Tag

GPU hardware

11 posts tagged with this topic.

Server rack with organized cables and blinking indicator lights

Nemotron 3 Nano and the Vessel AI Stack

NVIDIA's open-weight Nemotron 3 Nano activates 3B of 31B parameters at inference. Here is what that means for on-vessel deployment.

Ethan Marsh·June 22, 2026

Close-up of AI processor chips arranged on a circuit board

Infrastructure5 min read

JetPack 7.2: The Vessel Agent Stack Arrives

NVIDIA's JetPack 7.2 lands NemoClaw on Jetson Thor with multi-instance GPU. Here is what that means for running AI agents on a vessel.

Ethan Marsh·June 8, 2026

Multiple illuminated screens displaying data dashboards in a dark monitoring environment

Infrastructure12 min read

Edge AI Observability Without Phoning Home

How to monitor on-vessel AI inference, GPU health, and model performance using Prometheus, Grafana, and OpenTelemetry without any cloud dependency.

Ethan Marsh·June 7, 2026

Rows of illuminated servers in a data center

Edge AI4 min read

NVIDIA AI Grid and the Vessel Math

NVIDIA's AI Grid cuts edge inference costs by 76 percent. The math for sovereign vessel AI just got a lot harder to argue against.

James Calder·June 5, 2026

Close-up of illuminated technology hardware panels with blue indicator lights

Infrastructure12 min read

Vessel GPU Racks: Power, Cooling, Security

The unglamorous realities of deploying GPU compute on a vessel. Power budgets, thermal management, vibration isolation, and physical access control.

James Calder·May 17, 2026

Server racks packed with network switches and cables

Edge AI6 min read

Kimi K2.6: 1T Parameters, 8 H200s Required

Moonshot AI's Kimi K2.6 ships natively in INT4 at 1T parameters. Open weights, world-class agent benchmarks, and a hardware bill that starts at 8 H200s.

Ethan Marsh·April 21, 2026

Organized network cables connected to ports in a server rack

Edge AI12 min read

Deploying a 70B LLM Offline at Sea

Hardware, quantization, and serving stack for running Llama 3.1 70B on a vessel with no cloud fallback. Real specs and real benchmarks.

Ethan Marsh·April 19, 2026

Close-up of a graphics card with red and black cooling fans

Infrastructure18 min read

On-Vessel GPU Selection Guide

Blackwell B200, H200, H100, L40S, and A100 compared for vessel deployments. When each is the right call and the power realities.

James Calder·April 11, 2026

Modern vessel bridge with multi-screen navigation displays

Edge AI17 min read

KV Cache Quantization at Sea (3/3)

Google's TurboQuant compresses the KV cache to 3 bits with no accuracy loss. Here is what that means for a vessel GPU rack.

Ethan Marsh·April 10, 2026

Compact AI compute module in a rack enclosure

Infrastructure15 min read

Unified Memory for Vessel AI

AMD Strix Halo and NVIDIA DGX Spark bring 128 GB of unified memory to a shelf-sized box for under $4,000. When to buy one.

James Calder·April 10, 2026

Circuit board with illuminated AI processor

Infrastructure14 min read

1M Context: The GPU Math (2/3)

VRAM and KV cache math for models from 7B to 405B at 1M tokens. Per-user scaling for crew and guests. The numbers are ugly.

James Calder·April 7, 2026