Kioxia Doubles XL-FLASH SSD Speed to 20M+ IOPS at OCP
Kioxia's third-gen XL-FLASH boosts GP-series SSDs past 20 million IOPS, while the LD4 E1.L QLC drive on BiCS 8 218-layer flash scales to 122.88 TB.
- By
- Grace Kim
- Filed
- Channel
- Memory & Storage
- Read
- 3 min read
Kioxia unveiled a third-generation XL-FLASH SSD delivering more than 20 million IOPS in 512-byte random reads at the OCP Global Summit, doubling the figure of the GP1 drive the company announced in August. The new GP-series flagship targets AI-driven data center workloads. The Japanese memory maker paired the announcement with a longer QLC E1.L drive and a liquid-cooled E1.S demo.
The GP series sits in the Optane-class low-latency tier Kioxia has built around its proprietary XL-FLASH media. Gen 3 XL-FLASH triples read performance and lifts writes by roughly 150 percent, Kioxia said. The company attributes the gains to shorter word lines and bit lines, which reduce RC delay, and to more planes per die than the 16-plane Gen 2 architecture. Energy efficiency improves by about 40 percent for reads and 80 percent for writes, though Kioxia cautions that energy per I/O will still rise because performance scales faster than efficiency.
The underlying cell technology remains MLC with a faster SLC mode, and read latency stays under five microseconds. Kioxia has not yet disclosed endurance ratings for the new GP drive. Neville Ichhaporia, SVP and GM of Kioxia America's storage device business unit, framed the announcement in AI terms: "Kioxia is making game-changing high-performance flash storage solutions for AI infrastructure. As AI drives new performance demands, scalable, efficient storage becomes the foundation for innovation. We're helping data centers build that foundation by pioneering storage solutions designed specifically for AI-driven infrastructure."
How does the LD4 extend Kioxia's QLC capacity line?
The LD4 is a single-port QLC drive in the E1.L form factor with a PCIe 5.0 x4 interface. Kioxia is launching it at 15.36 TB and 30.72 TB. The architecture scales to 122.88 TB, the company said, and an intermediate 61.44 TB point appears likely based on the empty bays the smaller capacities leave inside the E1.L enclosure. The drive uses BiCS 8, Kioxia's 218-layer 3D NAND, leapfrogging the LD2 from October 2023, which used BiCS 5 (112-layer) flash on a PCIe 3.0 bus.
Kioxia has not released performance or endurance figures for the LD4. The company confirmed support for NVMe 2.0 and OCP Datacenter NVMe SSD Specification 2.6, plus end-to-end data loss protection. Compared with Kioxia's own LC9 — which reaches 246 TB using BiCS 8 in a denser package — the LD4 prioritizes the E1.L serviceable form factor that hyperscalers standardize on.
What is the competitive picture in E1.L PCIe 5.0 SSDs?
Three E1.L PCIe 5.0 QLC drives now compete for hyperscaler slots:
- Micron 6550 ION
- DapuStor R6060
- FADU Echo
The LD4 enters at the lower capacity points where these competitors already ship. Kioxia's ability to scale the same platform to 122.88 TB will determine whether it can capture high-capacity sockets currently served by denser QLC packages like the 246 TB LC9.
What else is Kioxia showing?
Kioxia is also demonstrating the liquid-cooled NX1 Series E1.S NVMe SSD, a data-center drive designed for direct-liquid-cooling sleds. The company did not disclose changes to the NX1 platform at the show.
When and where?
The OCP 2026 Global Summit runs October 12 to 16 at the San Jose Convention Center.
Whether Kioxia converts the LD4's scaling headroom and the GP-series IOPS lead into design wins will depend on pricing, endurance disclosures, and qualification cycles at the hyperscalers the LD4 explicitly targets.
Source: Blocks & Files
More from Grace Kim
Show full bio
Market editor covering industry trends and analytics at Chip Dispatch.
262 articles
Related articles
microsoft-s-2-599-surface-laptop-ultra-puts-nvidia-rtx-spark-in-first-arm-pc-35164c89
Microsoft's $2,599 Surface Laptop Ultra Puts Nvidia RTX Spark in First Arm PC
openai-s-jalapeno-chip-pressures-nvidia-s-ai-margins-6b702d45
OpenAI's Jalapeño Chip Pressures Nvidia's AI Margins
vdura-v12-ports-hyperscaler-storage-tactics-to-ai-factories-3fee5c6f
VDURA V12 Ports Hyperscaler Storage Tactics to AI Factories
xinnor-ships-xinas-3-13-previews-ai-assisted-management-via-mcp-a19c453a
Xinnor Ships xiNAS 3.13, Previews AI-Assisted Management via MCP


