deppman :
First, this is an impressive package. Great performance, and it beats the Tegra K1 at least in raw performance. But there are some points for caution:
1. Although the graphics scores look great, I wonder what the visual fidelity is on the Intel parts. So far, the K1 was had fantastic, full dp rendering just like on the desktop, and has not only outperformed the competition, but all while providing a higher quality image (see Anand). I wonder if the Intel had to make compromises to reach these performance numbers.
2. The cost is insane! It's about 6-7x the cost of a K1. One can buy a complete K1 tablet (miPad, Shield Tablet, the Nexus 9) for about the same or *less* cost than just the Core M board.
3. With Denver and Erista on the horizon, Core M perf and perf/W dominance might not be as long lived as Intel might hope, especially if the next gen parts can get to 16nm, and Denver turns out as impressive as NV hopes.
4. What's the GPGPU capability of this chip? The K1 can use the same CUDA code as used in supercomputers. What's Intel's answer - OpenCL? Is that well supported and competitive?
5. Perhaps the greatest sign of caution for me is the tablet performance demo video. Is the Intel solution so weak that they have to show it against an off-brand chinese A9 tablet? Any one with have an IT neron knows the 8-core aspect is simply a diversion, since the app almost certainly use at most 4 cores. My suspicion is that if the actually competed with post-2011 technology - like a miPad or Shield tablet (4 x A15r2p3 CPU, Kepler graphics with GPGPU acceleration for image processing) they might actually lose or see indistinguishable results. And then spectators might come to the conclusion that only fools would spend a ~$300 premium for little or no difference.
Again, an impressive package, and I might consider a Core-M for my next Linux laptop or desktop. But those are the areas that concern me.
You have to remember Intel is building iGPUs for the compute power more than for actual graphics outputs. That's just a bonus. Intel sees ARM and HSA as a serious threat to its interests and market share of servers/supercomputers/mainframes. Why do you think the top Broadwell SKU promises a whopping 2 Teraflops of performance? Though you can use a K1 for gpgpu compute, it's only rated slightly above Iris Pro 5200 at I think 1.06 Teraflops vs 832 Gigaflops.
There are probably some fidelity compromises, and of course we have to wait for Skylake before Intel puts a tessellation engine on its integrated GPU (which will make it capable of more serious gaming far beyond Broadwell's Iris Pro 6200).
As for perf/watt, you have big x86 cores in there at only 4.5 watt TDP. Intel is not going to let ARM keep claiming its amazing perf/watt numbers, especially when the thermals now say x86 is king by a huge margin. Intel says Skylake is again mobile-oriented, so I expect 3-watt parts by the end of 2016, and I expect ARM's TDP to continue creeping upwards as it has the last few years. Without better branch prediction they're against a brick wall to get more performance. It's either boost clocks or boost IPC, which requires good branch prediction, and the best ARM offering is 34% accurate on the AMD Opterons bs. Intel's now (I believe) 79% accurate branch predictor. That circuitry is thermally expensive, but boosting clocks is electrically AND thermally expensive. ARM can either make individual instructions take a bit less time (not much room on that given how rudimentary the instruction set is in the first place) or they improve the pipelining and branch prediction.
I don't know what the gigaflop rating is on the K1, but if Broadwell's top SKU has the claimed 2.03 Teraflops (GPU only) at 1GHz on 96 EUs, I suspect Core M, with 24 EUs, has about 440Gigaflops peak given the 850MHz boost clock, plus the 200 Gigaflops from the CPU cores themselves and their numerous SIMD units (4 per core).
OpenCL is used a ton in the supercomputer world. Intel's Xeon Phis outnumber Nvidia's Teslas by a huge margin due to their perf/watt advantage and raw performance advantage.
I'm assuming the Intel tablet had a very good aluminum backplate and heat spreader, but I'd also bet it has the graphics power of mullins easily. I somewhat doubt it can match the K1, but most tablet apps are still built on x86, so Intel has that advantage too.