Characterizing 2nd Law Efficiency of AI Datacenters
Datacenter workload fluctuations challenge the design and operation of our critical grid infrastructure. Significant gaps exist in assessment of these fluctuations as to their lifetime and short-term impact on the infrastructure and environment. Some mitigation approaches focus on throttling the datacenter workload, thus impacting the performance. Other approaches include use of alternate generation sources or energy storage systems. While disparate, these approaches are reactive and lack foresight and the intelligence to plan and strategize for optimal power management. We introduce an approach to quantify the transitional entropy generated during datacenter power fluctuations as a metric to evaluate datacenter performance using power demand measurements. A comparative assessment is provided between a BESS optimized datacenter and a regular datacenter to demonstrate the reduction of irreversibilities due to power fluctuations. Workloads are used to characterize the datacenter power demand at the point of interaction with utility. This approach can be scaled from datacenters to servers to chips.
Date and Time
Location
Hosts
Registration
-
Add Event to Calendar
- Zio Fraedo's
- 611 Gregory Lane
- Pleasant Hill, California
- United States 94523
- Starts 25 February 2026 01:00 PM UTC
- Ends 21 April 2026 05:00 AM UTC
- Admission fee ?
Speakers
Ratnesh K Sharma
Characterizing 2nd Law Efficiency of AI Datacenters
Biography:
Ratnesh K Sharma, PhD, P.E. is the Chief Technology Officer at Relyion Energy Inc., He has over 25 years' experience in energy management systems for Datacenters and Energy Storage Systems. He has worked on AI based algorithms for dispatch and life-optimization of energy storage systems. He has (co)authored several cited publications and granted patents.
Agenda
No-host social at 5:30pm
Presentation at 6:00pm
Dinner at 7:00pm
Presentation continues at 7:45pm
Adjourn by 8:30pm