The last few generations of GPU architectures have been increasingly optimized for massive throughput of low-precision integer arithmetic operations, though, which are not useful for any of those other applications.