Systems

A card in a box is a support problem.

Name: Umbra N8
Brand: ZipLogic

N4 and N8 ship configured, burned in, acceptance tested, and engineer-installed.

Why systems

Configured, tested, installed

Power, airflow, firmware, burn-in and acceptance each introduce variables a card in foam cannot control. Umbra nodes arrive as validated systems: not a parts bin and a manual.

Nodes are quoted per configuration. Typical programmes begin at a single evaluation card and scale to multi-node deployments.

Configurations

N4 and N8

	Umbra N4	Umbra N8
Cards	4	8
Throughput	160M ops/s estimated1	320M ops/s estimated
Wall power	~600 W	~1400 W
Form factor	4U rack-mount	4U rack-mount
Network	Dual 100 GbE	Dual 100 GbE
V-series ready	Yes	Yes
Delivery	Engineer installed	Engineer installed

Umbra N8

Eight cards. One system.

Accelerators	8 × Umbra U100
Throughput	320,000,000 ops/s estimated1
Efficiency · at wall	~230,000 ops/s per watt3
Node draw	~1.4 kW
Host	Dual-socket, 768 GB
Storage	2 TB boot (mirrored), 30 TB NVMe
Data plane	Dual 100 GbE
Management	Dedicated BMC, IPMI 2.0 + Redfish
Power	4 × 2000 W Titanium, 3+1 redundant, dual feed
Cooling	Front-to-back directed airflow
Security	TPM 2.0, signed firmware and bitstream, chassis intrusion detection
Upgrade	V-series card upgrade in the same chassis, Q4 2026

V-ready

Provisioned for the upgrade path

Every ZipLogic node chassis is provisioned for the V-series card upgrade from day one. That claim applies to nodes purchased from ZipLogic: not to cards dropped into an underspecified host.

Roadmap →

Configured systems, engineer installed.

Request a quotation for N4 or N8, or a briefing to size the deployment.

Request a quotation Umbra U100

Methodology

¹ Encrypted logic operations per second, measured on production silicon. Node figures estimated from measured single-card throughput.

³ N8 node wall power estimated for dual-socket host under load; efficiency derived from estimated node throughput ÷ wall power.