This company develops high-performance computing platforms optimized for deploying large language models in the cloud. Their hardware and software solutions reduce energy consumption and enhance AI security, enabling data centers and enterprises to improve AI performance while ensuring data privacy.
Funding
Funding not disclosed
Founders
Product
Problem
Data centers face increasing demands to deploy large language models (LLMs), which require significant hardware resources and consume substantial energy. Existing computing platforms often struggle to balance performance, energy efficiency, AI security, and data privacy when running these complex AI workloads.
Solution
Pascaline Systems offers heterogeneous computing platforms designed to accelerate LLM deployment in the cloud while minimizing energy consumption and enhancing AI security. Their data-centric hardware solutions, coupled with orchestration software, cater to the specific requirements of AI-native cloud services. By focusing on optimizing the cost of inference, Pascaline's architecture enables data centers and enterprises to deploy AI models with lower hardware demands, improved model quality, and reduced instances of AI hallucinations. The company's approach to data center design, starting from the chip level, allows for tailored, scalable, and eco-conscious data centers that meet the unique needs of AI.
Target Audience
Pascaline Systems targets data centers and enterprises seeking to deploy large language models (LLMs) efficiently and securely, with a focus on reducing hardware requirements and energy consumption.
Features
- Veloxity-100: High compute density card with a HHHL form factor, designed for easy integration into existing systems.
- Veloxity-210: High compute density board with up to 512GB on-board flash memory for localized data storage.
- High-speed network interface with 2x QSFP28 (Veloxity-100) or 4x QSFP28 (Veloxity-210) for fast data transfer.
- Category Theory (CAT)-based AI models: Provides a framework for defining relationships between data points, minimizing guesswork and the need for corrections, leading to smaller, more precise, and accurate AI models.
- Holistic thermal management: Optimizes thermal management at every level, from chip selection to building design, resulting in efficient airflow and better thermal interfaces.
- Power efficiency optimization: Selects the most efficient components for specific workloads, optimizing power distribution and reducing conversion losses.
- Scalable architecture: Designed with future chip densities and cooling requirements in mind, ensuring the facility remains cutting-edge for years.