Intel's Nvidia GPU licensing deal ends next month

[HEXUS]

Meanwhile chitter chatter about a possible Intel / AMD GPU licensing deal intensifies.

Read more.

[darcotech]

Some are saying that the validity of Intel-Nvidia agreement is til 2019.

Why would AMD do such contract? Get some cash but lose on APUs sell?

[HW90]

Some are saying that the validity of Intel-Nvidia agreement is til 2019.

Why would AMD do such contract? Get some cash but lose on APUs sell?

The value of the contract probably outweighs the cost to APU sales they think they're going to incur.

[Wrinkly]

Amazing how well AMD works with other companies.

[SHSPVR]

Some are saying that the validity of Intel-Nvidia agreement is til 2019.

Why would AMD do such contract? Get some cash but lose on APUs sell?

I should point out to you is that APU is CPU and a GPU within a single chip die fab, What Intel has plan for is separate CPU and GPU chip within single die socket.

[nitro912gr]

I don't know why this is big news, amd and intel cross-license a lot of techs anyway.

[DanceswithUnix]

Some are saying that the validity of Intel-Nvidia agreement is til 2019.

Why would AMD do such contract? Get some cash but lose on APUs sell?

It would make AMD money on each of these chips that Intel sell, and it would squeeze Nvidia in the process.

These will be, even by Intel standards, an expensive chip. I don't see it competing with the cheap APUs that AMD make.

[watercooled]

'Licensing technology' from AMD does not by itself imply we'll see anything like Radeon GPUs on Intel processors - much like the existing Nvidia deal, it could simply mean licensing core technologies and/or implementations and business as usual for Intel. If the source is correct in claiming AMD is specifically working on a product for Intel, as opposed to misunderstanding and over-extrapolating a rumour, then that would be something different of course.

[DanceswithUnix]

If the source is correct in claiming AMD is specifically working on a product for Intel, as opposed to misunderstanding and over-extrapolating a rumour, then that would be something different of course.

That is the rumour, specifically AMD supplying a graphics die for Intel to use in a MCM package. So not integrated graphics like we have gotten used to, but presumably a laptop part where PCIe lanes are directly tied to a GPU.

[watercooled]

In a product like that I assume it would have to be a fairly large GPU for it to make sense, as Intel seem to be doing well enough on the hardware side with their smaller IGPs. Their drivers still leave a lot to be desired though IMO; speaking of which, I wonder who would be responsible for drivers in this case?

[CAT-THE-FIFTH]

In a product like that I assume it would have to be a fairly large GPU for it to make sense, as Intel seem to be doing well enough on the hardware side with their smaller IGPs. Their drivers still leave a lot to be desired though IMO; speaking of which, I wonder who would be responsible for drivers in this case?

I think if it does exist - it will be some form of Xeon with an AMD GPU for compute purposes.

[watercooled]

I think if it does exist - it will be some form of Xeon with an AMD GPU for compute purposes.

Wouldn't Intel have that covered with their Phi line though? Plus an integrated GPU would be quite limited in size and lack the scaling capability of having many discrete cards in one rack. Unless they have some very specific workload in mind (like AMD's HSA?) where roughly equally sized CPU and GPU parts are desirable? I suppose Intel could thrown in an Altera FPGA for good measure too!

[CAT-THE-FIFTH]

Wouldn't Intel have that covered with their Phi line though? Plus an integrated GPU would be quite limited in size and lack the scaling capability of having many discrete cards in one rack. Unless they have some very specific workload in mind (like AMD's HSA?) where roughly equally sized CPU and GPU parts are desirable? I suppose Intel could thrown in an Altera FPGA for good measure too!

They don't have anything to really attack the lower end of the Nvidia market,so by making such an SKU available,they have their bases covered. Remember the Xeon D CPUs?? Imagine one of those with an AMD dGPU onboard??

Edit!!

Plus for AMD,it might sound like they are reducing the sales of their own Pro series cards,but on the flip side,more people developing for OpenCL and other frameworks the more likely they can start to attack the entrenched CUDA framework.

OFC,the rumour could all be untrue and AMD is just licensing their tech for a lower price than Nvidia.

[nobodyspecial]

Amazing how well AMD works with other companies.

https://pipedot.org/article/2BF9W
Right up until a lawsuit like NV (which was their first by them). Whatever...

[DanceswithUnix]

Plus for AMD,it might sound like they are reducing the sales of their own Pro series cards,but on the flip side,more people developing for OpenCL and other frameworks the more likely they can start to attack the entrenched CUDA framework.

There have been rumours that the only reason Intel graphics is as good as it is comes down to Apple nagging at Intel to do better.

If Apple are still nagging, and recent Apple products have been switching to discrete AMD parts, then it makes sense to embed an AMD part to keep Apple sweet. That could make such a part a custom Apple device, not available to anyone else to buy.

Or, as you said, it could all be horse feathers and someone was pulling Kyles leg

I wonder who would be responsible for drivers in this case?

I would hope it would be AMD, but then in the past Intel have bought in graphics hardware and insisted on doing their own broken drivers that couldn't drive parts some of the pipeline making it unnecessarily slow as well as crashy.

If this is just an RX460 bolted onto the side of an i5 for laptop use, then the graphics and CPU PCIe lanes can be connected up but that leaves the question of how much ram can you put in an MCM package.

But back to Cat's comment, if this were an Apple centric device then it wouldn't help drive OpenCL at all. In fact, the Intel chips often score well in OpenCL benchmarks.