Compute | New York // Major Panel: Reading between the lines: Striking the XPU/GPU/CPU balance to adapt to AI inference workloads?
- —
- Compute Main Stage
- Compute
- Open Session
Speakers
Editor-in-Chief & Publisher, DatacenterDynamics
SVP, Product & Strategy, Cerebras
VP of Product and Strategy for IBM Infrastructure Support, IBM
Founder & CEO, Lemurian Labs
Curtis Pulley
Vice President of Business Development for the Infrastructure Line of Business, Arm
GenAI is growing 800x every 2 years, but compute is growing at a rate of less than 2x and a GPU supply shortage looms, but is the rush for GPUs going to continue as AI inference becomes the dominant workload? Could CPUs act as cost-effecitve alternative?
Or will inference leave GPUs gathering dust?
How should you balance hardware performance in the short and long-term? Join our talk to find out:
- Primrose Path: As inferencing makes up 85% of the AI compute problem, is the focus on GPUs over CPUs leading everyone down the wrong path?
- Getting Lean: How will changes in model building from going bigger to higher efficiency models affect the choice of compute hardware?
- GP4U: How should you think about the rise of GPU-as-a-Service and its role in next gen AI workloads?