Engineering topics

Latif, Imran

Publications and source records attributed to Latif, Imran.

3 records

DOE OSTI2026

Cooling Matters: Benchmarking Large Language Models and Vision-Language Models on Liquid-Cooled Versus Air-Cooled H100 GPU Systems

The unprecedented growth in artificial intelligence (AI) workloads, recently dominated by large language models (LLMs) and vision-language models (VLMs), has intensified power and cooling demands in data centers. This study benchmarks LLMs and VLMs on two HGX nodes, each with 8× NVIDIA H100 graphics processing units (GPUs), using liquid and air cooling. Leveraging GPU Burn, Weights & Biases, and IPMItool, we collect detailed thermal, power, and computation data. Results show that the liquid-cooled systems maintain GPU temperatures between 41-50$^\circ$C, while the air-cooled counterparts fluctuate between 54-72$^\circ$C under load. This thermal stability of liquid-cooled systems yields 17% higher performance (54 TFLOPs/ GPU vs. 46 TFLOPs/GPU), performance-per-watt, reduced energy overhead, and greater system efficiency than the air-cooled counterparts. These findings underscore the energy and sustainability benefits of liquid cooling, offering a compelling path forward for hyperscale data centers seeking to optimize AI infrastructure. https://github.com/iscaas/Cooling-Matters.

Latif, Imran↗

DOE OSTI2025

Empirically-calibrated H100 node power models for accurate AI training energy estimation

Accurately quantifying the energy use of artificial intelligence (AI) training is critical for infrastructure planning, carbon accounting, and sustainable data center operation, but few studies have directly measured the power consumption of production workloads on contemporary hardware. By combining empirical measurements from Brookhaven National Laboratory during AI training on 8-graphics-processing-unit H100 systems with open-source benchmarking data, we develop statistical models relating computational intensity to node-level power consumption. We measure the gap between manufacturer-rated thermal design power (TDP) and actual power demand during AI training. Our analysis reveals that even computationally intensive workloads operate at only 76% of the 10.2 kW TDP rating. Our architecture-specific model, calibrated to floating-point operations, predicts energy consumption with 11.4% mean absolute percentage error, significantly outperforming TDP-based approaches (27%–37% error). We identified distinct power signatures between transformer and convolutional neural network architectures, with transformers showing characteristic fluctuations that may impact grid stability. These results provide a measurement-grounded basis for improving AI training energy estimates, enabling more reliable infrastructure sizing, cost projections, and environmental impact assessments.

Newkirk, Alex C↗

DOE OSTI2023

Finalizing Transition to the New Data Center at BNL

Computational science, data management and analysis have been key factors in the success of Brookhaven National Laboratory's scientific programs at the Relativistic Heavy Ion Collider (RHIC), the National Synchrotron Light Source (NSLS-II), the Center for Functional Nanomaterials (CFN), and in biological, atmospheric, and energy systems science, Lattice Quantum Chromodynamics (LQCD) and Materials Science, as well as our participation in international research collaborations, such as the ATLAS Experiment at Europe's Large Hadron Collider (LHC) at CERN (Switzerland) and the Belle II Experiment at KEK (Japan). The construction of a new data center is an acknowledgement of the increasing demand for computing and storage services at BNL in the near term and enable the Lab to address the needs of the future experiments at the High-Luminosity LHC at CERN and the Electron-Ion Collider (EIC) at BNL in the long term. The Computing Facility Revitalization (CFR) project is aimed at repurposing the former National Synchrotron Light Source (NSLS-I) building as the new data center for BNL. The construction of the new data center was finished in 2021Q3, and it was delivered for production in early FY2022 for all collaborations supported by the Scientific Data and Computing Center (SDCC), including STAR, PHENIX and sPHENIX experiments at RHIC collider at BNL, the Belle II Experiment at KEK (Japan), and the Computational Science Initiative at BNL (CSI). This paper highlights the key mechanical, electrical, and networking components of the new data center in its final configuration as used in production since 2021Q4 and gives an overview for the extension of the central network systems into the new data center and the migration of a significant portion of IT load and services from the old data center to the new data center carried out in 20212023, with expected completion of the main phase of the gradual IT equipment replacement and migration from the old data center into the new one set to the end of FY2023 (Sep 30, 2023).

99 GENERAL AND MISCELLANEOUS↗