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

You mean like how AMD is making a killing with the 460???
Oh wait it is the CPU division that helped save AMD both in terms stock price and returning to profit due to the margins.

 

We're not talking GPUs. That 45W+ firestarter with EMIB is niche beyond niche and will be widely used only by Apple.
Calm down already.
Its merely another way to paint GPU market red.

 

You said their focus should be volume...
I gave you a nice example of a volume sale product that is well known, it did not help in the way you suggest relative to what else was happening within AMD.
Nvidia makes a lot of their consumer profit from the 970/1070/1080, these are far from volume products and with a great profit margin ratio.
For mobile they also have nice revenue for their performance models again due to margins.

So yeah I do feel the Radeon division has sort of cut a strategic market off from the CPU Division because lower price volume GPU deal is not where the profit and sustainability is for AMD as a whole.

 

Its a product that competes with iGPUs in its purpose.

Unfortunately, it's 1050/1050ti/1060 And sometimes 1070.
Please stop screaming murder at the very basic semi-custom deal or take your pills already.

 

I actually do not understand why Intel is choosing to do this, instead of spinning their own, more powerful, Gen GPU?
One thing I can imagine is a customer request (e.g. Apple).

 

There's no denial, it doesn't matter to me which it is.
Intel has only confirmed it's 8th gen Core H -series, nothing about Kaby Lake-H. Intels 8th gen Core includes both Kaby and Coffee Lake chips

 

I'm not sure that's a prize I would want to win...

It's possible. On thinking about EMIB's ubump interface further, I wonder if that portion of the GPU needs significant customization. The idea would be that the HBM interface would treat the bridge silicon as if it were like any other interposer. The AMD GPU could use a similar or the same HBM2 interface, and with Intel managing the packaging AMD would only need to care about providing an otherwise unremarkable discrete GPU.
If using PCIe or a modified version, that's another section of the GPU that wouldn't need to change much. That would further isolate the GPU from Intel's tech, and simplify what AMD's semi custom engineering would have to handle.

AMD's revenue is split more along the lines of client and enterprise/semi custom.
If this falls under semi custom, it bolsters the segment that is a big target for AMD's long-term growth.
The client side has seen success with Ryzen, the GPU group is whatever from the x86 standpoint, and non-console APUs just don't seem to be garnering much affection.
If AMD's proposed plans are to be believed, the monolithic APUs are going the way of the dodo anyway.

This chip might be for most purposes internally like a special model of discrete, and if semicustom someone else paid for the NRE and is doing all the work in managing the packaging and supply chain. It could be outside AMD's ability to manufacture something along those lines given its likely market penetration, someone paying for the development, and packaging tech.

Without EMIB or a vastly improved business case for a product using interposer/chiplets, I am curious if AMD can make something practical given its limited resources and internal demand.

 

And that is a good one.
AMDs graphics solution is a proven one. It will work, within reasonably well established boundaries. Nobody says this marriage will last forever, it is a product almost worth buying from a purely industry mile stone point of view.

 

Source for Intel's 10nm being "trash", please...

AMD spent resources re-engineering polaris for HBM interface when they already have vega?

...Why?

You need something more concrete than a geekbench entry to back that claim up.

 

I think cheking semiwiki and looking at production targets will do?

You do understand that this design goes back for quite some time?
Oh boy the denial is sweet.

 

Probably zero research and development on Intel's part other than the substrate itself, which is established tech at this point (and supposedly if not low cost, so at least cost-effective). So, really rather cheap on the whole.

They slap on kabys or coffees straight off the production line, some minor modifications by AMD to integrate into Intel's side-bus interface thing, and...done.

 

If this is a semi custom division product, the selling point is that the customer pays for the engineering work. The downside would be that the actual margins of the produced silicon would be relatively poor. However, a likely benefit here is that AMD wouldn't be trying to balance production/inventory between HBM and non-HBM APUs, given its generally bad history at such management and the iffy balancing point its APUs exist at.

The customer would take on the particulars of selling/writing down a given batch.

There would be a cost in terms of organizational bandwidth (only so many engineers for everything, so much silicon going in all directions, etc.), which might be harder to quantify from the outside.

 

It is, Intel calls it "custom-to-Intel".

 

i think because intel gpu have bad aspect on area size and scaling performance(performance improvement only around 1.5x from GT2 HD 620 to GT3e iris plus 650).


from 1 to 2 processor graphics(192 to 384 core) intel add 51mm2 area
polaris radeon rx560 1024 core only 123mm2 on area size.
intel 960 core will have around 255mm2(5*51), over 2 times compare 1024 core rx560.

 

The revenue may be reported like that but from a division and VP perspective success is based upon one's product-margins/project contracts-IP-sales-market reached, internally it looks like this deal would go back to the Radeon division and VP.
Just going by experience working with a few international VPs and divisions in a few large tech manufacturing companies.
Each division is competing on budget/resources/success, and critically future planning; the balance is between what is good for the division and what is good for the business as a whole (CEO hopefully managing well the VPs and their strategies).

 

This may be the case, although the x86 group's future seems rather secure given the markets the Zen core is gaining share in.
If this deal is mostly about the monetary upside, and AMD's position in terms of integration technology and manufacturing isn't given a leg up by Intel's input, then this might be treated as a chance at getting cash that AMD wouldn't be able to get at anyway.

The encapsulation of the product makes it behave like a discrete GPU, albeit a more involved version of the Apple-centric Tonga discrete--particularly if Apple is one of the drivers for this.
AMD's logistics and integration tech seem like they aren't up to the task for a non-custom APU at this level of complexity and cost.
If AMD's integration/manufacturing doesn't improve, then at some point the x86 group could just assume that it will not be a factor in this particular space--and AMD overall is going to get what income it can.

 

Yeah, this deal only makes sense if the chips are going to be used exclusively by Apple, otherwise AMD would be shooting its CPU division in the head.
And it also shows that Apple doesn't care about their remaining "pro" users, who are asking for Nvidia GPUs since most pro apps use CUDA. I guess that most pro users have already jumped ship anyway.

 

Since Intel is paying for the chip, AMD might not have too much say in where the final assembly is sold. It does seem like it's aligned with Apple's specific balance of disproportionately using AMD GPUs and yet being invested in Intel's CPUs, though.

Even if not exclusive, it would be a product that would need to have similar price, performance, and power needs that AMD's currently disclosed APU solutions do not mesh with.

 

I find the relationship realy intresting. What is Intel doing when AMDs own APUs are stronger than Intel ones? AMD for sure knows now what APU Intel will bring! Intel is not knowing what AMD will throw into the APU market.

 

what Intel said:

And its family members are:
KBL-R like i5-8250U, Coffee Lake (desktop only?) and mobile Cannon Lakes. Cannon Lake H will come next year.