Rumor: AMD gets priority advantage to Hynix HBM2

[Tuna-Fish]

Patents don't work that way at all at this level. There are thousands and thousands of patents that nV holds that could be used to block anyone else from making GPUs. AMD also has equivalent patent weapons. What this means is that they both just use each other's patents without care. Effectively, everyone cross-licenses everything at nominal cost. Patents are only used to block new players from entering the field.

 

[Tuna-Fish]

Also, jedec's patent policy.

 

[Silverforce11]

We don't know the inner workings so just stop pretending or claiming with certainty. We can only speculate for fun*.

* It would be funny if AMD gets first dibs on HBM2 and stall Pascal because that's AMD playing by NV's books.

From the patent text:

FIG. 1 illustrates a perspective view of a first multi-chip module implementing physical memory according to some embodiments. Multi-chip module 100 generally includes a multi-core processor chip 120 and a memory chip stack 140. Memory chip stack 140 includes a plurality of memory chips stacked on top of each other. As illustrated in FIG. 1, memory chip stack 140 includes a memory chip 142, a memory chip 144, a memory chip 146, and a memory chip 148. Note that, in general, memory chip stack 140 may include more or fewer memory chips than illustrated in FIG. 1. Each individual memory chip of memory chip stack 140 is connected to other memory chips of memory chip stack 140, as desired for proper system operation. Each individual memory chip of memory chip stack 140 also connects to multi-core chip 120, as desired, for proper system operation.

FIG. 2 illustrates a perspective view of a second multi-chip module 200 implementing physical memory according to some embodiments. Multi-chip module 200 generally includes an interposer 210, a multi-core processor chip 220, and a memory chip stack 240. Interposer 210 is connected to the active side of multi-core chip 220. Memory chip stack 240 includes a plurality of memory chips stacked on top of each other. As illustrated in FIG. 2, memory chip stack 240 includes memory chip 242, memory chip 244, memory chip 246, and memory chip 248. Note that, in general, memory chip stack 240 may include more or fewer memory chips than illustrated in FIG. 2. Each individual memory chip of memory chip stack 240 is connected to other memory chips of memory chip stack 240, as desired for proper system operation. Each individual memory chip of memory chip stack 240 is also connected to multi-core chip 220, as desired for proper system operation. In some embodiments, memory chip stack 240 includes a single memory chip. In some embodiments, the multi-chip module 200 includes more than one memory chip stack like memory chip stack 240.

With talks of memory chip stacks that link with each other on an interposer...

Smells like HBM MC.

@Sabrewings
If NV isn't doing ECC, they aint viable for Teslas.

 

[Keysplayr]

Unless they come up with their own unique MC for HBM (that won't infringe AMD's patents), they may have to do that.

Imagine that, each NV SKUs sold, AMD earns some $.

Would be the first smart thing AMD did in a LONNNG time. Are you going to call this dirty? Or smart?
Jeopardy music.

 

[Keysplayr]

@Sabrewings
If NV isn't doing ECC, they aint viable for Teslas.

Not sure Teslas have, or ever will (for the forseeable future) need such high bandwidth memory. They can probably use GDDR5 for years to come. I think the amount of memory is more important and the GPU compute processing power is more important on Tesla products.
And Nvidia won't need ECC for GeForce GPUs obviously.

So. Disaster averted? Or are we all wrong to take this information the way we took it?

 

[Flapdrol1337]

This rumor seems complete nonsense to me, they invested in it and put it in jedec, they want it to be used by others, so they get the jedec license fees. Nvidia may be the competition but probably also the only ones interested for now.

Hynix will want to sell to nvidia too, why would they agree to an exclusivity deal? Amd doesn't have the money to do this kind of trolling.

 

[Wall Street]

I could see them agreeing to this because it looks like they had exactly one customer for HBM 1. It is always valuable to get someone to step up to be the guinea pig, to work through the teething pains, and to act a proof that HBM is a viable product. AMD probably saw they were the only one asking for HBM 1, so they were in a position to negotiate to get both HBM 1 and HBM 2 rights.

 

[cytg111]

Omg AMD exclusive on HBM2, HBM2 on Zen, Zen kicks Intels arse. Lotsa lolz.
Later

 

[Silverforce11]

Would be the first smart thing AMD did in a LONNNG time. Are you going to call this dirty? Or smart?
Jeopardy music.

