I am using this repository to try out a few approaches to later PC emulation.
The first target is a machine I wanted in 2007: a Core 2 system with a GeForce 8800 GTX, running Windows XP and the original drivers. The point is not simply to run old XP games. I want the machine itself to feel convincing, from setup and Device Manager through to 3DMark06.
There are several possible ways to get there. QEMU may be the right base. Parts of 86Box may still be useful. The graphics card may need a separate hardware model with a translated rendering backend. I do not want to choose between those options until small experiments have shown what works.
Each experiment starts with a GitHub issue that states what I want to find out, how I will test it, and what would make me stop. Results belong in the issue and in a short report under experiments/. Larger decisions are recorded under docs/decisions/.
The work started in four paths:
- Define the exact machine and what counts as convincing.
- Measure how close stock QEMU gets before changing it.
- Find out how far the original ForceWare driver gets with a minimal G80 device.
- Check whether a small host-time CPU experiment in 86Box avoids a much larger rewrite.
See docs/target.md for the current target and docs/testing.md for the common test ladder.
This repository will not contain operating-system media, product keys, proprietary drivers, VBIOS images, firmware, benchmark installers, or virtual-machine disks. Tests that require them will document what the user must supply and how the files are identified locally. The first set of period-correct sources and hashes is recorded in docs/media-sourcing.md.
Small prototypes may live here at first. Changes that clearly belong to 86Box or QEMU will move to branches in the appropriate fork and link back to the issue that motivated them.
QEMU is the primary probe platform. The first G80 device shell is now running in QEMU as a secondary display adapter. Its PCI identity, apertures, capabilities, reset behavior, and interrupt paths are covered by focused tests. The 86Box path is closed for this phase because a CPU timing experiment would not remove the larger Core 2, chipset, interrupt, and PCI Express gaps.
The architecture choice and its revisit conditions are recorded in decision 0001. The G80 device-shell result, QEMU host baseline, and 86Box source review are under experiments/.