Pure Storage He has developed one High performance data storage platform Designed to meet the needs of companies that have to work with artificial intelligence and high performance computing. The platform, called FlashBlade//EXAIt has an architecture developed to solve bottleneck problems with metadata thanks to the fact that it is intended for high concurrence and for massive amounts of typical metadata operations of the workloads of AI and high -scale high performance computing.
According to the first tests carried out by the company to the platform, Flashblade // Exa will offer a reading mode of more than 10 terabytes per second in a single Namespace. Thanks to its architecture, it is able to climb data and metadata independently, offering an almost limit scale with OTT third -party data nodes that allow highly escahating multidimensional performance and reduce the complexity in the implementation, management and escalation through the use of standard protocols and networks.
Flashblade // Exa has a disaggregated and parallel architecture, which provides flexibility to large -scale storage, which allows companies to adapt to the evolution of multimodal models of AI, optimize reliability and reduce dead times to accelerate the training and inference of AI models. All while optimizing the use of GPUS. It also combines the Pure Storage metadata engine with the Purity operating system, for greater performance adjustment.
The platform also incorporates Nic Nvidia Connectx of high speed, Spectrum switches, linkx cables and accelerated communications bookstores. It will be available, predictably, throughout next summer 2025.
Rob Lee, CTO de Pure Storagehas highlighted this platform that «Flashblade // Exa offers a massively parallel architecture that allows independent data and metadata scaling to provide customers, scalability and unique adaptability for some of the largest and most demanding data environments in the world ». The storage is accelerating the rhythm of the evolution of the HPC and the AI on a large scale«.
