According to NForce2 config guide
http://www.planet3dnow.de/artikel/diverses/nf2config/3.shtml#config_apic
is is not recommended to use APIC mode with NForce2 motherboards.
I switched from APIC to PIC following instructions it that config guide.
Performance dropped abt 5 % in CPU intensive applications, games stuttering etc. TaskInfo 2003 reported that 3 % of CPU power was wasted in Interrupts Time thread, even with no activity.
Switched back to APIC (rollback driver and BIOS setup) and the problem was gone.
My IRQ settings with APIC are as follows:
0 : System timer
1 : Standard 101/102-Key or Microsoft Natural PS/2 Keyboard
6 : Standard floppy disk controller
8 : System CMOS/real time clock
9 : Microsoft ACPI-Compliant System
12 : NVIDIA nForce PCI System Management
13 : Numeric data processor
14 : Primary IDE Channel
15 : Secondary IDE Channel
18 : Silicon Image SiI 3112 SATARaid Controller
19 : RADEON 9800 XT
20 : NVIDIA(R) nForce(TM) MCP Audio Processing Unit (Dolby(R) Digital)
20 : Standard OpenHCD USB Host Controller
21 : NVIDIA(R) nForce(TM) Audio Codec Interface
21 : Standard Enhanced PCI to USB Host Controller
22 : NVIDIA nForce MCP Networking Controller
22 : Standard OpenHCD USB Host Controller
That makes 17 IRQ, which is absolute minimum with this configuration. All unnecessary devices are disabled. Use of APIC somehow gives smoother performance, and this is also seen in games, and 5 % better performance (vs PIC) in CPU tests (Sandra2004, CPUMark).
Sometimes switching into PIC mode _after installing OS with APIC mode_ may be good idea, but at least in my case it deteriorated performance.
System: Abit NF7-S 2.0, K7 Barton 2500+ @ 3200+, 1 Gb KHX 3500, Radeon 9800 XT, 120 Gb Hitachi Sata drive w/ 8 mb buffer.
Earlier I had a reduced system with EPOX RDA & ThB 2100 & Gf4 Ti4600, with clean install of XP in PIC mode. Everything worked smooth. I also tried clean installs with APIC and PIC, and there were no measurable difference in CPU or grapic performance. I/O subsystems were not tested.