hello guyz:
i want to run some socket7 benchmarks of my own, with following cpus being tested:
133mhz(66*2) group: p54c, p55c, 6x86, m2, winchip, k6, k6-2-26050
300mhz(100*3) group: p55c, m2, k6, k6-2-26050, k6-2-26351, k6-2+, k6-3
max clock group: k5-125, p54c-250, winchip-250, k6-2-550, k6-2+550, k6-3+550

and with following benchmarks:
superpi
linx
prime95
winrar
quake(software renderer)
quake2&3
3dmark2000&01

test platform is mvp3 with 2mb cache, 256mb sdram and geforce2ultra.

the problem is, videocards and drivers can affect superpi and other non-graphical benchmarks. for example, for a 6x86-133, superpi without installing videocard driver(unknown vga) or an old 8.05 driver(last of the pre-10 series) scores around 20:50, while a 30.82 driver slows down to 21:50, which makes a whole minute difference. however the 30.82 driver allows 3dmark01 to run on this cpu, while 8.05 hangs up.

now, would you use a same videocard driver for all cpus and benchmarks and live with the slowdowns, or use old driver or no driver to run the non-graphical tests to get the fastest results(which would fail some or all 3d tests after all)? which way would you consider to be fair?