The field of computer science is moving towards more sustainable and efficient computing solutions, particularly in the areas of AI and data centers. Researchers are exploring innovative strategies to reduce the carbon footprint of AI workloads, such as the 'Follow-the-Sun' strategy, which involves dynamically moving workloads to regions with cleaner energy sources. Additionally, there is a growing interest in optimizing energy efficiency in machine learning retraining, with techniques such as retraining with only the most recent data and retraining only when necessary. Furthermore, the development of new data center architectures, such as the FullFlat network architecture, is enabling more scalable and efficient support for large language models. Noteworthy papers in this area include 'On the Effectiveness of the 'Follow-the-Sun' Strategy in Mitigating the Carbon Footprint of AI in Cloud Instances', which demonstrates the effectiveness of the 'Follow-the-Sun' strategy in reducing carbon emissions, and 'Scaling Intelligence: Designing Data Centers for Next-Gen Language Models', which provides a comprehensive co-design framework for designing data centers that can efficiently support trillion-parameter models.
Sustainable and Efficient Computing in AI and Data Centers
Sources
On the Effectiveness of the 'Follow-the-Sun' Strategy in Mitigating the Carbon Footprint of AI in Cloud Instances
A4: Microarchitecture-Aware LLC Management for Datacenter Servers with Emerging I/O Devices
DPUV4E: High-Throughput DPU Architecture Design for CNN on Versal ACAP
Topology-Aware Virtualization over Inter-Core Connected Neural Processing Units
Capsule: Efficient Player Isolation for Datacenters
Sustainable Machine Learning Retraining: Optimizing Energy Efficiency Without Compromising Accuracy
Empirically-Calibrated H100 Node Power Models for Reducing Uncertainty in AI Training Energy Estimation
Efficient Serving of LLM Applications with Probabilistic Demand Modeling
Scaling Intelligence: Designing Data Centers for Next-Gen Language Models
LiteGD: Lightweight and dynamic GPU Dispatching for Large-scale Heterogeneous Clusters
CXL-GPU: Pushing GPU Memory Boundaries with the Integration of CXL Technologies
From Block to Byte: Transforming PCIe SSDs with CXL Memory Protocol and Instruction Annotation