Enakta Labs Collaborates with KIOXIA to Advance Storage Solutions

We've published a Reference Architecture whitepaper with KIOXIA Corporation, combining their CD8P-R NVMe drives with the Enakta Storage Platform. The short version: up to 200 GiB/s reads from a single server, on commodity hardware.

The whitepaper validates a production-ready architecture that pairs Enakta Labs' storage software with KIOXIA's PCIe 5.0 NVMe technology. Lower latency, higher throughput, and a clear path to scaling for AI, HPC, media, and enterprise workloads.

Key Takeaways

The numbers from the reference architecture speak plainly. Using the DAOS core and KIOXIA CD8P-R NVMe drives, the Enakta Storage Platform hits these figures on standard commodity server hardware:

KIOXIA CD8P-R NVMe

The CD8P-R is a read-intensive data centre NVMe SSD built on PCIe 5.0. Sequential reads hit 12,000 MB/s, writes reach 5,500 MB/s. That's a 60-80% improvement in sequential reads over the PCIe 4.0 predecessor, with consistently low latency under load.

Available in capacities up to 30TB and in both 2.5-inch and E3.S form factors, the CD8P-R provides the flexibility to scale storage capacity while taking full advantage of the PCIe Gen5 bandwidth.

Reference Architecture: The Building Block

The reference architecture is based on the Supermicro SYS-211GT-HNTR, a 2U chassis housing 4 single-socket nodes, providing an ideal balance of density, performance, and cost.

Component	Specification
Chassis	Supermicro SYS-211GT-HNTR (2U, 4 nodes)
CPU	Intel Xeon Gold 6430 (32 cores per node)
Memory	16 × 96GB ECC Registered DDR5 (1.5TB per node)
Storage	6 × KIOXIA CD8P-R NVMe 15TB (90TB raw per node)
Network	NVIDIA ConnectX-6 VPI dual port (2 × 100 Gbps per node)
Boot	2 × M.2 NVMe 480GB (per node)

This means the base 2U building block delivers 360 TB of raw capacity with up to 800 Gbit/s of network bandwidth, with theoretical throughput of 288 GB/s reads and 132 GB/s writes based on the peak performance of KIOXIA CD8P-R devices.

Scale-Out Design

The Enakta Storage Platform supports flexible data protection including N-way replication and erasure coding. For example, with 6+2 erasure coding the usable capacity is 75% of raw, and different datasets on the same platform can use different protection schemes.

Performance scales linearly across servers. The same DAOS engine has been proven at 1,000+ nodes in production at Argonne National Laboratory's Aurora supercomputer.

AI/ML Storage Challenges

With GPU performance doubling every two years, the bottleneck is shifting from compute to storage I/O. The faster data can reach GPUs, the less time they spend idle. DAOS with KIOXIA CD8P-R NVMe drives enables the throughput required for modern AI checkpointing, training datasets, and model serving, all at speeds that keep GPU clusters fully utilised.

What This Means

This validates the Enakta Storage Platform on enterprise-grade NVMe hardware from one of the world's leading flash manufacturers. The combination delivers high throughput at commodity pricing, and the reference architecture gives teams a concrete starting point for their own deployments.

We'll have more to share as the collaboration progresses. In the meantime, reach out if you'd like to discuss sizing for your workloads.