What type of study is this?

A Thermal-Aware Workload Scheduler for High-Performance LLM Inference in Cooling-Regulated Datacenters

View Full Paper

Ask AI

Bookmark

Authors

Observational analysis shows a new scheduling method reduces thermal throttling risk in GPUs, suggesting enhanced performance.

A thermal-aware workload scheduler increases the throughput of LLM inference by 40.94% at 41°C ambient temperature.
Existing schedulers can increase the probability of thermal throttling by 10 times and degrade performance by 34.2%.
Effective cooling regulation is crucial for high-performance computing, particularly for GPU workloads in AI datacenters.
The proposed solution accounts for GPU voltage and frequency to optimize job assignments in thermal-regulated environments.

Lu et al. (2025) studied this question.

View Full Paper

Ask AI

Bookmark

View Full Paper

Ask AI

Bookmark

Authors

Observational analysis shows a new scheduling method reduces thermal throttling risk in GPUs, suggesting enhanced performance.

A thermal-aware workload scheduler increases the throughput of LLM inference by 40.94% at 41°C ambient temperature.
Existing schedulers can increase the probability of thermal throttling by 10 times and degrade performance by 34.2%.
Effective cooling regulation is crucial for high-performance computing, particularly for GPU workloads in AI datacenters.
The proposed solution accounts for GPU voltage and frequency to optimize job assignments in thermal-regulated environments.

Lu et al. (2025) studied this question.

View Full Paper

Ask AI

Bookmark