Need smarter insights in your inbox? Sign up for our weekly newsletter to get only the things that matter to enterprise AI, data and security leaders. Subscribe now
From enhancing patient care, to sophisticated medical imaging, to power complex fraud detection models, to helping to preserve wildlife, key bottlenecks often emerge as AI applications become increasingly permeable in enterprise operations: data storage.
At VentureBeat's Transform 2025, Greg Matson, Head of Product and Marketing, Soligm, and Greg Matson of Roger Cummings, CEO of Peak: AIO: AIO spoke with M12 managing partners.
The Monai framework is a breakthrough in medical imaging, building faster, safer and safer. Advances in storage technology enable researchers to build on this framework, quickly iterate and innovate. Peak: AIO partnered with SolidGM to integrate power-efficient, performance, and large-capacity storage that allows storage of over 2 million full-body CT scans on a single node in an IT environment.
“As enterprise AI infrastructure evolves rapidly, storage hardware needs to be tailored to specific use cases, depending on where your AI data pipeline is,” says Matson. “The types of use cases we spoke with Monai, edge application cases, and training cluster feeding are well provided by very large capacity solid state storage solutions, but actual inference and model training require very high performance. Software.”
Improved AI inference at the Edge
For peak performance at the edge, it is important to reduce storage to a single node to bring inference closer to the data. And the important thing is to remove the memory bottleneck. It can be done by making memory part of the AI infrastructure to scale along with data and metadata. The proximity of the data dramatically increases the time to insight.
“We see all the huge deployments, big greenfield data centers of AI. We can use a very specific hardware design to get data as close as possible to the GPU,” says Matson. “They have been building data centers with very large capacity solid state storage to provide very high-speed storage with petabyte-level storage that has very access to GPUs. The same technology is happening at the edge and in the enterprise with microcosmetics.”
It is important for AI system buyers to get the best performance from the system by running it in all solid states. This provides a huge amount of data and allows for incredible processing power in small systems at the edge.
The future of AI hardware
“It's essential to provide a solution at memory speed, open, scalable, memory speed, using the latest and best technology to do that,” Cummings said. “That's our goal as a company by providing its openness, its speed, and the scale that an organization needs. I think we'll see the economy align that.”
Within the overall training and inference data pipeline, and inference itself, the hardware needs continue to increase, whether it's a very fast SSD or a very large capacity solution with high power efficiency.
“Whether it runs at very low power and can basically replace a 4x hard drive or a very high performance product that is close to memory speed, I think it will move further towards very large capacity,” Matson said. “We see that large GPU vendors are considering how to define the next storage architecture, which allows them to scale the HBM of their systems very closely. Cloud computing generic SSDs are divided into capacity and performance.
Source link
