No you don't, you use an 8-bit ECC value for each 64-bit data word.
While this requires just as many extra memory bits as does classic 8+1 parity, the math is different. It allows detecting more errors, and even auto-correcting some.
8+1 can detects single-bit errors, misses many multiple-bit errors, and allows no corrections.
64+8 allows detecting and correcting all single- and some multi-bit errors, and detects any multi-bit errors.
Regards, Peter