Intel and AMD’s custom SOC has no IP sharing elements

Intel and AMD Radeon announced yesterday that they have been working together to create a CPU+GPU solution in a single package, facilitating the creation of ultra-compact gaming/productivity notebooks. 

As part of this deal, AMD’s semi-custom division has created a GPU that is designed to match Intel’s specifications which is allegedly based on the company’s Polaris GPU architecture, much like today’s PS4 Pro and Xbox One X consoles. In effect, this deal has AMD create a custom GPU solution for Intel, with no IP being shared between both companies. 

One of the most interesting parts of Intel’s design is the fact that the GPU section of Intel’s new product will not be using a silicon interposer, unlike any of AMD’s previous GPU+HBM products. Intel instead has decided to use their own EMIB technology as an alternative, greatly decreasing the complexity and cost of creating a GPU with an HBM2 memory stack. 

 

 

When Intel announced that they would be using an AMD graphics product, one of the biggest questions that arose was if AMD were able to license Intel’s EMIB technology, allowing them to create future GPU+HBM2 memory solutions without the need for a complex and costly silicon interposer. Sadly, this deal has no IP sharing or licensing elements, which means that AMD does not have access to this exclusive Intel technology. 

While EMIB would be hugely useful to AMD, it is understandable why Intel would not want to share this technology, especially given how beneficial it would be to AMD when creating other multi-die products. A faster interconnect for AMD’s Threadripper and EPYC CPUs would be bad news for Intel in the server market. 

With the manufacturing problems that AMD has been experiencing with their Vega 10 series products, AMD should be looking for an alternative to Silicon interposers, to allow them to create HBM2 GPUs with greater reliability and reduced cost. It seems that Intel’s EMIB is not the answer here, but that does not mean that AMD isn’t creating their own IP to solve the same problem.  

 

 

Intel and AMD Radeon announced yesterday that they have been working together to create a CPU+GPU solution in a single package, facilitating the creation of ultra-compact gaming/productivity notebooks. 

As part of this deal, AMD’s semi-custom division has created a GPU that is designed to match Intel’s specifications which is allegedly based on the company’s Polaris GPU architecture, much like today’s PS4 Pro and Xbox One X consoles. In effect, this deal has AMD create a custom GPU solution for Intel, with no IP being shared between both companies. 

One of the most interesting parts of Intel’s design is the fact that the GPU section of Intel’s new product will not be using a silicon interposer, unlike any of AMD’s previous GPU+HBM products. Intel instead has decided to use their own EMIB technology as an alternative, greatly decreasing the complexity and cost of creating a GPU with an HBM2 memory stack. 

 

 

When Intel announced that they would be using an AMD graphics product, one of the biggest questions that arose was if AMD were able to license Intel’s EMIB technology, allowing them to create future GPU+HBM2 memory solutions without the need for a complex and costly silicon interposer. Sadly, this deal has no IP sharing or licensing elements, which means that AMD does not have access to this exclusive Intel technology. 

While EMIB would be hugely useful to AMD, it is understandable why Intel would not want to share this technology, especially given how beneficial it would be to AMD when creating other multi-die products. A faster interconnect for AMD’s Threadripper and EPYC CPUs would be bad news for Intel in the server market. 

With the manufacturing problems that AMD has been experiencing with their Vega 10 series products, AMD should be looking for an alternative to Silicon interposers, to allow them to create HBM2 GPUs with greater reliability and reduced cost. It seems that Intel’s EMIB is not the answer here, but that does not mean that AMD isn’t creating their own IP to solve the same problem.