Intel Kaby Lake + AMD Radeon product *spin-off*

  • Thread starter Deleted member 13524
  • Start date
Already posted in the Kaby Lake thread, but it appears that Intel+AMD solution is coming after all

https://archive.fo/Gn813

GPU codename clearly points at something in the Vega family, and 4GB dedicated RAM.

If true, I'm thinking it's a MCM carrying a 4C+GT2 Kaby Lake + Interposer with Vega 11 + single-stack HBM2.

In one hand, it's a shame that AMD couldn't make their own Zen+Vega11 solution. On the other hand, the solution itself might be spectacular on efficiency metrics.

And I'll say right from the start that the origin for this alliance was most probably a special order from Apple for macbook pros and imacs.
 
But AMD does have their own Vega, Ryzen APUs, don't they?

With a dedicated HBM2 stack, the performance level of this GPU is clearly way above Raven Ridge's.


To me it looks like Apple just won't let go of using Intel for CPUs and AMD for higher-performance GPUs (i.e. they're not fully convinced with Zen's performance). And in the off-chance Apple would rather go ARM + nvidia for laptops and AiOs, Intel and AMD had to work together to give Apple what they wanted.
 
„… a custom-to-Intel third-party discrete graphics chip from AMD’s Radeon Technologies Group“
 
Where would this chip be fabbed? Somewhat ironic if @ Intel's (and would hope for more to come then)
 
There's a lot of responses on various threads. We need a dedicated thread for this.
 
Where would this chip be fabbed? Somewhat ironic if @ Intel's (and would hope for more to come then)

That Vega (probably-)11 is fabbed at either GF or TSMC. Probably GF.
I don't believe AMD wants their GPU in Intel's fabs, or Intel wants AMD in their fabs. Chips are built separately and a later fab (probably from Intel?) is used to placed them all into the EMIB.
 
Already posted in the Kaby Lake thread, but it appears that Intel+AMD solution is coming after all

https://archive.fo/Gn813

GPU codename clearly points at something in the Vega family, and 4GB dedicated RAM.

If true, I'm thinking it's a MCM carrying a 4C+GT2 Kaby Lake + Interposer with Vega 11 + single-stack HBM2.

In one hand, it's a shame that AMD couldn't make their own Zen+Vega11 solution. On the other hand, the solution itself might be spectacular on efficiency metrics.

And I'll say right from the start that the origin for this alliance was most probably a special order from Apple for macbook pros and imacs.

"Apple" was my first thought as well, for the first few million units anyway.
 
If we look back at this statement from Lisa Su, it seems the MCM will only be available during H2 2018:

http://www.barrons.com/articles/amd...nother-semi-custom-project-in-2018-1501022700
I asked if it’s too soon to talk about next semi-custom projects. She said there is “one additional design that will ramp in the second half of 2018,” and said AMD will wait until the customer is ready to discuss that project. I asked what industry it might be in, and Su replied “let’s just call it embedded."
 
Hoping for any company to just give away major tech/IP like EMIB is probably a bit too much; however I don't think Intel has said anything at all about licensing this stuff, when they could be making loads of dosh if they did. Smart devices today are super highly integrated, oftentimes these days they're bare silicon dies soldered straight onto PCBs because the density is too high for BGA packaging. Then there's the crazy stacked PCB used in iPhone X; EMIB could probably improve integration further still, not to mention reduce power usage.

Except, who knows if Intel feels like sharing? At least they're actually using this tech (in more than one product now even), unlike BBUL from years ago which never amounted to anything from what I recall.
 
Hoping for any company to just give away major tech/IP like EMIB is probably a bit too much; however I don't think Intel has said anything at all about licensing this stuff, when they could be making loads of dosh if they did. Smart devices today are super highly integrated, oftentimes these days they're bare silicon dies soldered straight onto PCBs because the density is too high for BGA packaging. Then there's the crazy stacked PCB used in iPhone X; EMIB could probably improve integration further still, not to mention reduce power usage.

Except, who knows if Intel feels like sharing? At least they're actually using this tech (in more than one product now even), unlike BBUL from years ago which never amounted to anything from what I recall.
EMIB is their only notable achievement in ages, doubly so considering their 10nm is sorta trash and is very late to the market.
I think they'll hold it thinking it will amount to something.
 
Back
Top