The Intel DAP box is not delivering the 2 instructions per cycle throughput in my simple micro benchmark. The same problem was previously observed on knl00 rack mounted system before the BIOS update, which has fixed the problem. More details. A program with a known mixture and number of independent instructions should provide the peak 2 IPCs throughput. This was previously confirmed on the knl00 machine with 1, 2 and 4 threads running on a core: knl00: 1 thread per core: 4512: optimized time 405239 result 0.00000000 IPC = 1.9911 SF/cycle = 58.2539 Peak rate = 91.0217 4528: optimized time 408436 result 0.00000000 IPC = 1.9825 SF/cycle = 58.0029 Peak rate = 90.6296 4544: optimized time 400927 result 0.00000000 IPC = 2.0268 SF/cycle = 59.2980 Peak rate = 92.6532 2 threads per core: 4512: optimized time 405415 result 0.00000000 IPC = 1.9902 SF/cycle = 58.2286 Peak rate = 90.9822 4528: optimized time 408624 result 0.00000000 IPC = 1.9816 SF/cycle = 57.9762 Peak rate = 90.5878 4544: optimized time 401053 result 0.00000000 IPC = 2.0262 SF/cycle = 59.2794 Peak rate = 92.6241 4 threads per core: 4512: optimized time 407851 result 0.00000000 IPC = 1.9783 SF/cycle = 57.8808 Peak rate = 90.4388 4528: optimized time 408742 result 0.00000000 IPC = 1.9810 SF/cycle = 57.9594 Peak rate = 90.5616 4544: optimized time 401151 result 0.00000000 IPC = 2.0257 SF/cycle = 59.2649 Peak rate = 92.6015 Same executable on Intel DAP box: 1 thread per core: 4512: optimized time 461229 result 0.00000000 IPC = 1.7494 SF/cycle = 51.1823 Peak rate = 79.9724 4528: optimized time 464888 result 0.00000000 IPC = 1.7418 SF/cycle = 50.9596 Peak rate = 79.6243 4544: optimized time 456222 result 0.00000000 IPC = 1.7811 SF/cycle = 52.1110 Peak rate = 81.4234 2 threads per core: 4512: optimized time 461132 result 0.00000000 IPC = 1.7498 SF/cycle = 51.1930 Peak rate = 79.9891 4528: optimized time 464981 result 0.00000000 IPC = 1.7414 SF/cycle = 50.9494 Peak rate = 79.6084 4544: optimized time 456272 result 0.00000000 IPC = 1.7809 SF/cycle = 52.1053 Peak rate = 81.4145 4 threads per core: 4512: optimized time 461073 result 0.00000000 IPC = 1.7500 SF/cycle = 51.1996 Peak rate = 79.9993 4528: optimized time 464830 result 0.00000000 IPC = 1.7420 SF/cycle = 50.9659 Peak rate = 79.6343 4544: optimized time 456215 result 0.00000000 IPC = 1.7812 SF/cycle = 52.1117 Peak rate = 81.4246 Same issue is confirmed when running HACCmk. On knl00, it is delivering 67% of peak: 2944: optimized time 858274 result 1354864.52650320 IPC = 1.8052 SF/cycle = 42.6229 Peak rate = 66.5983 2960: optimized time 860064 result 1367795.59051550 IPC = 1.8112 SF/cycle = 42.7654 Peak rate = 66.8209 2976: optimized time 863812 result 1380780.63478190 IPC = 1.8131 SF/cycle = 42.8099 Peak rate = 66.8905 2992: optimized time 865702 result 1393825.84251841 IPC = 1.8189 SF/cycle = 42.9462 Peak rate = 67.1034 Intel DAP provides less then 60% of peak: 2944: optimized time 982627 result 1354864.52650320 IPC = 1.5767 SF/cycle = 37.2289 Peak rate = 58.1702 2960: optimized time 989038 result 1367795.59051550 IPC = 1.5750 SF/cycle = 37.1886 Peak rate = 58.1072 2976: optimized time 991330 result 1380780.63478190 IPC = 1.5799 SF/cycle = 37.3032 Peak rate = 58.2862 2992: optimized time 995241 result 1393825.84251841 IPC = 1.5821 SF/cycle = 37.3563 Peak rate = 58.3693 In these results, the time is measured in cycles, therefore the possible differences in frequency should not lead to performance difference. Vitali