-
Notifications
You must be signed in to change notification settings - Fork 0
ARM NVIDIA Port
This page defines the fourth platform target for XAI OS: ARM/NVIDIA N1X-compatible and GB10-class systems.
- Why ARM/NVIDIA Last
- Platform Assumptions
- Boot and Interrupts
- Memory Behavior
- Core Policy
- Success Criteria
ARM/NVIDIA comes after x86 QEMU, Intel Desktop, and Intel Xeon because it needs a stable OS architecture before platform-specific AArch64 work. The goal remains CPU-only AI, not CUDA or GPU dependency.
Target assumptions:
- AArch64;
- UEFI/SBSA where available;
- Device Tree fallback later;
- GIC interrupt controller;
- SMMU for DMA protection;
- LPDDR or unified memory behavior;
- mixed core or cluster topology.
Generic AArch64 work should include exception levels, page tables, timers, GIC, cache maintenance, atomics, and ABI boundaries. NVIDIA-specific work should be isolated to platform discovery, firmware quirks, board-level device setup, and any documented memory-controller behavior.
The first ARM path should prefer UEFI plus ACPI/SBSA where possible. GIC support and timer support are required before meaningful service work.
LPDDR/unified memory systems need explicit bandwidth measurement. Shared model weights, prefaulted arenas, and no-swap AI memory remain required.
Default policy:
- highest-performance cores: AI inference;
- middle cores: network and tokenizer workers;
- efficiency cores: SSH, logging, Git, source indexing, build/test;
- cluster migration forbidden for hot AI workers.
XAI OS boots, runs AI Cell lifecycle tests, keeps hot workers on intended cores, and reports platform-specific memory and latency counters.
Additional criteria:
- no CUDA, GPU, NPU, or accelerator runtime is required;
- GIC interrupt routing avoids hot AI cores;
- SMMU protects DMA into AI Cell memory;
- LPDDR/unified-memory bandwidth is measured;
- cluster/core policy is visible in telemetry.
- Vendor firmware may hide memory-controller and power controls.
- UEFI/SBSA support may be incomplete.
- Device documentation may lag hardware availability.
- Fallback plan is to keep generic AArch64 boot and QEMU-style driver work independent from board-specific optimization.
This page defines the GitHub Wiki navigation sidebar.
- Architecture
- AI Cells
- CPU AI Runtime
- App Agents
- Memory System
- Networking
- Scheduler and Core Isolation
- Filesystem and Storage
- Driver Model
- Security Model
- Build System
- Build System
- Project Tracker
- Implementation Plan
- QEMU Full OS Core Workdown
- QEMU 100 Completion Plan
- Example Apps
- Codex Work Packages
- Testing and Benchmarking