gpu architecture is the only reason our company's "ai solution" doesn't melt the datacenter. cpu people need to take a long hard look at themselves