The card

Umbra U100

One accelerator. Full-height, single-slot, passively cooled. 40M encrypted operations per second.

40M ops/s1

Per card

25–75 W

Power envelope

PCIe ×8 Gen3

Host interface

Specification

Public fields

Throughput	40,000,000 encrypted logic ops/s1
Host interface	PCIe ×8 Gen3
Form factor	Full-height, full-length, single-slot
Power envelope	25–75 W
Cooling	Passive: requires directed chassis airflow
Attachment	Host-attached: no on-card network interface
Runtime	Umbra runtime · Umbra SDK
Operating system	Ubuntu 22.04 LTS · RHEL 8 / 9

The U100 is passively cooled and depends on chassis airflow. Intended for rack-mount servers with directed front-to-back cooling: not workstation chassis.

Workloads

What the card runs today

RainDB: fully FHE-encrypted database queries
ZChat: private encrypted AI (reference: 6 × U100 + 2 × NVIDIA GPUs)
Encrypted inference on the Umbra runtime
Custom workloads via the Umbra SDK

Use today →

Evaluation

Start with one card.

A single U100, the runtime, and a defined workload against your own data. Thirty days, with our engineers. Evaluation units credit against a production order.

Request evaluation

Product visualisation: Umbra U100 and V-series accelerator cards

Upgrade path

Same chassis. Next generation.

Umbra V-series arrives Q4 2026 as a card upgrade in the same chassis: preliminary ~1.5G ops/s per 8-card node4. Today's N8 is estimated at 320M ops/s1.

See N4 / N8 Roadmap

Full specification sheet

Complete mechanical, electrical and thermal specifications. Sent to your work address.

Interrogate the card with your workload.

Technical evaluation and quotation. Response within two business days.

Request a briefing Integrate

Methodology

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

⁴ V-series ~1.5G ops/s per 8-card node is a preliminary engineering estimate (equivalent ~150–200M per card); not a committed specification.