Intel Arc Celestial Xe3 GPUs Confirmed Ready For Tape-Out Soon
Image: Haze (@Haze2K1 on Twitter)
What does this mean? Basically, Intel's giving OEMs and BIOS vendors a way to test their firmware and software stacks remotely, through a secure cloud platform that simulates real hardware pretty convincingly. No lab access? No problem. Behind the scenes, they're probably using massive FPGA setups to emulate the chips—basically standing in for the as-yet unavailable Xe3 'Celestial' GPUs.
Right now, Intel's pre-silicon validation team is deep in the weeds tweaking how the chip handles power and voltage across different states—sleep, idle, boost—you name it. They're dialing in the frequency curves that define how the GPU behaves under varying loads, and making sure it all plays nice with power delivery and voltage scaling.
Since the Xe3 IP is meant to scale across everything from laptops to full-on discrete graphics cards, the engineers are testing it in every power class they can. A lot of the data flow inside the chip depends on how these frequency and power curves are tuned, so it's all tightly linked. That phase of work is pretty much done now, and the team's moving on to other optimizations. Once the actual silicon comes back from production, they'll do a final tuning pass to get everything running exactly how they want it.
Given this, assuming there are no major issues, we might see Celestial GPUs at CES next year. That isn't soon, but remember, Intel still has another Battlemage GPU to launch—or so it seems, anyway. Of course, that won't be the first Xe3-based GPU we encounter; Panther Lake's integrated graphics are known to be based on the Xe3 architecture, and while rumors have planted the idea that Xe3 will come in multiple architectural forms (like Xe1 but unlike Xe2, which is identical between Lunar Lake and Battlemage), it's unclear how different Xe3 and Xe3P will really be.
