We asked some people inside AMD how to actually get the 1.2 TeraFlops number and we got a simple mathematical explanation.

RV770, as you all know by now, has 800 Shaders. Each one of the 800 processors is a Floating Point Multiply and Accumulate unit (FMAC) and in each cycle each Shader can perform a multiply and an adder instruction.

Because it can perform both of these operations in a single cycle, this is counted as two floating-point operations. This gets you to the following equation.

2 (Floating point instructions per processor) x 800 (processors) x 750 (Engine clock in MHz) = 1200 GLFOP = 1.2 TFLOPS.

Quite logical, isn’t it?