It's smart and dirty at the same time.

If AMD actually manage to leverage their MC patent, they are basically sidelining JEDEC rules and obligations. ie. RAM spec is open but the method to use it isn't.

 

[twjr]

Would be the first smart thing AMD did in a LONNNG time. Are you going to call this dirty? Or smart?
Jeopardy music.

Why do they have to be mutually exclusive? Many smart business decisions would seem dirty from a competitors perspective.

 

[Silverforce11]

Not sure Teslas have, or ever will (for the forseeable future) need such high bandwidth memory. They can probably use GDDR5 for years to come. I think the amount of memory is more important and the GPU compute processing power is more important on Tesla products.
And Nvidia won't need ECC for GeForce GPUs obviously.

So. Disaster averted? Or are we all wrong to take this information the way we took it?

Yes, NV should stick with GDDR5 for years to come. It will definitely be an advantage for them.

ps. I've said it in the past many times, GameWorks is a brilliant/smart move, but a very dirty move. Because I have a self-interest as a PC gamer, I don't want companies to compete via closed source software and bribing devs to implement stuff that cripples competitor hardware performance. The reason is plain obvious for anyone with a semblance of logic, if AMD, Intel & NV all behave dirty and push GW-like features, we'll get mostly AAA titles that are poorly optimized, all of which run even worse if you don't have the correct brand/generation of hardware. Or the simple thing: I pay $60 for a game, but its got 4-5 features I can't actually enjoy because I don't have the correct brand/gen of hardware, so I get LESS value out of that game for the same money paid. Compare that to open source features where all the vendors can optimize it easily and we can all enjoy great looking & performing games. Clearly one scenario benefits PC gamers much more than the other.

 

[railven]

Why do they have to be mutually exclusive? Many smart business decisions would seem dirty from a competitors perspective.

The question is legit, and comical, because of Silver's stance on "dirty" tactics Nvidia employs. Haha.

AMD blocking Nvidia to HMB2 would be a good move, but something tells me people will still wait for Nvidia cards. Haha.

 

[Keysplayr]

Yes, NV should stick with GDDR5 for years to come. It will definitely be an advantage for them.

For Tesla. GeForce would require HBM sooner than later. I think FuryX 4K performance is a testament to this.

 

[Vesku]

AMD invests and assists in the development of Hynix version of HBM. Has to deal with half the expected chip memory capacity and low yields. Ends up being the sole customer for HBM1, helping sustain development of much more useful HBM2.

Given this why is having priority on initial Hynix HBM2 production "dirty"? If they don't have priority I'd see that as a big snub from Hynix.

 

[frozentundra123456]

Yes, NV should stick with GDDR5 for years to come. It will definitely be an advantage for them.

ps. I've said it in the past many times, GameWorks is a brilliant/smart move, but a very dirty move. Because I have a self-interest as a PC gamer, I don't want companies to compete via closed source software and bribing devs to implement stuff that cripples competitor hardware performance. The reason is plain obvious for anyone with a semblance of logic, if AMD, Intel & NV all behave dirty and push GW-like features, we'll get mostly AAA titles that are poorly optimized, all of which run even worse if you don't have the correct brand/generation of hardware. Or the simple thing: I pay $60 for a game, but its got 4-5 features I can't actually enjoy because I don't have the correct brand/gen of hardware, so I get LESS value out of that game for the same money paid. Compare that to open source features where all the vendors can optimize it easily and we can all enjoy great looking & performing games. Clearly one scenario benefits PC gamers much more than the other.

Or you could look at it as extra features that nVidia users can use that would not have been developed at all had not nVidia provided the resources. As long as gameworks features can be turned off on AMD cards, I dont see how extra features for their cards takes away anything from AMD users. If AMD want some exclusive features, let them develop them. Now would it be "nice" if nVidia developed the features to run on both cards? Of course, but I dont see that they have any obligation to do so.

 

[Headfoot]

Pretty sure AMD and Nvidia already have cross licensing agreements for graphics.

 

[xthetenth]

The only customers hurt by such a move would be the ones who refuse to consider buying AMD, and frankly, I don't care about them. Brand loyalists distort the market, harm the mechanisms that provide benefits to other customers, and encourage problematic decisions by companies. I feel no solidarity with people working against my (and their) long term interests.

Probably moot though.

