In benchmarking 2D graphics accelerators, we've found that the ability to quickly copy data from the host processor to the screen is a critical operation. Many benchmarks focus on testing all kinds of BLTs and ROPs and other specialized graphics operations. While these are important, actual Windows applications perform a lot of host to screen transfers.

We wrote two simple benchmarks that you can download in the in the (500KB) file - PhotoStones and Cascade.

PhotoStones simply copies multiple true-color images from system memory to the screen. It measures the number of pixels-per-second that the system is achieving. From past experience, ISA bus computers will achieve about 1 MPix/sec, EISA about 2 MPix/sec, VL-Bus about 4 MPix/sec, and PCI about 12 MPix/sec. The theoretical peak for PCI/33 should be 44 MPix/sec. This maximum assumes the system handles 24 bit true color intelligently and that the graphics card and computer are perfect. We haven't found one yet.

Cascade (also in the file) pretends to be the very end of Windows Solitaire - if you double click the obvious king, the cards will "cascade" off the screen. This test shows the speed of both color and monochrome bitmaps being sent from the processor to the screen. Hint: notice the speed of the black non-face cards - they're mono-bitmaps. The remaining cards are color-bitmaps. Windows uses mono-bitmaps more than you'd think. Cascade doesn't produce any kind of timing number - it's more of a gee-whiz to look at.