Autonomous Vehicle Technology Leader Benefits from High Bandwidth I/O to Its GPU Cluster
WekaIO (Weka), the innovation leader in high-performance, scalable file storage for data-intensive applications, announced that Innoviz, a leading manufacturer of high-performance, solid-state Light Detection and Ranging (LiDAR) sensors and Perception Software that enables the mass-production of autonomous vehicles, has selected the Weka File System (WekaFS) to accelerate its Artificial Intelligence (AI) and deep learning workflows. WekaFS has been chosen by Innoviz to improve application performance at scale and deliver high bandwidth I/O to its GPU cluster.
“Weka solves these issues, with WekaFS presenting a shared POSIX file system to the GPU servers and delivering extreme performance to keep data-intensive applications compute-bound.”
Innoviz’s solid-state LiDAR sensors are key to the future of autonomous cars. The sensors and Perception Software, which identifies, classifies, segments, and tracks objects to give autonomous vehicles a better understanding of the 3D driving scene, rely heavily on AI.
Having recently closed its Series C funding round with $170M secured, Innoviz is choosing and developing the right technologies to empower it to realize its expansion plans and enhance its manufacturing capabilities. Part of Innoviz’s vision is to accelerate the progress to mass production and to meet the growing demand for affordable sensing solutions that enable autonomy.
After comparisons with legacy NFS-based NAS storage solutions, Innoviz selected WekaFS because performance improvements with WekaFS matched the company’s needs.
“Weka’s storage scalability and ability to grow the infrastructure without losing performance, was a key factor in the decision to select the Weka file system,” said Oren Ben Ibghei, IT manager, Innoviz.
“Customers who are like Innoviz in terms of managing multiple petabytes of data on-premises and supporting I/O-intensive workloads are looking for alternatives to legacy storage systems that are costly at scale and deliver sub-optimal performance,” said Doron Zuberman, executive vice president, E&M Computing (Emet), the integrator that delivered the solution in partnership with Weka. “The larger the dataset, the better the AI outcomes, so immediate access to data lakes is a critical requirement for a storage solution. Being a software-only solution, Weka is the perfect storage alternative for AI workloads as it allows for the most economic build-out of infrastructure at scale.”
“Weka is being used by many AI companies to significantly reduce AI training Epochs. We can help companies shorten wall clock time by ensuring the GPU cluster is fully saturated with as much data as the application needs. Managing large amounts of data is challenging when the AI training system spans multiple GPU nodes. A shared file system eliminates this challenge, but legacy NFS-based NAS can cause I/O starvation to the GPUs,” said Liran Zvibel, co-founder and CEO, WekaIO. “Weka solves these issues, with WekaFS presenting a shared POSIX file system to the GPU servers and delivering extreme performance to keep data-intensive applications compute-bound.”
WekaFS was architected to leverage the benefits of NVMe flash technology, delivering high-performance, high-bandwidth, and low-latency storage infrastructure to meet the demands of today’s GPU-enabled AI and High-Performance Computing (HPC) workloads in the data center and in the cloud. WekaFS is the world’s fastest and most scalable file system for AI systems, which generate unpredictable I/O workloads with highly random-access patterns across both small and large files. Weka has achieved proven scalable performance of over 70 GBytes per second bandwidth to a single GPU node, 10x more than NFS and 3x more than a local NVMe SSD can deliver.