Specifications
Interface
PCIe 6.0 x4, NVMe 2.2
Flash Memory
KIOXIA XL-FLASH Gen 2
Low-latency flash
Random Read IOPS
Up to 10,000,000 IOPS
(512-byte block size)
Endurance
Up to 50 DWPD
Form Factors
E3.S, E1.S 9.5mm, E1.S 15mm
Cooling
Cold-plate liquid cooling (E3.S, E1.S 9.5mm)
+ traditional air cooling
Use Case
GPU memory extension tier
AI HBM capacity expansion
Architecture
Flash as memory extension
512-byte fine-grained access
Scalability
Target: 100M IOPS future gen
from 10M IOPS today
Power Efficiency
Lower power per I/O vs. TLC SSDs
Availability
Evaluation samples: late 2026
Award
FMS 2026 Best of Show
Specialized Storage Category
Overview
The KIOXIA GP1 Series represents a paradigm shift in enterprise storage architecture — the industry's first PCIe 6.0 NVMe SSD purpose-built for GPU direct memory access. Powered by KIOXIA's second-generation XL-FLASH low-latency flash memory, the GP1 delivers an unprecedented 10 million random read IOPS with 512-byte data access, enabling flash to serve as a high-performance memory extension tier for GPUs. This fundamentally changes the economics of AI infrastructure by extending High Bandwidth Memory (HBM) with a cost-effective flash tier, allowing AI systems to efficiently access larger datasets without relying solely on costly in-node HBM or DRAM expansion.
Unlike conventional enterprise SSDs designed for block storage, the GP1 is architected as a GPU memory extension — positioned between HBM and traditional NVMe storage in the memory hierarchy. GPUs directly access the GP1 for model weights, context data, and intermediate results that exceed onboard HBM capacity, delivering significantly lower cost-per-gigabyte than HBM or DRAM expansion while maintaining the low latency required for AI workloads. The architecture scales from today's 10M IOPS to future generations targeting 100 million IOPS.
Key Benefits
GPU Memory Extension: The GP1 bridges the gap between HBM capacity limits and the massive datasets required by frontier AI models. Rather than scaling expensive HBM, data centers can deploy GP1 drives as a high-performance flash memory tier directly accessible by GPUs. XL-FLASH Gen 2 Technology: KIOXIA's proprietary low-latency flash delivers significantly higher IOPS, finer-grained 512-byte data access, and lower power consumption per I/O operation compared to conventional TLC-based enterprise SSDs. 50 DWPD Endurance: With up to 50 drive writes per day, the GP1 is engineered for the intensive, sustained write workloads of AI training and inference pipelines. PCIe 6.0 & NVMe 2.2: The next-generation interface doubles the bandwidth of PCIe 5.0, ensuring the IOPS capability is never bottlenecked by the host connection. Liquid Cooling Ready: E3.S and E1.S 9.5mm form factors support cold-plate liquid cooling for dense multi-GPU server deployments.
Request a Quote — KIOXIA GP1 Series
QS Compute — NVIDIA Goldendisk Partner. Global B2B supply. Evaluation samples available late 2026.
Get Your Quote →