Computex 2024: Kioxia and Xinnor Collaborate to Deliver High Performance PCIe 5.0 NVMe SSD RAID Solution for Enterprise and Data Center Applications

Kioxia Corporation announced that its PCIe 5.0 NVMe SSDs have been successfully tested for compatibility and interoperability with the Xinnor, Ltd. RAID solution and achieved up to 25x higher performance running PostgreSQL than software RAID solutions with the same hardware configuration ⁽¹⁾.

This solution was demonstrated in the company’s booth at Computex 2024 Taipei, which held from June 4 to June 7.

PostgreSQL (with the pgvector extension) and vector databases are becoming more important for generative AI and RAG (Retrieval Augmented Generation) systems than before, and these results demonstrate the performance gains utilizing Xinnor’s xiRAID Opus and the firm’s PCIe 5.0 NVMe SSDs solution for a GenAI and RAG application.

New servers with the PCIe 5.0 interface and corresponding high-speed SSDs are in demand for high performance applications, such as GenAI, and the importance of PCIe 5.0 compatible SSDs to support this demand is increasing. The partners’ high performance software RAID solution maximizes the performance of PCIe 5.0 SSDs for AI, ML, and data analytics applications in on-premises enterprise data centers. The company’s CM7 Series SSDs successfully completed compatibility testing performed by both parties.

Success of next-gen data center infrastructures will be dependent on ecosystem collaboration and interoperability testing efforts to ensure that current and future products and technologies work together and deliver as expected. As a leader in enterprise and data center class SSDs, Kioxia is committed to driving the industry forward with memory solutions that power the next wave of applications and services. The company will continue to support the PCIe 5.0 ecosystem and maximize the value of high performance PCIe 5.0 NVMe SSDs.

⁽¹⁾ Compared to a standard RAID solution in Linux (mdraid/mdadm), in degraded mode with one drive failure, in database read (query) operation.