Kioxia Europe has introduced the GP1 Series SSD, boasting a record 10 million random read IOPS and a remarkable endurance rating of up to 50 drive writes per day. Designed to support emerging AI storage architectures, this PCIe 6.0 NVMe SSD aims to optimize GPU utilization and reduce latency in AI training and inference systems.
- 10 million IOPS using PCIe 6.0 and NVMe 2.2 standards
- Exceptional endurance of up to 50 DWPD for sustained heavy workloads
- Designed for GPU direct access and flash-based memory extension in AI
What happened
Kioxia Europe GmbH unveiled the GP1 Series SSD built on PCIe 6.0 and NVMe 2.2 protocols, targeting high-performance AI infrastructure. The drive achieves up to 10 million random read IOPS at a 512-byte block size, using second-generation XL-FLASH memory, an SLC NAND combined with a proprietary controller.
This SSD is offered in E3.S and E1.S form factors with liquid cooling compatibility and air-cooled options. Evaluation samples began distribution to select customers by the end of 2026, with a public showcase planned at the Flash Memory Summit conference in California.
Why it matters
AI workloads often require rapid parallel data transfers between storage and GPUs to avoid performance bottlenecks. Kioxia’s GP1 Series addresses this by providing extremely high IOPS rates and durable endurance of 50 drive writes per day, which equates to over 100 petabytes of writes on a 3TB drive over two years.
This capability allows AI systems to extend memory capacity using flash storage tiers without compromising on GPU data supply, improving GPU utilization and minimizing idle time. The endurance and speed also support extensive, sustained AI training and inference workloads at scale.
What to watch next
Kioxia aims to advance the GP Series further, targeting a future SSD generation capable of 100 million IOPS, aligning with demands from major AI hardware companies like Nvidia. Achieving this will likely require innovations beyond current NAND flash technologies to meet escalating AI infrastructure needs.
Industry observers should monitor how these advancements influence AI hardware architectures and adoption, especially as AI systems grow increasingly data-intensive and demand faster, more durable storage solutions integrated directly with GPUs.