Image credit: CentML
CentML, an AI startup focused on optimizing machine learning models, emerged from stealth today. The Toronto-based company aims to solve the global shortage of GPUs necessary for training and inference of generative AI models.
CentML believes that the scarcity of accessible compute is a significant hurdle to AI development, and this problem will only intensify as the demand for inference workloads grows. By improving the efficiency of existing AI chips and legacy inventory, the company hopes to increase access to computational resources, especially in the current "broken" GPU market.
Addressing the GPU Accessibility Gap
CentML secured a $3.5 million seed round in 2022 led by Radical Ventures. Cofounder and CEO Gennady Pekhimenko emphasizes the need to democratize access to GPU resources, especially given the dominance of giants like Nvidia that control about 80% of the GPU market.
According to Pekhimenko, Nvidia's business model, focused on selling its most expensive chips, limits access for smaller companies. However, less expensive, underutilized chips could be optimized more efficiently using CentML's software, thereby serving a larger market.
Innovative Optimization Techniques
As the costs of running models like ChatGPT skyrocket, CentML uses an open-source compiler to automatically tune optimizations for a company's specific inference pipeline and hardware. Pekhimenko points out that while competitors like OctoML use similar approaches, CentML's technology does not suffer from the same deficiencies, offering them a competitive edge.
AI startup CentML emerged from stealth mode, aiming to optimize machine learning models and address the global GPU shortage.
CentML's software optimizes less expensive, underutilized GPU chips, making compute resources more accessible to a larger market.
By using a unique open-source compiler, CentML can tune optimizations for specific inference pipelines and hardware, offering a competitive edge over similar solutions.
David Katz, a partner at Radical Ventures, likens the scramble for AI chips to a "Game of Thrones"-like competition. He believes that CentML’s approach could "increase the supply of chips in the market" and potentially revolutionize the accessibility and efficiency of AI compute resources. Source