Based on the amd sdk app3.0 , the global memory benchmark program from amd:
my test results shows:
fury 's float4 random performace is
225.512 Gbps
225/512 = 44% of the peak bandwidth.
mean while , the 280x 's float4 random formance is
192.248 Gbps.
192/288 = 66% of the peak bandwidth.
Does this a driver bug or the 1st generation HBM defect?
the opencl drvier is 15.30.1025 , the latest one on ubuntu 14.04.