Intel Tech Tour 2025 in Arizona was primarily focused on disclosures around Xeon 6+ Clearwater Forest and Panther Lake / Xe3 but during the opening keynote was also teasing a yet-to-be-announced inference-optimized GPU.
On Intel’s AI Execution Roadmap during the Intel Tech Tour keynote they shared a new inference-optimized GPU:
“Inference-optimized GPU with enhanced memory, bandwidth and enterprise-level inference”
It will be very interesting to see what this new inference-optimized GPU is… From the open-source Linux driver activity, there hasn’t been any recent new indicators surfacing of some additional hardware on the horizon. My guess would just be an inference-optimized product based on the BMG-G31 Battlemage GPU, a.k.a. “B770” type product. It’s too early for any Xe3 / Celestial dGPU product as so far in that context we have only seen Linux driver patches pertaining to Xe3 integrated graphics with Panther Lake and Wildcat Lake platforms.
Whatever the case this new inference-optimized enterprise-level GPU should fit in nicely given their ongoing Project Battlematrix Linux driver efforts for multi-GPU compute, SR-IOV support, etc.
We are eager to learn more soon about this new Intel GPU offering. Intel is expected to reveal more at the OCP Global Summit coming up next week in San Jose.
