Hi. Are there any specs for fp64 support on Kabini's GPU in terms of speed? A slide posted at Anandtech lists 8 fp64 ADDs per cycle per CU which looks fine. However, they also list 4 DP FMAs every 16 clocks on each CU. That means a total throughput of only 0.5 DP FMAs per cycle (4/16 FMAs per CU per cycle* 2 CUs) for the GPU. Is this correct?
Please post your clinfo output. My clinfo seems to be supporting DP for iGPU.