Blab Tech
KIOXIA Launches GP1 Series PCIe 6.0 NVMe SSDs Targeting AI GPU Memory Extension
KIOXIA unveiled its GP1 Series today, the first production PCIe 6.0 NVMe SSDs that promise to rewrite the rules for AI memory expansion.
Built on the company’s second‑generation XL‑FLASH, the GP1 drives deliver blistering low‑latency performance that can feed the massive data streams required by modern AI workloads. KIOXIA positions the series as a flash‑based memory‑extension tier that sits between DRAM and conventional NAND, giving GPUs a new avenue to scale capacity without the cost of adding more HBM.
The drives connect via PCIe 6.0 and support the NVMe 2.2 protocol. Each unit can achieve up to 10 million random‑read IOPS at a 512‑byte block size—a best‑case figure that can vary with host hardware, drivers and workloads. The architecture is designed to grow to 100 million IOPS in future iterations.
Key specifications include: - Flash memory: KIOXIA XL‑FLASH generation 2, a low‑latency, high‑performance storage class memory based on 96‑layer BiCS4 NAND. - Random read performance: Up to 10 million IOPS (512‑byte). - Endurance: Up to 50 drive writes per day (DWPD). - Form factors: E3.S and E1.S, available in 9.5 mm and 15 mm heights. - Cooling: Cold‑plate liquid cooling is supported on the E3.S and E1.S 9.5 mm models; all models can operate in standard air‑cooled server environments.
Unlike the capacity‑oriented CM10 Series—announced earlier this year and built on 332‑layer BiCS FLASH generation 10 TLC NAND—the GP1 is a latency‑centric solution that trades capacity for speed.
"AI memory wall is a critical challenge for our industry," said Neville Ichhaporia, senior vice president and general manager of KIOXIA’s SSD business unit. "KIOXIA’s XL‑FLASH is uniquely positioned to deliver the extremely high performance and low‑latency data access needed to support GPUs as a memory extension tier. It also offers a significantly lower cost per gigabyte than expanding in‑node HBM or DRAM capacity."
By extending High Bandwidth Memory with a high‑speed flash layer, AI platforms can use flash memory to expand capacity and improve GPU utilization, potentially reducing the cost of scaling AI memory.
KIOXIA plans to provide evaluation samples of the GP1 Series to select customers by the end of 2026. The drives will be showcased at the Future of Memory and Storage (FMS) conference, scheduled for August 4‑6 in Santa Clara, California.
The announcement follows KIOXIA’s March reveal of GP Series technology, which was described as a prototype that now has shipping silicon. The company has also announced an XL1 series memory expansion module that will be presented at the same FMS event.
Industry observers note that the GP1 Series is part of a broader trend toward storage class memory solutions that bridge the performance gap between DRAM and conventional NAND flash. By offering a flash tier that can be accessed directly by GPUs, KIOXIA aims to provide a cost‑effective alternative to expanding HBM in data‑center AI workloads.
At this time, KIOXIA has not released pricing or mass‑production timelines for the GP1 Series. The company’s focus remains on delivering a functional evaluation platform for early adopters and on refining the technology for future high‑performance iterations.
The GP1 Series represents KIOXIA’s first step toward a line of flash‑based memory extension products that could reshape how AI systems handle large data sets, offering a balance of speed, capacity, and cost that is tailored to GPU‑centric workloads.
Built on the company’s second‑generation XL‑FLASH, the GP1 drives deliver blistering low‑latency performance that can feed the massive data streams required by modern AI workloads. KIOXIA positions the series as a flash‑based memory‑extension tier that sits between DRAM and conventional NAND, giving GPUs a new avenue to scale capacity without the cost of adding more HBM.
The drives connect via PCIe 6.0 and support the NVMe 2.2 protocol. Each unit can achieve up to 10 million random‑read IOPS at a 512‑byte block size—a best‑case figure that can vary with host hardware, drivers and workloads. The architecture is designed to grow to 100 million IOPS in future iterations.
Key specifications include: - Flash memory: KIOXIA XL‑FLASH generation 2, a low‑latency, high‑performance storage class memory based on 96‑layer BiCS4 NAND. - Random read performance: Up to 10 million IOPS (512‑byte). - Endurance: Up to 50 drive writes per day (DWPD). - Form factors: E3.S and E1.S, available in 9.5 mm and 15 mm heights. - Cooling: Cold‑plate liquid cooling is supported on the E3.S and E1.S 9.5 mm models; all models can operate in standard air‑cooled server environments.
Unlike the capacity‑oriented CM10 Series—announced earlier this year and built on 332‑layer BiCS FLASH generation 10 TLC NAND—the GP1 is a latency‑centric solution that trades capacity for speed.
"AI memory wall is a critical challenge for our industry," said Neville Ichhaporia, senior vice president and general manager of KIOXIA’s SSD business unit. "KIOXIA’s XL‑FLASH is uniquely positioned to deliver the extremely high performance and low‑latency data access needed to support GPUs as a memory extension tier. It also offers a significantly lower cost per gigabyte than expanding in‑node HBM or DRAM capacity."
By extending High Bandwidth Memory with a high‑speed flash layer, AI platforms can use flash memory to expand capacity and improve GPU utilization, potentially reducing the cost of scaling AI memory.
KIOXIA plans to provide evaluation samples of the GP1 Series to select customers by the end of 2026. The drives will be showcased at the Future of Memory and Storage (FMS) conference, scheduled for August 4‑6 in Santa Clara, California.
The announcement follows KIOXIA’s March reveal of GP Series technology, which was described as a prototype that now has shipping silicon. The company has also announced an XL1 series memory expansion module that will be presented at the same FMS event.
Industry observers note that the GP1 Series is part of a broader trend toward storage class memory solutions that bridge the performance gap between DRAM and conventional NAND flash. By offering a flash tier that can be accessed directly by GPUs, KIOXIA aims to provide a cost‑effective alternative to expanding HBM in data‑center AI workloads.
At this time, KIOXIA has not released pricing or mass‑production timelines for the GP1 Series. The company’s focus remains on delivering a functional evaluation platform for early adopters and on refining the technology for future high‑performance iterations.
The GP1 Series represents KIOXIA’s first step toward a line of flash‑based memory extension products that could reshape how AI systems handle large data sets, offering a balance of speed, capacity, and cost that is tailored to GPU‑centric workloads.
