本帖最后由 gtx9 于 2017-7-8 17:00 编辑
intel 官方的测试,算上核心数的差异
LINPACK 的AVX512效率是80%左右
Through its integration of the Intel® Advanced Vector Extensions 512 (Intel® AVX-512), the platform generates 2X FLOPs/clock-cycle peak improvements, offering a boost to performance for demanding use.1 Intel AVX-512 combined with improvements in cores, cache and memory, delivers up to 2.27x more performance than today’s Intel Xeon processor E5 v4 (formerly codenamed Broadwell), and up to 8.2x more double precision GFLOPS/second when compared to a 4-year old Intel Xeon processor E5 family in the installed base.
Baseline config: 1-Node, 2 x Intel® Xeon® Processor E5-2699 v4 on Red Hat Enterprise Linux* 7.0 kernel 3.10.0-123 using Intel® Distribution for LINPACK Benchmark, score: 1446.4 GFLOPS/s vs. estimates based on Intel internal testing on 1-Node, 2x Intel Xeon Scalable processor (codename Skylake-SP) system. Score: 3295.57
Baseline config: 1-Node, 2 x Intel® Xeon® Processor E5-2690 based system on Red Hat Enterprise Linux* 6.0 kernel version 2.6.32-504.el6.x86_64 using Intel® Distribution for LINPACK Benchmark. Score: 366.0 GFLOPS/s vs. 1-Node, 2 x Intel® Xeon® Scalable process on Ubuntu 17.04 using MKL 2017 Update 2. Score: 3007.8
|