Intel’s first wave of Xe graphics cards is supposed to launch this year. However, a lot of questions are being asked about the viability and efficiency of these GPUs. Recently, an official Intel document was leaked that showed an RTX 2060 (160-180W) level GPU expected to draw nearly 500W. That’s just not going to cut it. The DG1 sample being shipped to developers is expected to be a 96 to 128EU GPU. That’s the same shader count as the Tiger Lake Gen12 graphics, expected to be on par with the NVIDIA GTX 1050/1050 Ti.
Have a look at the FP32 and FP16 performance of the Intel DG1 GPU. The FP32 performance is quite mediocre, marginally higher than the GeForce GTX 650 Ti, but the FP16 score is more than twice as much.
FP32 is mainly used in gaming and other traditional workloads. FP16 or half-precision compute is often leveraged in MATLAB and neural network training which don’t require a high level of precision. This means that Intel is probably focusing more on AI and neural network performance, rather than gaming at this point.
There’s a good chance that the 1st Gen Intel Xe graphics cards will be mostly for Data Centers and other professional workloads, with little to no focus on gaming. The lack of ray-tracing and any gamer-centric announcements about Xe further back this claim up. Ponte Vecchio too, for example, is also an Exascale processor.
In the beginning, Intel had promised that Xe would scale from integrated laptop graphics to gaming and all the way to Data Centers. However, it seems that the company has run into some roadblocks, and it’ll take a while for that roadmap to become a reality.