Challenge:
With over 3 MW of power consumption during peak HPC workloads like HPL, LRZ was seeking a way to cap energy usage on SuperMUC-NG to 2 MW, without sacrificing performance.
Solution:
LRZ deployed EAR to dynamically manage and limit energy usage across its high-density Intel CPU cluster. EAR’s smart power capping and runtime optimization features allowed LRZ to meet its energy goals while maintaining system throughput. Following the success of this first deployment, LRZ extended EAR to its second flagship cluster, SuperMUC-NG2, which integrates Intel Data Center GPUs (Ponte Vecchio) to accelerate scientific research at scale.
Results::
- Peak energy usage reduced by 33% (3 MW → 2 MW)
- Minimal performance degradation on large workloads
- Transparent operation with no impact on user experience
- Successful extension to GPU-accelerated workloads on SuperMUC-NG2
- Ongoing collaboration with EAS for tuning and support
“We are fully satisfied that EAR meets our goal and by the quality of professional services that EAS is delivering.”
Dr. Herbert Huber, Head of High Performance Systems Department, LRZ
3 Comments
This is exactly what i was looking for, thank you so much for these tutorials
It would be great to try this theme for my businesses
What a nice article. It keeps me reading more and more!