Intel Readies Adoption of AMD HBM Stacked RAM Tech

[ARMD Supremacy]

Development of a technology that Intel is to facilitate with adoption of HBM

Future of Intel's high-end GPU core built-in CPU is, in the current eDRAM not, there is a possibility that becomes equipped with the HBM (High Bandwidth Memory). In that case, not only the memory chip is changed, the amount of memory and use changes. Rather than 128MB, it is now put the memory number in GB is likely to be to be used as part of the main memory rather than cache.

Since HBM was created from R&D of AMD and Hynix, there can be an exchange of tech...such as Intel grants AMD use of CPU cache tech and utilization of 14nm foundries. From these tech exchanges, can true progress continue to be bourn to consumers.



http://pc.watch.impress.co.jp/docs/column/kaigai/20150203_686619.html

 

[NTMBK]

Since HBM was created from R&D of AMD and Hynix, there can be an exchange of tech...such as Intel grants AMD use of CPU cache tech and utilization of 14nm foundries. From these tech exchanges, can true progress continue to be bourn to consumers.

Yeah, no, that isn't going to happen. Hynix will sell some memory to Intel, the end.

Also, for the love of god please stop messing with font size and formatting.

 

[III-V]

I think they're more likely to use Micron's HMC, seeing as they're already doing that.

 

[ARMD Supremacy]

Yeah, no, that isn't going to happen. Hynix will sell some memory to Intel, the end.

Also, for the love of god please stop messing with font size and formatting.

Then the irony? When Intel or Nvidia use millions of their R&D, they make sure proprietary and expensive to consumers...CUDA, G-sync, Gameworks to ambush AMD users.

Also, sorry for fonts but direct copy paste from article and tried to keep to font size 3 for all paragraphs except title of article.

 

[Tsavo]

pfff. Stacked is old hat. I'm waiting for Parallel Arrays of Ram.

THAT, my friends, will be something.

 

[Arachnotronic]

 

[ShintaiDK]

I think they're more likely to use Micron's HMC, seeing as they're already doing that.

^^ This.

Stacked DRAM is...just stacked DRAM. Many names for the same thing.

 

[bronxzv]

Since HBM was created from R&D of AMD and Hynix,

why will Intel switch from HMC to HBM according to you ?

 

[Erenhardt]

^^ This.

Stacked DRAM is...just stacked DRAM. Many names for the same thing.

Yep. That is why after amd burned their $ for R%D for HBM, intel will adopt the tech and push their branding name with more $. Then everyone forgets about HBM and hail intel for coming up with HMC stacked dram and pushing the industry forward.

Even if you have tons of $, there is no reason to burn it if you don't have to

 

[ShintaiDK]

Yep. That is why after amd burned their $ for R%D for HBM, intel will adopt the tech and push their branding name with more $. Then everyone forgets about HBM and hail intel for coming up with HMC stacked dram and pushing the industry forward.

Even if you have tons of $, there is no reason to burn it if you don't have to

Your post wouldnt be so silly if you had done your homework.

Even SK Hynix is a HMC backer.

http://hybridmemorycube.org/about.html

The companies behind HMC is:
Altera
ARM
IBM
Micron
Open Silicon
Samsung
Semtech
SK Hynix
Xilinx

 

[ARMD Supremacy]

why will Intel switch from HMC to HBM according to you ?

Depending on applications, for routers and servers Intel may choose HMC, but for high volume mobile and semi-mobile apps HBM may win on cost and customization. Already, company called Thruchip demonstrate low cost method for HBM attachment to APUs/GPUs.

http://forums.anandtech.com/showthread.php?t=2416366 "Thruchip Demos Low Cost Method For Superfast HBM APUs/GPUs"

Customers like Sony, Nintendo and Microsoft will likely choose a semi-custom version HBM for next generation mobile and desktop products that market to 50 to 100+ million unit sales. Intel has design that is used in Microsoft Hololens product and next generation Hololens and similar products very likely to use HBM tech.

 

[ShintaiDK]

Who else but SK Hynix makes or will make HBM chips?

Vs this:

HBM needs a miracle to win over HMC. Even SK Hynix isnt willing to bet the house on it.

