Specifications
Product
Xsight Labs E1 DPU
CPU Cores
Up to 64 x Arm Neoverse-N2
Memory Controller
DDR5-5200
PCIe
40 x PCIe Gen5 lanes
Networking
Dual 400G MACs — 800G aggregate
Compute Comparison
Roughly half the compute of Microsoft Azure Cobalt 100 (128 Neoverse-N2 cores)
Software Stack
Runs a vanilla Ubuntu LTS ISO installed directly on the DPU
Significance
No vendor-specific OS or image required — long support horizon
Development Platform
Standalone dev board accepting add-in cards and drives
Measured Power
Approximately 100-104 W at the wall (development platform)
Memory Population (tested)
Four DDR5-5200 ECC modules
Storage Role
2x 400G bandwidth roughly matches 8x PCIe Gen5 NVMe SSDs
Deployment Form
PCIe add-in card and custom OCP sled variants exist
Application
Hammerspace AI storage solution shown at OCP Summit 2025
Class
Linux-capable DPU — unlike many data-path processors that require proprietary images
Thermal
Large passive heatsink on the development board; it is audibly loud under load
Relevance to AI Racks
Offloads storage stacks and data path so host CPUs stay on inference
Supply
B2B volume supply; quote per configuration
Overview
The Xsight Labs E1 is an 800G DPU built around up to 64 Arm Neoverse-N2 cores with DDR5-5200 controllers, 40 PCIe Gen5 lanes and dual 400G MACs. For scale, that is roughly half the compute of Microsoft's 128-core Azure Cobalt 100 — enough Neoverse-N2 silicon to run real general-purpose work, not just packet forwarding.
The detail that separates the E1 from most data-path processors is software. A vanilla Ubuntu LTS ISO installs directly onto the DPU. Many Linux-running DPUs cannot do this and instead require vendor-supplied OS images, which leaves the operator dependent on the vendor for support as long as the device is in service. A standard install gives the platform a much longer useful life and removes a procurement risk that shows up years after purchase.
Architecturally the E1's bandwidth match is deliberate: dual 400G MACs line up with 8x PCIe Gen5 NVMe SSDs, so a sled built around the chip can move storage traffic at wire speed without the host CPU touching it. That is the shape of the Hammerspace AI storage solution Xsight demonstrated at OCP Summit 2025. In a test development platform the whole setup drew around 100-104 W at the wall with four DDR5-5200 ECC modules populated.
Key Benefits
64 Arm Neoverse-N2 cores — enough general-purpose compute to run storage stacks and services on the DPU, not just the data path. Vanilla Ubuntu LTS support, which removes vendor-image lock-in and lengthens the support horizon. 40 PCIe Gen5 lanes and 800G of networking sized to match eight Gen5 NVMe drives. Multiple form factors — PCIe add-in card, standalone development platform and OCP sled.
Applications
AI storage nodes offloading NVMe and file stacks from host CPUs; KV-cache and data-path acceleration in GPU clusters; DPU-based storage shelves matching 8x Gen5 NVMe; OCP sled deployments; software-defined storage appliances; network function offload in virtualised infrastructure.
Request a Quote — XSIGHT LABS E1 DPU — 64 ARM NEOVERSE N2 CORES, DUAL 400G, 800G TOTAL
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →