Intel has a history of being so bad at making integrated GPUs.
I'm not getting my hopes up on them making dedicated GPUs.
Intel is doing several different things with their discrete GPUs.
1. They are fabing it all outside of Intel for now (at least for these Arc Xe-HPG products ). So on TSM N6 and should be transitioning to N3 in 2023.
2. With their tile/chiplet disaggregation strategy some of these smaller 'external' GPU probably are going to get coupled on package in 2023-24 with an Intel CPU tiles. So can conceptually push the iGPU onto a denser fab processor if that means more sense.
3. Long term Intel does want a wider set of products to make for themselves in their fab. They'll push dGPUs aggressively and weave them back in when it make sense. ( not likely they are going to 'quit' in next 3-4 years even if it takes that long to weave back into "in house" development. )
At the initial stages Intel's iGPUs were allocated "spare die space" from the primary focus on CPU cores and its memory subsystem. When Intel 'paused' CPU cores at a '4 count' in the mainstream dies the iGPUs started to make more progress. Intel has had leading edge video en/decode.
What they were not willing to do is "blow up" the memory controller to focus on GPU rather than CPU workloads. For a discrete GPU that isn't going to be a problem. Not really 'sharing' the transistor budget or local memory controller.
Over time Intel have been growing their ability to handle a broader and deeper software stack.
Discrete GPUs from AMD and Nvidia have been moving to higher power budgets. Intel hasn't had a major problem keeping up in the higher power budget products.
Intel recently took out some job ads for a ultra low power GPU that is different set of products than their dGPU group. That would be the more skeptical path. Intel has said was going to get a super low power GPU going ever since got skipped for the iPhone and haven't delivered since.
