NVMesh

Build powerful NVMe over Fabrics high-performance shared storage solution with MayaScale. Consolidate direct attached NVMe resources into shared storage pool and provision flexible NVMe namespaces to clients demanding high performance with low latency. When finished the clients can return NVMe storage space back to storage pool. No more over-provisioning or stranded NVMe storage space as is the case with Direct Attached storage. This network agnostic solution uses RDMA for on-premises and standard TCP on cloud. True NVMe device is visible to client using standard NVMe driver stack without need for any proprietary driver extensions. Configure and deploy NVMe over Fabrics SAN infrastructure at rack scale in your data center by pooling heterogeneous NVMe devices over RDMA capable interconnect. This includes ROCE, iWARP or Infiniband. Even on public Cloud, experience NVMe over Fabrics using the standard TCP/IP protocol without need for any specialized RDMA hardware or SRIOV virtualization.

Hammerspace is a revolutionary storage platform that unlocks unused local NVMe storage in GPU servers to accelerate AI training and checkpointing. It transforms siloed, stranded storage into a shared, ultra-fast tier that dramatically increases GPU utilization and reduces the need for costly external storage systems. By using a standards-based parallel file system, Hammerspace delivers low-latency, high-throughput data access that scales to thousands of GPU servers. The platform helps cut power consumption and infrastructure costs while boosting AI workload performance. Leading organizations like Meta rely on Hammerspace to optimize their AI infrastructure. With easy deployment and rapid scaling, Hammerspace enables teams to get AI models trained faster and more efficiently.