Principal Investigator: Prof. Kayode P. Ayodele

Institution: Dept. of Electronic and Electrical Engineering, Obafemi Awolowo University (OAU), Nigeria

Lab Designation: The AMD Pervasive AI & Robotics Lab

1. Executive Summary

The Lace Network is a pilot initiative to solve the “AI Compute Gap” in the Global South without relying on centralized supercomputers or foreign cloud infrastructure. By aggregating high-performance edge workstations into a federated mesh, we aim to demonstrate that African institutions can collectively train sovereign foundation models on-premise.

This proposal requests 10x AMD Strix Halo (Ryzen AI Max+ 395) workstations to establish the network’s anchor node at OAU. These devices will serve a dual purpose: powering the AMD Pervasive AI & Robotics Lab for 300+ undergraduate students annually, and acting as the primary aggregation cluster for “Naiden,” our multi-modal healthcare foundation model.

2. The Problem: Sovereignty vs. Compute

African AI research faces two significant constraints:

1. The Compute Gap: Frontier model training typically requires multi-million-dollar GPU clusters (e.g., H100s), which are financially inaccessible to most African universities.
2. Data Sovereignty: Strict data protection laws (such as the Nigeria Data Protection Act 2023) prohibit the export of patient medical records to foreign cloud servers.

This creates a deadlock: We cannot use the cloud due to privacy laws, and we cannot afford on-premise supercomputers. African healthcare AI remains stalled.

3. The Solution: A Federated “Lace” Mesh

Instead of a single massive supercomputer, The Lace Network proposes a “fabric” of powerful, distributed nodes.

• Federated Learning Architecture: Individual nodes (hospitals/labs) fine-tune models locally on private data. Only encrypted weight updates (never raw patient data) are transmitted to the OAU Aggregation Node.
• The “Naiden” Model: A multimodal healthcare agent designed to diagnose region-specific pathologies (e.g., Malaria, Lassa Fever) and interpret Nigerian Pidgin English.

4. Technical Justification: Why AMD Strix Halo?

The success of this project hinges specifically on the AMD Ryzen AI Max+ 395 (Strix Halo) architecture. Standard discrete GPUs are unsuitable for our specific constraints:

• 128GB Unified Memory is Critical: To run and fine-tune large multimodal models (7B–40B parameters) locally, we require massive video memory. Discrete consumer GPUs (limited to 16GB–24GB VRAM) cannot load these models without extreme quantization. The Strix Halo’s 128GB unified memory allows us to run “datacenter-class” workloads on an edge device.
• The NPU & Intelligent Control: For our robotics curriculum, the integrated NPU allows for low-latency inference of computer vision models directly on the robot control loop, bypassing the latency of cloud API calls.
• Energy Efficiency: As an APU, Strix Halo offers high performance per watt, a critical feature for deployment in regions with unstable power grids where running 1000W+ discrete GPU rigs is impractical.

5. Deployment Plan: The 5+5 Strategy

The 10 requested units will be deployed in a hybrid configuration to maximize impact:

Cluster A: Education (5 Units)

• Location: The OAU AMD Pervasive AI & Robotics Lab.
• Curriculum: These units will support EEE 577: Intelligent Control and EEE 401: Mechatronics.
• Impact: 300+ students annually will transition from legacy control theory to modern Embodied AI.
• Ecosystem Shift: This lab will be the first in the region dedicated to the AMD ROCm software stack, training a new generation of engineers to develop on open hardware rather than CUDA-exclusive environments.

Cluster B: Research (5 Units)

• Location: Distributed Research Nodes (Simulated & Physical).
• Function: These units will form the “Lace Network” testbed.
• Workflow:
  • Node 1 (Aggregator): Compiles model weights.
  • Nodes 2-5 (Clients): simulate distinct hospital nodes, training locally on partitioned datasets (Malaria RDT images, Clinical Notes) to validate the federated protocol’s resilience to high-latency African network conditions.

6. Expected Outcomes (Year 1)

1. Sovereign AI Prototype: A working, fine-tuned version of the “Naiden” medical model that respects Nigerian data residency laws.
2. Curriculum Modernization: Full integration of AMD ROCm and PyTorch into the senior-year Robotics capstone projects.
3. Reference Architecture: Publication of a technical blueprint for “The Lace Network,” demonstrating how other universities in the Global South can replicate this high-performance mesh using affordable AMD APUs instead of prohibitive server clusters.

---

For inquiries regarding this proposal or the technical architecture, please contact the Principal Investigator at kayodele@oauife.edu.ng