Intel Desktop Port

Intel Desktop Port

Purpose

This page defines the second platform target for OSAI.

Contents

• Why Desktop Second
• Current Bring-Up Gate
• Placement Policy
• Hardware Focus
• Benchmark Focus
• Success Criteria

Why Desktop Second

Intel Desktop is the first real performance target after QEMU correctness. It offers high single-core performance, common developer access, P-core/E-core scheduling challenges, NVMe, and PCIe NIC options.

Current Bring-Up Gate

Milestone 43 adds the first x86_64 Intel Desktop boot path:

• x86_64 UEFI loader build;
• x86_64 kernel ELF build;
• FAT boot image at build/osai-x86_64.img;
• QEMU/OVMF boot through qemu-system-x86_64;
• deterministic serial markers for loader handoff, boot-info validation, COM1 serial, and milestone success.

The local gate is:

• make qemu-x86_64-smoke

This is a correctness gate only. It does not claim Intel Desktop hardware performance.

Placement Policy

Default policy:

• P-cores: AI inference and low-latency network workers;
• E-cores: SSH, logging, source indexing, Git, build/test, background services;
• SMT siblings: disabled or reserved until explicitly enabled;
• migration: forbidden for hot workers.

Hardware Focus

Desktop validation should cover DDR bandwidth limits, PCIe NIC latency, NVMe behavior, thermal sustain, and core isolation under build/test load.

Benchmark Focus

Measure TCP/UDP latency, effective CPU-AI memory bandwidth, sustained usable CPU-core performance, scheduler jitter, core migration, context switches, and page faults.

Success Criteria

OSAI must measure against a tuned Linux/BSD baseline for the target CPU-only embedded app-agent workload. Claims remain design targets until measured.

Concrete acceptance gates:

• make qemu-x86_64-smoke passes for the QEMU x86_64 boot path;
• core_migrations = 0 on hot AI threads;
• post_ready_page_faults = 0 on hot AI paths;
• P-core/E-core placement matches the service manifest;
• SMT policy is documented and measured;
• TCP/UDP latency is compared with tuned kernel TCP/UDP;
• effective CPU-only AI bandwidth is measured separately from raw DRAM bandwidth;
• sustained core performance is measured under build/test interference;
• NVMe and PCIe NIC behavior are included in logs.

Related Pages

• Hardware Readiness Gate
• Scheduler and Core Isolation
• Testing and Benchmarking
• Benchmark Result Format