus received their royalties as assigned by a court because the JEDEC standard violated the patents of that non-JEDEC member.

You are correct that it isn't free among JEDEC members. However, JEDEC policy states that it must be fair under RAND. Thus, unless AMD wants to shoot themselves in the foot and get booted from JEDEC (meaning that a lot of their patent royalty costs will go up tremendously) they have to license it fairly to anyone who wants to use it within JEDEC.

Since AMD is forced (under JEDEC policy) to disclose all essential patents (and HBM could easily be argued to be an essential patent of the future), they have no choice in the matter.

 

[beginner99]

Why restrict HBM to dGPU only? HBM can be used in every other segment I would imagine. Mobile. Professional. Commercial/Compute. Super computing. Automotive.

Because it's way more expensive than DDR/GDDR. So for sure a no-go in mobile/automotive also because mobile does not need the high bandwidth anyway.

Of course it will be in more products but such products are years out, Zen APUs in 2017 will probably be the first non-dGPU using it. Meaning demand will be low for years to come. So no, I don't see a huge earning potential now, but yeah maybe in 3-4 years demand will rise, costs will get down. Still, till you need that kind of bandwidth in mobile, it will be a decade easily.

 

[Abwx]

That doesn't matter. What does matter is that AMD/SK Hynix submitted HBM to JEDEC as a standard. They can hold all the patents they want, but since they're also a part of JEDEC (required as they use patents held by many other companies within JEDEC, including Nvidia) they have to license it fairly.

No one said that they dont have to licence it that s not the point..

That's not how that went. JEDEC invited Rambus to participate, but since they didn't agree with JEDEC's patent policy they left. Rambus received their royalties as assigned by a court because the JEDEC standard violated the patents of that non-JEDEC member.

I used Rambus as exemple even if it is not a typical case, we know that Rambus entered in the Jedec just to troll the whole thing, that is, to patent future implementations still under discussion when they joinded this organ...

You are correct that it isn't free among JEDEC members. However, JEDEC policy states that it must be fair under RAND. Thus, unless AMD wants to shoot themselves in the foot and get booted from JEDEC (meaning that a lot of their patent royalty costs will go up tremendously) they have to license it fairly to anyone who wants to use it within JEDEC.

Since AMD is forced (under JEDEC policy) to disclose all essential patents (and HBM could easily be argued to be an essential patent of the future), they have no choice in the matter.

It is obvious that there are already licencees, Samsung has undoubtly bought a licence, otherwise they wouldnt announce starting production in 6 months.

As for licence fees they are generaly quite low, either you manage to get your idea largely used and you ll get some money or you wont recoup the RD cost, in any case the cost for the licence is negligible for firms like Samsung or Intel, i dont have numbers at hand but for DDR RAM it s something like 0.6%..

 

[Keysplayr]

Because it's way more expensive than DDR/GDDR. So for sure a no-go in mobile/automotive also because mobile does not need the high bandwidth anyway.

Of course it will be in more products but such products are years out, Zen APUs in 2017 will probably be the first non-dGPU using it. Meaning demand will be low for years to come. So no, I don't see a huge earning potential now, but yeah maybe in 3-4 years demand will rise, costs will get down. Still, till you need that kind of bandwidth in mobile, it will be a decade easily.

Every new tech is expensive when it first arrives. Over time it levels out just as every other tech has. Look at fast DDR4. Mucho bucks right now for the speeds that actually shows performance gains over DDR3. That'll come down also in time.

 

[Keysplayr]

I m talking of the fact that it is patented, the contract between Hynix and AMD is another matter, that said it s not difficult to guess that AMD will have the priority...

That said i guess that you are worried that AMD could get a definitive lead, and indeed that would be fair that they get back some marketshare, at some point their innovation will allow the GPUs to keep evoluting and the gamers will benefit from it, including you because if it was Nvidia you would see nothing at all other than moves to rak in more profits at the expense of the consumer..

Of course this doesn not apply for whom manage to get free gear in exchange of his support in various forums...

What difference does it make if it is patented? Royalties can be paid and use can be made of the tech. What is the issue here?

 

[Despoiler]

There is a lot more to consider with HBM than what's covered in the spec. The spec is how to build the physical product. It covers logic, but not logic strategies/optimization, optimal component placement, interposer design, packaging, or component validation. The component validation strategy alone is a licensable product. Do you test every single component? Do you package and then test? Do you use a hybrid approach? What and when you test is a giant question that if done intelligently has the opportunity to save a lot of money.

 

[Abwx]

What difference does it make if it is patented? Royalties can be paid and use can be made of the tech. What is the issue here?

There s no issue, whoever want to manufacture the thing using this methodology will pay some bucks to the patent holders, and as said fees are negligible compared to the RD cost.

A single licencee along with AMD s own production wont be enough to amortize the RD costs, one way or another they knew from the start that broadly licencing the tech was mandatory..

As for the induced cost it should be 1-2$ for a 500$ card, that s about negligible in respect of the manufacturing induced savings.

Well, dont worry, you ll have Nvidia HBM equipped cards next year, just that AMD will also be first to HBM2 and will have a few months advantage as they can use experimental HBM2 dies with their next GPU without having to wait for the process being validated, that s essentialy what they did with Fury.

 

[frozentundra123456]

You'd think. I've lost count how many times certain individuals have posted that Nvidia is not a charity and their hard work and money should not be expected to benefit AMD. So the reverse must be true.

