Hardware

Intel to Use ill-fated 10nm Process node for its Xe Architecture; First GPU to Launch in Mid-2020

We know that Intel plans to enter the GPU market in 2020. They have already unveiled their GPU architecture called Xe. DigiTimes has reported that Intel is looking for a mid-2020 release date for its first product in the GPU market. It means Intel can either use Computex or E3 as a launch event to unveil their products. Intel has already cleared that they are not going to be competing against the flagship products from Nvidia or AMD. Their first discrete GPU will be a mediocre device with performance comparable to the GTX 1050.

Xe architecture

A little is known about the Gen 12 GPU architecture from Intel. Intel is formally calling it Xe (e is the superscript) architecture; the meaning of Xe has been left for readers to contemplate. Intel’s graphics team is working under Raja Koduri, who has served as the lead architect for AMD Radeon. Intel brought him over specifically to improve their gen architecture and to head their graphics team. The salient feature of Xe architecture is its ability to scale both mobile and desktop components. It means rather than developing another architecture for the mobile GPU; they can use a single architecture to develop both desktop and Mobile GPUs.

We have already seen AMD’s RDNA architecture with the same capabilities, though RDNA’s scalability is much higher. It could potentially end up in our mobile devices; Samsung is already working on it. On the other hand, Xe architecture is still a mystery to most. The comparison with RDNA architecture will be unfair for Intel.

Intel is planning to use the 10nm process node to manufacture the architecture and its early products. It will be competing against AMD RDNA and Nvidia’s upcoming Ampere architecture. Both of which will be based on the 7nm manufacturing process. Intel may shift to the 7nm process during 2021, as it plans to make products for data center, AI, and HPC markets. It will use Foveros 3D technology to stack GPU, memory, and memory controllers on a single die. It will potentially remove the bandwidth problems that most GPUs face; however, the technology is still in its incubation state.


Leave a Reply

Your email address will not be published.

Close