Hi there,

I am started a little project where I set the VIA C3 Nehemiah to all settings available in SETMUL (multipliers 4 to 14 when FSB is 100 or 4 to 10.5 when FSB is 133, L1 cache L2 cache instruction cache and branch prediction toggled on or off), which amounts to a total of 224+336 = 560 different settings it can be run at, varying from about 8086 speed to a Pentium III Coppermine 700.

My problem is: I need a benchmark to protocol all of the results, in order to find out which settings equal important other settings ( the speed of a 386SX-25, DX2-66 and such).
At first I went with Chris's 3D Benchmark, which gives out an FPS number, but it gave me the exact same result (6.9 FPS) at 4x133 MHZ with instruction cache disabled, instruction cache and L2 cache disabled, and instruction cache and branch prediction disabled.
So that's clearly not accurate enough.

Landmark System Speed Test 6.00 said my CPU was an Intel 80486DX at anything around 600 MHz, sometimes higher MHz despite slower settings while at the same time it claimed a performance of 5000 or so MHz, so tenfold what the CPU was falsely characterized as, so that was out the window instantly.

Topbench gave me a score of 128.5 with instruction cache disabled, but 129.5 when additionally the L2 cache got disabled to that clearly also isn't an option.

DOOM: have seen the DOOM benchmark miss keystrokes on another system of mine, leading to Doomguy walking into walls, getting killed and respawning, getting completely wrong readings.

So here I am, humbly asking for your ideas what benchmark could be accurate enough at low speeds, but not glitch out at high speeds.