Intel CEO: AI Shift to Inference Could Boost CPU to GPU Ratio to 4:1

Intel CEO Chen Liwu says AI transition from training to inference could push CPU to GPU ratio to 4:1. Intel 18A yields improve 7% monthly, 14A PDK 0.5 released.

Intel CEO: AI Shift to Inference Could Boost CPU to GPU Ratio to 4:1

CEO Chen Liwu stated at the JPMorgan 54th Global Technology, Media and Communications Conference that the CPU to GPU configuration ratio is moving from 1:8 towards 1:1, and could even reach 4:1 as AI shifts from training to inference. The company is also pursuing an ASIC business for custom solutions.

Intel 18A process yields improve 7% monthly

Intel's 18A process node now supports production of Panther Lake processors, with yield improving approximately 7% per month. For the 14A node, the 0.5 PDK has been released, and a 0.9 PDK is planned for external release in October. Additionally, Intel is developing 10A and 7A nodes.

IC substrate supply remains tight, and customers are prepaying to secure capacity. Intel is adapting its product mix to the changing AI landscape, where inference workloads may demand a higher proportion of CPU cores relative to GPUs.

The company has not confirmed specific pricing or availability for products based on these nodes. Intel's process roadmap and AI inference trends were discussed at the JPMorgan conference.

Discussion

0 comments

Log in to join the thread with a thoughtful take, question, or correction.

Add to the discussion