I've been working on a Direct3D Immediate Mode project. Running in 640x480x16bpp I get a respectable 60 fps on my machine (Celeron 450, 128 Mb RAM, TNT2 w/32 Mb). On a lower-end test machine I get 30-35 fps (Pentium-II 233 Mhz, 64 Mb RAM, Voodoo Graphics 1 w/4 Mb). These are good results. On my friend's AMD K6-2 (450 Mhz, 196 Mb RAM, Voodoo Banshee with 8 Mb) things slow to a crawl: 18-20 fps.

Is this because of the K6-2's notoriously slow FPU? If so, is there a good way to minimize its impact? Much of the render code uses floats (since they are Direct3D's instrument of choice).

Any advice would be appreciated.

K6-2 supports 3DNow!, so if you optimize vertex buffers containing static geometry, D3D's TnL code will take advantage of that.

Use lower resolution models, less vertices to transform.

Prevent retransformation if you're doing it (for things like multipass multitexturing) by using ProcessVertices(), and then reusing the buffer.