Graid Technology Achieves Industry-First 100 Million IOPS on a Protected RAID 5 Volume for GPU-Initiated I/O

Graid Technology announced an industry-first result: 100 million random 512-byte IOPS on a protected storage volume built from 32 Kioxia XD8 NVMe SSDs.The milestone highlights Graid Technology’s ability to deliver extreme storage performance with data protection for next-generation GPU-driven computing environments.

Beyond the raw I/O result, Graid Technology demonstrated that its protected RAID-5 storage volume can support demanding AI training workloads without measurable performance compromise. In benchmark testing with a WholeGraph training workload, this configuration delivered training performance on par with the baseline environment, showing that high-speed data protection does not have to come at the expense of application throughput.

This result marks an important milestone for AI infrastructure, where modern workloads, including graph analytics, large-scale training pipelines, retrieval-intensive inference, and emerging agentic frameworks, increasingly depend on massive volumes of fine-grained, random I/O to keep GPUs fully utilized. As AI systems scale in model complexity, data size, and concurrency, storage can no longer be treated as a secondary consideration behind compute.

This level of performance shows that resilient NVMe storage can now operate at the scale current and future AI deployments require. This is especially significant for environments that require both extreme performance and enterprise-class data protection, where downtime, rebuild risk, or storage bottlenecks can directly impact productivity and infrastructure efficiency.

“With this milestone, Graid Technology continues to prove that customers do not need to choose between performance and protection,” said Leander Yu, CEO, Graid Technology. “As AI infrastructure evolves toward more data-intensive and agent-driven workloads, storage architectures must keep pace with GPU performance. These results show that Graid Technology’s GPU-accelerated, and now GPU-initiated, approach to storage performance and data protection can deliver the resiliency enterprises require, while sustaining the ultra-low-latency, high-IO/s access that next-gen AI workloads demand.”