that is device-to-device memory bandwidth, computed by
(MHz Memory Clock Speed) x (512-bit Memory Interface) / (8 bits per byte ) x 2 (double data rate).
for example, ASUS ENGTX285
1.3 GHz x 512/8 x 2 = 166.4 GB/s
this number is just a theoretical maximum bandwidth, you should use SDK/bandwidthTest to measure effective bandwidth,
which is upper bound of memory-bound problem.
I don’t evaluate GTX285, but in my experience effective bandwidth is about 0.8 * theoretical maximum bandwidth