Intel made headlines in late 2017 after they hired AMD's Raja Koduri to be their new lead GPU architect. This was followed with information that Intel was planning on developing a line of discrete GPUs. Japanese site PC.Watch has received some more details on just what Intel has been working on.

Intel hasn't been competitive in the GPU market for more than two decades and the closest thing to a discrete GPU they have currently is the Xeon Phi compute card. We're not sure how much progress Intel has made with its discrete GPUs, but these details show that they have definitely been working hard at it. That being said, Intel has made it clear that this is just a prototype and will not be a future product.

The GPU will be built in a 14nm process and will consist of just over 1.5 billion transistors. For comparison, the GTX 1080Ti has 12 billion transistors and the GT 1030 has 1.8 billion transistors. The current prototype has a clock range from 50MHz - 400MHz although the final design may be clocked much higher.

This diagram shows that it will be a two-chip system. The GPU will be on one chip while there will also be a second FPGA to communicate with the host PC over PCIe.

Since this design is just for research purposes and isn't likely to ever make its way to market, it's hard to say what this all means. While this chip is certainly not aimed at gamers or other high-end GPU applications, the knowledge learned from it could very likely make its way into consumer cards.