Some people may remember that I rebuilt my old Windows 98 PC from 1999. It has the following parts:

ATX Standard beige case - like any other. It has 4x 5.25" bays, 3x 3.5" bays - two of them external)
ATX Power Supply ~450 watts (modern)
ASUS P2B Revision 1.02 (Intel 440BX)
Western Digital Caviar WD400 40GB PATA IDE hard drive
IOMEGA ZIP 100MB ATAPI PATA internal drive
Standard 3.5" 1.44MB Floppy drive
CD-RW ATAPI drive
PS/2 Keyboard
PS/2 Mouse
Gravis GamePad Pro (gameport)
SMC PCI 10/100 Ethernet card

The graphics cards used in this rig and for this benchmark test:

AGP 2x: Diamond Viper V770 (nVidia Riva TNT2) 32MB AGP
PCI1: STB V2-1000 Voodoo2 12MB PCI 3D Accelerator
PCI2: STB V2-1000 Voodoo2 12MB PCI 3D Accelerator

The Voodoo2 cards are in SLI using the 3dfxzone.it FastVoodoo2 4.6 suggested by other forum members here. These drivers are the only ones I have been able to use that got SLI working completely stable.

The Processors for this test:
1: Intel Pentium II Deschutes 350MHz
2: Intel Pentium !!! Katmai 600MHz

Both are using the same 2.0V Slot 1 and 100MHz frontside bus, so the only thing that changed is the CPU.

Why am I doing this bench? Because lots of people will bench different video cards with the same CPU. I've only ever seen Phil's Computer Lab run a CPU comparison with the same cards. And considering how much the CPU is used with these older cards (before the GPUs from nVidia and ATI started coming out in the GeForce/Radeon lines), a CPU bench is just as important as a graphics card bench for comparison.

For the results, I will format them with the P2 first and the P3 second in italycs. I'll then add a % increase in bold per line.

VOODOO2 SLI 640 x 480
Quake 1 Timedemo1: 99.1 FPS 105.0 FPS 5.9%
Quake 1 Timedemo2: 98.6 FPS 103.0 FPS 4.5%
Quake 1 Timedemo3: 93.1 FPS 101.3 FPS 8.8%

VOODOO2 SLI 640 x 480 x 16
Quake 2 Timedemo1: 73.1 FPS 111.7 FPS 52.8%
Quake 2 Timedemo2: 67.5 FPS 104.3 FPS 54.5%

DIAMOND VIPER V770 640 x 480 x 16
Quake 2 Timedemo1: 57.7 FPS 84.8 FPS 47.0%
Quake 2 Timedemo2: 55.2 FPS 80.3 FPS 45.5%

DIAMOND VIPER V770 800 x 600 x 16
Quake 2 Timedemo1: 54.1 FPS 60.5 FPS 11.8%
Quake 2 Timedemo2: 51.3 FPS 58.1 FPS 13.3%

DIAMOND VIPER V770 1024 x 768 x 16
Quake 2 Timedemo1: 35.7 FPS 36.0 FPS 0.8%
Quake 2 Timedemo2: 34.1 FPS 34.6 FPS 1.5%

3DMark 99 Max 3DMarks / CPU 3DMarks
Voodoo2 640x480x16: 2737/3129 4598/8250 68.0%/163.7%
Voodoo2 800x600x16: 2733/3122 4549/8229 66.4%/163.6%
Voodoo2 1024x768x16: 2721/3129 3837/8239 41.0%/163.3%
Riva TNT2 640x480x16: 2811/3127 5165/8238 83.7%/163.4%
Riva TNT2 640x480x32: 2817/3123 5041/8200 78.9%/162.1%
Riva TNT2 800x600x16: 2815/3129 4965/8189 76.4%/161.7%
Riva TNT2 800x600x32: 2815/3126 4342/8249 54.2%/163.9%
Riva TNT2 1024x768x16: 2811/3129 4156/8229 47.8%/163.0%
Riva TNT2 1024x768x32: 2508/3127 3116/8153 24.2%/160.7%

So on the CPU bench score, there's a consistent ~162% increase in power for only a 71.4% increase in clock speed.. The video card benches all went up accordingly, though not quite so high.

My findings suggest that Quake 1 is already limited by the graphics engine itself, as the increase in CPU speed had a negligible effect on the framerates during the time demo. I only tested Quake 1 via the 3dfx miniGL mode with GLQuake.

Quake 2 was more configurable within the game itself, so I had more options. It also had a sizeable jump in framerates consistent with the suggested 3DMark scores until we hit 1024 x 768, where the graphics card hits a bottleneck or a cap, whichever. There was only the stated 1% increase in framerate at that resolution, even though the benchmark suggested there should have been a ~30% increase.

Merry Christmas 2016!