or that allows expanding flash-based memory to help meet the growing performance and memory capacity needs of AI infrastructures.
The KIOXIA GP1 series is designed to support new storage architectures for AI, which extend high-bandwidth memory (HBM) by using a fast, flash-based memory layer. This approach allows AI systems to access larger datasets at a significantly lower cost than adding more HBM[4], while improving GPU utilization.
The KIOXIA GP1 series SSDs leverage high-performance, low-latency KIOXIA XL-FLASH flash memory to deliver significantly higher IOPS and more granular data access with 512-byte blocks. The architecture is designed to scale from the current 10 million random read IOPS to future generations, with the goal of reaching 100 million IOPS.
“The increasing performance and memory capacity demands of AI infrastructures require a broader memory hierarchy to efficiently deliver data to GPUs,” said Paul Rowan, Vice President and Head of Marketing at KIOXIA Europe GmbH. “The KIOXIA GP1 series enables flash-based memory expansion and direct GPU access, helping to reduce CPU latency and overhead, and boosting performance for AI, HPC, and data analytics workloads.”.
Key features of the KIOXIA GP1 series include:
• Designed to PCIe 6.0 and NVMe 2.2 specifications
• Utilizes low-latency KIOXIA XL-FLASH flash memory
• Up to 10 million random read IOPS with a 512-byte block size[2]
• Available in 9.5 mm and 15 mm E3.S and E1.S designs, with cold plate liquid cooling capability in the 9.5 mm E3.S and E1.S designs (all designs are compatible with conventional air-cooling environments)
• Up to 50 DWPD endurance[5]
