Nvidia CEO Jensen Huang confirmed today that Nvidia will contribute “GPU chiplets” that Intel can place alongside its x86 CPU cores instead of the Arc integrated graphics it develops in-house today.
They could buy AMD Radeon chiplets too, that would be something.:)
Intel actually bought AMD Radeon GPUs for their Hades Canyon (Kaby Lake G) platform. It was a NUC mainboard with a full Intel platform, combined with an AMD Radeon GPU. The Intel CPU and the GPU (including HBM2 memory for the GPU) was all on one package soldered to the mainboard.
I think they did a couple of follow ups on that as well, because it worked very well.
AMD’s GPUs were much faster than Intel’s, and making GPUs for this kind of application was something AMD already did. Nvidia didn’t, so would have to design a whole chip from scratch, and didn’t really have a power efficiency advantage (in recent generations where AMD’s desktop cards have run hot, it’s because they’ve been clocked high to keep up with Nvidia’s cards, but the same architecture runs cool when clocked lower for mobile applications, e.g. Vega was notoriously inefficient on the desktop due to being delayed two years and having to compete with a different generation than it was designed to, but was great in laptop APUs). Intel would also have gained experience with chiplets and packaging a fast GPU with a CPU. It let everyone involved make more money than doing it any other way.
What? Ah here it is:
They could buy AMD Radeon chiplets too, that would be something.:)
Intel actually bought AMD Radeon GPUs for their Hades Canyon (Kaby Lake G) platform. It was a NUC mainboard with a full Intel platform, combined with an AMD Radeon GPU. The Intel CPU and the GPU (including HBM2 memory for the GPU) was all on one package soldered to the mainboard.
I think they did a couple of follow ups on that as well, because it worked very well.
Interesting. What a strange companionship. :)
There was a time when you could have a platform that ran with components from all 3. Intel CPU, AMD GPU, and nvidia’s nForce chipset on the mobo.
AMD’s GPUs were much faster than Intel’s, and making GPUs for this kind of application was something AMD already did. Nvidia didn’t, so would have to design a whole chip from scratch, and didn’t really have a power efficiency advantage (in recent generations where AMD’s desktop cards have run hot, it’s because they’ve been clocked high to keep up with Nvidia’s cards, but the same architecture runs cool when clocked lower for mobile applications, e.g. Vega was notoriously inefficient on the desktop due to being delayed two years and having to compete with a different generation than it was designed to, but was great in laptop APUs). Intel would also have gained experience with chiplets and packaging a fast GPU with a CPU. It let everyone involved make more money than doing it any other way.
And it fixes nothing. Who is going to buy an already less powerful chip now half powered by Intel. Not that this is find to fix anything.