Datacenter fabric startup GigaIO lands $14.7M

GigaIO, a startup growing high-speed interconnect {hardware} for computing clusters, right this moment introduced that it raised $14.7 million in a funding spherical led by Impression Ventures. The Carlsbad, California-based firm says that it’ll put the proceeds towards increasing its advertising, gross sales, associate, and channel groups in addition to supporting its product R&D efforts.

It’s typically troublesome to plan IT infrastructure for and round altering AI workflows. For instance, coaching algorithms can require specialised {hardware} like accelerators, which span GPUs, discipline programmable gate arrays (FPGAs), and customized chips. Embracing the cloud is one solution to obtain scalability, however doubtlessly at a excessive price. The opposite is to make on-premises {hardware} extra versatile by adopting software-defined infrastructure — particularly composable compute, which makes community, storage, and compute assets obtainable over a community.

GigaIO, which Joey Maitra based in 2012, gives datacenter “cloth” {hardware} — interconnecting switches — that allow composable compute. Maitra, previously head of engineering for a agency making pc servers and gear for the army, prototyped GigaIO’s first product with an exterior “field of slots” and a PCI Specific (PCIe) with cabling to connect with it.

“One of many primary methods we democratize entry to AI and machine studying is by [delivering] the identical infrastructure for use by totally different groups and departments to run various kinds of workflows,” Maitra advised VentureBeat by way of e-mail. “Infrastructure necessities are very totally different for information ingest part versus the information coaching part. With out GigaIO, particular person techniques are [designed] to deal with every part. The info then strikes from system to system for information ingest, cleansing and tagging, coaching, and inference. Sources are idle as much as 85% of the time for GPUs … and require a major footprint.”

Optimizing infrastructure

GigaIO’s cloth product can repurpose one set of datacenter infrastructure “at will” for an AI workflow utilizing the fitting mixture of {hardware} for the job, in accordance with Maitra. For instance, firms utilizing Google’s TensorFlow framework for predictive analytics, which requires a particular CPU-GPU ratio and particular sorts of GPUs, can leverage GigaIO’s know-how to optimize the allocation of those on-premises assets.

“Our primary ‘competitor’ is the cloud,” Maitra mentioned. “The pandemic slowed preliminary testing that the majority clients require. Nevertheless, for the reason that begin of the 12 months, we’ve got seen exercise improve dramatically, and buyer engagement is at document ranges.”

Maitra declined to call clients. Nevertheless, GigaIO earlier this 12 months introduced that its cloth will make up part of the forthcoming Prototype Nationwide Analysis Platform (NRP), a computing platform designed by the College of California, San Diego for scientific analysis. The NRP will likely be underwritten by a $5 million five-year grant from the Nationwide Science Basis, with matched funding offered for techniques operation.

“Complicated computational and information workflows underpin most of the scientific analysis challenges we hope to handle with NRP,” Dr. Frank Würthwein, interim director of the San Diego Supercomputer Heart, mentioned in a press release. “In areas as various as public well being, excessive vitality physics, and wildfire response, this analysis requires that we combination disparate computational components, resembling FPGAs, GPUs, x86 processors, and storage techniques into extremely usable and reconfigurable techniques. GigaIO’s … know-how makes it doable to dynamically carry these components collectively in a really low-latency, high-performance interconnect whereas permitting for distinct, non-interfering workflows to co-exist on the identical infrastructure.”

Up to now, 30-employee GigaIO has raised $22.5 million in enterprise fairness financing.