Pepperdata launches AI tool to boost GPU use & cut costs by 30%
Pepperdata has released a new solution designed to cut costs and improve resource efficiency for artificial intelligence (AI) workloads running on graphics processing units (GPUs). The offering, pepperdata.ai, is now generally available and targets organisations striving to maximise the utilisation of their GPU assets both in the cloud and on-premises.
AI GPU utilisation
The challenge of underutilised and idle GPU hardware is increasing as businesses expand their infrastructure to support a growing number of AI projects. Organisations investing heavily in GPU resources are seeing parts of their costly infrastructure sitting unused or not fully leveraged, leading to higher operating bills and slower progress on strategic initiatives. Poor GPU allocation also complicates the work of platform operators, who must balance user demand and performance with budget constraints.
"As enterprises scale up GPU infrastructure to support AI initiatives, they are discovering a painful truth: Most GPUs are underutilised. Expensive hardware sits idle or fragmented, while operators struggle to balance performance, cost, and access," said Ash Munshi, Chief Executive Officer, Pepperdata.
Automated optimisation
The pepperdata.ai solution applies automation to both the allocation and management of GPU resources, with the goal of ensuring high-priority workloads such as real-time inference, batch inference, and development tasks receive appropriate computing power. This approach can result in substantial savings, with Pepperdata citing cost reductions of up to 30% on GPU expenditure for users of its optimisation solutions.
Pepperdata said its product functions as an automatic optimisation layer, relieving end users from having to manually tune or configure how workloads are distributed across different GPU resources within their environment. The company notes that this frees technical staff to focus on higher-value activities.
"We consider Pepperdata to be the optimisation layer for all our platforms, including our GPU environments. Our end users can rely on Pepperdata to do all the optimisation for them, automatically-which frees them to focus on more strategic value-add initiatives for our company," said a Cluster Operations Manager at a Fortune 10 enterprise.
Product features
The new offering comprises two core components, designed to work independently and in conjunction with each other. The first, GPU Demand Optimisation, allows platform owners to identify mismatches between GPU supply and demand, and then shift workloads according to availability by time or GPU type. This helps spread demand more evenly across a company's available GPU assets, ensuring maximum resource utilisation and efficiency.
The second component, GPU Resource Optimisation, leverages NVIDIA's Multi-Instance GPU (MIG) capability. MIG allows a single GPU to be partitioned into multiple secure, isolated resource pools, enabling a more granular allocation of resources to individual workloads. This results in more effective infrastructure usage, the ability to run a greater number of workloads to completion, and further potential for cost savings.
Industry partnerships
Pepperdata has reported that customers across various sectors, including global financial services and technology firms, use its resource optimisation tools to manage complex, large-scale compute environments. The company states that it has helped customers save a combined amount of over USD $250 million since its founding.
"Pepperdata intelligently allocates and manages GPU resources, ensuring that critical AI workloads, including Real-Time Inference, Batch Inference and Jupyter Notebooks, receive the necessary computing power while eliminating waste and delivering substantial savings and faster time-to-insight," said Munshi.