 

[rainy]

Who else but SK Hynix makes or will make HBM chips?
HBM needs a miracle to win over HMC. Even SK Hynix isnt willing to bet the house on it.

Quite obviously HBM is bad because it will be used by AMD, isn't?

 

[el etro]

Wide I/O(that is like a Low-power version of HBM) will be the main DRAM solution in the long run, but i don't see HMC and HBM being competitors. HMC is targeted at highest tier of computing solutions, while HBM can be on mobile solutions.

 

[ShintaiDK]

Quite obviously HBM is bad because it will be used by AMD, isn't?

Is that your entire rationale?

HMC:
SK Hynix, Micron, Samsung.

HBM:
SK Hynix

Did I miss anything?

 

[DownTheSky]

SK Hynix is BOSS.

 

[DeathReborn]

I think Intel might adopt HBM for Xeon Phi successors but not for CPU's.

OP, might want to add Hynix to the thread title, after all they did have a large part (more than AMD infact) in HBM's development.

 

[rainy]

Did I miss anything?

Yes, don't even try to pretend that you're an objective person - "unfortunately" you can't hide history of your posts.

 

[bronxzv]

http://forums.anandtech.com/showthread.php?t=2416366 "Thruchip Demos Low Cost Method For Superfast HBM APUs/GPUs"

the original ThruChip paper compares their solution to connect layers (TCI coils) with previous solutions (TSV), HBM is used merely as an example IIRC, no reason to not use TCI with an HMC interface isn't it ?

 

[senseamp]

Development of a technology that Intel is to facilitate with adoption of HBM



Since HBM was created from R&D of AMD and Hynix, there can be an exchange of tech...such as Intel grants AMD use of CPU cache tech and utilization of 14nm foundries. From these tech exchanges, can true progress continue to be bourn to consumers.



http://pc.watch.impress.co.jp/docs/column/kaigai/20150203_686619.html

HBM is a JEDEC standard, with corresponding IP commitments by the entities pushing it. Intel is not going to exchange its unencumbered technologies and production advantages for pledged IP it can get on FRAND terms.
The big question is price. HMC has multiple competing vendors, whereas HBM is a single vendor solution at this point. Not only is it single vendor solution, it's also doesn't have a large customer base aside from a small part of AMD's business at this point to amortize the R&D and fixed costs.
It is likely that AMD got favorable pricing terms from Hynix for collaborating on HBM, but Intel will look at whatever will get it the most favorable pricing, and it's likely to be multi-vendor multi-customer HMC, and not HBM.

 

[Phynaz]

JEDIC standard. You can download it here:

http://www.jedec.org/standards-documents/results/jesd235

There's nothing for AMD to "grant" to Intel.

 

[bronxzv]

JEDIC standard. You can download it here:

http://www.jedec.org/standards-documents/results/jesd235

There's nothing for AMD to "grant" to Intel.

as is Wide I/O 2 btw

http://www.jedec.org/standards-documents/docs/jesd229-2

there is a lot of functional overlap between all these interfaces (HBM, HMC, WIO2) apparently, I'll be glad to read an article contrasting all pros/cons of these

EDIT, just found this article: http://www.extremetech.com/computin...es-between-wide-io-hbm-and-hybrid-memory-cube

 

[Elixer]

pfff. Stacked is old hat. I'm waiting for Parallel Arrays of Ram.

THAT, my friends, will be something.

Parallel has failed to show vast speed gains in the past...

 

[Phynaz]

Exactly, that's why everything has moved to serialized interfaces. Clocking parallel interfaces is too slow.

 

[Shehriazad]

I don't care who adapts what...I just want to get rid of DDR3 and DDR4 asap...which is indirectly APU,CPU and GPU related...but if that kind of memory is able to showcase versatility and far superior performance...we could probably hope to finally move on from DDR as main memory, no?

No matter what Intel chooses to adapt (HMC, HBM...whatevs)...with Nvidia and AMD both swinging toward HBM and Intel also using either one of them...we can finally move on. xD No matter the outcome..I want to be able to push tiny memory blocks into my mboard...not some huge sticks belonging to an old standard that has only questionable performance gains over the years.