You are missing the point. The ironic part is that posters who continually complain about nVidias supposedly anti-competitive practices suddenly think it will be a great thing if AMD does it.

 

[tential]

You are missing the point. The ironic part is that posters who continually complain about nVidias supposedly anti-competitive practices suddenly think it will be a great thing if AMD does it.

The irony of the situation is utterly hilarious.
Trying to explain to them reality of the situation is not.....

 

[Keysplayr]

There s no issue, whoever want to manufacture the thing using this methodology will pay some bucks to the patent holders, and as said fees are negligible compared to the RD cost.

A single licencee along with AMD s own production wont be enough to amortize the RD costs, one way or another they knew from the start that broadly licencing the tech was mandatory..

As for the induced cost it should be 1-2$ for a 500$ card, that s about negligible in respect of the manufacturing induced savings.

Well, dont worry, you ll have Nvidia HBM equipped cards next year, just that AMD will also be first to HBM2 and will have a few months advantage as they can use experimental HBM2 dies with their next GPU without having to wait for the process being validated, that s essentialy what they did with Fury.

Why do you choose to pull numbers out of thin air? Like I said before, and this is the last time I'll say it to you..... You don't know what you're talking about. You have no data. Just guesses. Stop. Please.
Thanks.

 

[Keysplayr]

You are missing the point. The ironic part is that posters who continually complain about nVidias supposedly anti-competitive practices suddenly think it will be a great thing if AMD does it.

I know right? They think we don't notice or something?

 

[Keysplayr]

The irony of the situation is utterly hilarious.
Trying to explain to them reality of the situation is not.....

Oh yeah that reminds me of that saying.

It's like when you're dead.
You don't know you're dead.
It's only difficult for others.
It's the same when you're stupid.

Not calling anyone stupid here.

 

[tential]

Why do you choose to pull numbers out of thin air? Like I said before, and this is the last time I'll say it to you..... You don't know what you're talking about. You have no data. Just guesses. Stop. Please.
Thanks.

Where are the assumptions amd will hit hbm2 first coming from? Why is that even a bonus of who hits hbm2 first.

Amd was first with hbm. The best card we recommend is a 390 a rebrand from awhile ago.

If hbm is so important why didn't it boost amd performance above nvidia this generation?

 

[Abwx]

Why do you choose to pull numbers out of thin air? Like I said before, and this is the last time I'll say it to you..... You don't know what you're talking about. You have no data. Just guesses. Stop. Please.
Thanks.

If you want to contradict me then show your own estimations, i guess that it s easy to criticise but it s much harder to play the game, that is , take some risks rather than staying out of the playing field.

Essentialy you have no other "arguments" than saying that i know nothing, if that s the case then your posts show that you know even less...

Where are the assumptions amd will hit hbm2 first coming from? Why is that even a bonus of who hits hbm2 first.

There are people on the know at SA, but i guess that most here wont register there by fear of being ridiculed...

Word is that HBM2 is for late 2016, mark this sentence if you want...

 

[Azix]

What really gets me is that all of these people who are praising hbm, saying it's amazing and that we need to buy amd products because of hbm. That we should hope amd has some type of exclusivity with hynix despite it clearly shown that amd will be a small part of hbm production.

None of them have an hbm product but they are specifically talking about how amazing hbm is and how we need to support amd since they were the first to make it.

If hbm is so amazing why don't any if them have it.....

who is saying we should go out and buy it?

You are missing the point. The ironic part is that posters who continually complain about nVidias supposedly anti-competitive practices suddenly think it will be a great thing if AMD does it.

you fail to appreciate the difference. If AMD were to use HBM to somehow reduce the performance of nvidia cards, that would be anticompetitive. Inventing something with no adverse effects on the competition and taking advantage of that innovation, is not anti-competitive.

 

[Piroko]

You are missing the point. The ironic part is that posters who continually complain about nVidias supposedly anti-competitive practices suddenly think it will be a great thing if AMD does it.

I agree. Nothing to be won here.

Where are the assumptions amd will hit hbm2 first coming from? Why is that even a bonus of who hits hbm2 first.

Amd was first with hbm. The best card we recommend is a 390 a rebrand from awhile ago.

If hbm is so important why didn't it boost amd performance above nvidia this generation?

A couple of years ago ATI pushed ahead with GDDR4. They didn't realize any significant gains with that either iirc. A year later though the fairly similar GDDR5 was released and AMDs transition to that standard was much smoother than Nvidias. Memory controllers certainly aren't easy to develop and a head start on that can be worth a lot. It took Nvidia until Kepler to catch up and that was one of the strongest market positions AMD has had.

 

[Keysplayr]

If you want to contradict me then show your own estimations, i guess that it s easy to criticise but it s much harder to play the game, that is , take some risks rather than staying out of the playing field.

Essentialy you have no other "arguments" than saying that i know nothing, if that s the case then your posts show that you know even less...



There are people on the know at SA, but i guess that most here wont register there by fear of being ridiculed...

Word is that HBM2 is for late 2016, mark this sentence if you want...

Tential. He just claimed Semi-Accurate as a resource for his information.

He doesn't need education
He claims citations aren't needed
And when he is cornered and is pressed for a source, he says SA.

El fini

 

[5150Joker]

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.

Oh you mean like how HBM 1 did for Fury X? :hmm: