Results: SiSoft Sandra 2016
The latest version of Sandra recognizes and utilizes most of the cutting edge features of the latest processors. AVX and FMA are integrated into these tests, as well as a handy OpenCL test. While Sandra is not a productivity app, it does have a nice grasp on theoretical performance.
We see some awfully interesting results here. The 7850K and 7670K are very close together. The 7860K is consistently slower than the others, but not by much. The most likely explanation here is that the other chips are staying faster, longer than the lower TDP 7860K. Keeping the turbo speeds up as long as possible obviously helps to finish the workload faster. Once TDP limits are reached, then the chip will slowly start clocking down. In pure INT performance the AMD parts are competitive with the older i3-4330. What I was most surprised at was the very large jump in performance from the i3-4330 to the i3-6100. Yes, the 6100 does run 200 MHz faster, but even if it were downclocked it would be much faster than the other products here still.
The AMD parts, based on the latest GCN architecture, handle the GPGPU portion mostly well. The i3-4330 does not do so great. What I find interesting here is that it does not look like AMD currently supports half precision with these chips, just as the i3-4300 does not either. The i3-6100 appears to support half precision in this case! Intel improves support in this particular GPGPU workload with the i3-6100, but it is still well behind what AMD shows.
We also see the continuing dominance of the Intel parts when it comes to memory streaming numbers. The DDR-4 results are another jump up for the company, even with the lower spec’d DDR-4 2400 modules. Memory speed is still important, but both AMD and Intel institute other mechanisms that help hide the latency and bandwidth limitations of even modern memory modules.