Research & Innovation

New Supermicro Server Supercharges School of Business Research

Nvidia H100 GPU server from Supermicro will provide significant boost in AI and fintech research.

Black and white photo of student using a computer from the 1950'sIn a 1956 edition of Stevens: A College for Engineers, the publication highlighted advanced work opportunities that allowed students to take the theories learned in the classroom and apply them through laboratory work. At the time, Stevens boasted access to an “ELECTRONIC COMPUTER, with the capacity sixty percent greater than any of its kind,” that allowed students to assist with their research and designs.

Nearly 70 years later, students, along with faculty and corporate partners, will now have access to a Nvidia H100 GPU server from Supermicro through the Hanlon Financial Systems Center in the Stevens School of Business to enhance their own work in the areas of artificial intelligence and financial technologies.

Access to this powerful new server, will allow greater experimentation that wasn’t possible before, allowing faculty and students to develop breakthrough research at a much faster pace and keep the School of Business at the forefront of producing leading-edge research. The speed and new capabilities will also allow corporate partners to scale up their work. The addition of this unit comes after the School of Business added its first H100 server more than a year ago.

AI use case benefits

  • GPU Power: The Nvidia H100 GPUs are built on the Hopper architecture, providing exceptional computational power with a focus on AI tasks. This enables faster model training and inference, crucial for developing and deploying complex AI models.

  • Tensor Cores: H100 GPUs come with Tensor Cores, optimized for deep learning operations, significantly accelerating matrix computations and improving throughput for neural networks.

  • Mixed Precision Computing: H100 GPUs support mixed precision (FP16/FP32) computing, which speeds up computations while maintaining accuracy, essential for training large AI models.

  • AI Optimizations: The server integrates software optimizations and AI frameworks, such as CUDA, cuDNN, and TensorRT, which are optimized for Nvidia GPUs, enhancing performance and ease of development.

Finance use case benefits

  • Low Latency: The server’s high-performance capabilities ensure low latency, which is crucial for high-frequency trading algorithms that need to process and react to market data in microseconds.

  • Data Processing Speed: The H100 GPUs can handle large datasets and complex calculations required for risk analysis, portfolio optimization, and predictive modeling, delivering results faster than traditional CPU-based systems.

  • Parallel Computing: Financial models often involve Monte Carlo simulations and other computationally intensive methods, which benefit from the parallel processing power of multiple GPUs.

  • Enhanced ML Capabilities: For tasks like sentiment analysis, market prediction, and anomaly detection, the server’s AI capabilities enable more accurate and faster model training and inference.

  • Big Data Analytics: Financial research often involves analyzing large volumes of structured and unstructured data. The H100 server’s ability to process big data efficiently aids in uncovering insights and trends.

Stevens H 100-Projects

FinMem: A Performance-Enhanced LLM Trading Agent with Layered Memory and Character Design

  • 1.1 Arxiv: 2311.13743

  • 1.2 Venues:

  • 1.3 How the hardware is utilized in the project: In this paper, we established a framework that utilizes large language models (LLMs) to power individual agents. With our novel memory database design, these trading agents outperform existing reinforcement learning counterparts. The use of H100 GPUs provides a highly efficient inference endpoint, supporting our main experiments.

The FinBen: A Holistic Financial Benchmark for Large Language Models

  • 1.1 Arxiv: 2402.12659

  • 1.2 Targeted Venue: NeurIPS 2024

  • 1.3 How the hardware is utilized in the project: This paper proposes a new benchmark framework for evaluating large language models (LLMs) within a financial context. In collaboration with researchers from Yale, Stony Brook, NYU and other institutions, we investigated the use of LLMs for named entity recognition, classification and trading tasks. Numerous models were deployed on the H100 GPUs to support these diverse tasks.

FINCON: A Synthesized LLM Multi-Agent System with Conceptual Verbal Reinforcement for Enhanced Financial Decision Making

  • 1.1 Venue: NeurIPS 2024, under review

  • 1.2 How the hardware is utilized in the project: In this project, we further developed our FinMem framework to design multi-agent systems. In this complex system, different agents collaborate towards a unified goal, each with delegated functionalities. The H100 hardware enables the deployment ofsuch an integrated and complex system.

Let Clickstream Talk: A Graph Neural Network Approach to Sales Forecasting

  • 1.1 Venue: POMS, under review

  • 1.2 How the hardware is utilized in the project: In this project, we designed a time series deep learning model that uses a click graph as its input fea- ture structure. During the experiments, the H100 GPU helped scale the number of concurrent training iterations, reducing the time needed to benchmark a model by 5 to 10 times.

Bi-channel Multi-modal Object Matching with VLM-powered Open-ended Concepts Generation (BiOCGen)

  • 1.1 Targeted Venue: AAAI 2024

  • 1.2 How the hardware is utilized in the project: In this project, we plan to pretrain the BLIP 2 model to enable the vision-language model to generate complementary fashion product recommendations. The H100 GPU will serve as the hardware for hosting the model during training andinference.

IJCAI Competition: Fine Tuning LLM for Financial Applications

  • 1.1 Venue: Financial Challenges in Large Language Models - FinLLM, Competition, 4th grade

  • 1.2 How the hardware is utilized in the project: With the H100 GPU, fine-tuning state-of-the-art models becomes feasible. In this project, we fine-tuned the Llama 3 8B model and the Mistral 7B model for classification, summarization and trading tasks within the financial domain. Our team achieved fourth place among all competitors and has been invited to submit a paper to the IJCAI proceedings.