 

[Stuka87]

Pretty sure AMD and Nvidia already have cross licensing agreements for graphics.

For GPU patents they do. But memory is not part of the GPU. Its a another component that can be used on GPUs.

 

[Genx87]

You guys are dreaming if you think Nvidia cant design a MC for HBM2.

 

[Unoid]

Or you could look at it as extra features that nVidia users can use that would not have been developed at all had not nVidia provided the resources. As long as gameworks features can be turned off on AMD cards, I dont see how extra features for their cards takes away anything from AMD users. If AMD want some exclusive features, let them develop them. Now would it be "nice" if nVidia developed the features to run on both cards? Of course, but I dont see that they have any obligation to do so.

hows that PS4 rain effect in batman vs the gameworks(TM)(R)(BS) working out better for nvidia users?

gameworks is crap for all gamers.:whiste:

edit: Referencing: http://www.eurogamer.net/articles/d...rkham-knight-pc-lacks-console-visual-features

If the PC Port people would have just taken the PS4 version and straight ported it to DX11, with adding pc specific option sliders, everything would have been awesome. No, Nvidia had to pay them to use gameworks to make the PC port AWESOME!!111.

 

[Stuka87]

You guys are dreaming if you think Nvidia cant design a MC for HBM2.

Has nothing to do with if they can or cannot. They would have to do it without infringing on AMD's patent, that or license the patent from them.

 

[Azix]

Or you could look at it as extra features that nVidia users can use that would not have been developed at all had not nVidia provided the resources. As long as gameworks features can be turned off on AMD cards, I dont see how extra features for their cards takes away anything from AMD users. If AMD want some exclusive features, let them develop them. Now would it be "nice" if nVidia developed the features to run on both cards? Of course, but I dont see that they have any obligation to do so.

Not true and definitely not better that way.

eg. Say witcher 3 devs used tressfx instead of hairworks. That experience would last them for their next game, their expansions etc. They could freely use that tech anywhere, anytime. Now they have to go to nvidia for permission everytime and they don't even have freedom with the tech.

The same with Physx. Havok is better, bullet is better. Tons of games have solid physics without physx. Same with shadow tech, PCSS does not look better than softshadows and performs worse than AMDs CHS and Softest shadow setting

http://www.hardocp.com/article/2015...video_card_performance_preview/6#.Vae_WvlVhBc

In reality its less performance for inferior tech.

The contrast here is that one is genuine hardware innovation and a huge boon for gamers. The other is just bloatware pretending to be added value.

 

[Silverforce11]

Or you could look at it as extra features that nVidia users can use that would not have been developed at all had not nVidia provided the resources. As long as gameworks features can be turned off on AMD cards, I dont see how extra features for their cards takes away anything from AMD users. If AMD want some exclusive features, let them develop them. Now would it be "nice" if nVidia developed the features to run on both cards? Of course, but I dont see that they have any obligation to do so.

Why can't NV play nice like AMD, they develop the features, make it open, push forward great optimized features for all to enjoy?

As said, only one scenario benefits PC gamers. GW doesn't even benefit NV gamers, because its horribly optimized. A frequently complaint on the steam forum is from NV owners why they can't get great smooth performance with it enabled.

If NV make it open, devs would be better able to optimize it themselves for all hardware. So instead of a major performance hit, you could have a very light hit like TressFX. Now all gamers can enjoy.

 

[tviceman]

Omg AMD exclusive on HBM2, HBM2 on Zen, Zen kicks Intels arse. Lotsa lolz.
Later

This post wins the thread.

As others have said, if AMD has first exclusivety, then it will be short-lived. If Nvidia is many months ahead of AMD's next-gen launch, then I don't see how Hynix sits on their hands losing money, unless AMD signed some sort of agreement with Hynix similar to their GloFo write-downs where they pay for the chips whether or not the chips are needed.

 

[tviceman]

Not sure Teslas have, or ever will (for the forseeable future) need such high bandwidth memory. They can probably use GDDR5 for years to come.

I don't think it's a viable strategy to stick with GDDR5 because of the much improved efficiency HBM brings. Nvidia dominates dGPU notebook sales because of it's superior efficiency. Perhaps only the flagship Pascal die uses HBM, but I think after Pascal all chips will be using HBM, even the low end ones. Being able to stay comfortably ahead of Intel's Iris Pro is important for Nvidia as well.