Intel NUC NUC6i7KYK Skull Canyon Mini PC Review

Jump To:

NUC6i7KSYK General Performance Testing

cinebench-corei7

The Intel NUC Kit NUC6i7KYK with the Intel Core i7-6770HQ processor scored 86.07  FPS on the OpenGL benchmark, 711 points on the multi-core CPU test and 130 points on the single-core CPU test.

3dmark-nuc-skydiver

In 3DMark we scored 8,596 points in Sky Diver, the benchmark for mid-range PCs, specifically 37.29 FPS in Graphics Test 1 and 39.58 FPS in Graphics Test 2.

x264-corei7

On the x265 HD Benchmark 64-bit video encoding benchmark we found the NUC with the Intel Core i7-6770HQ quad-core processor was capable of roughly 15 FPS with regards to the average frame rate for encoding HD (1080p) video in the new H.265 / HEVC format.

handbrake-corei7

We ran Handbrake 0.10.5 and used the Big Buck Bunny 1080P 60 FPS movie as our test file. We were able to complete the transcode with an average speed of 42.2 FPS in 15 minutes and 3 seconds.

aida64-nuc

AIDA64 v5.70 beta showed that the dual channel 2400MHz DDR4 memory had 14-14-14 2T timings and was capable of 41,489 MB/s read and 60,580 MB/s write speeds with a latency of 59.7 ns. The copy test bandwidth was found to be 57,528 MB/s. These are very respectable scores and the highest that we have ever seen on an Intel NUC model since we aren’t limited to 2133MHz by a dual-core processor. If you plan on purchasing this model we highly suggest going with a 2400MHz DDR4 SO-DIMM memory kit.

cpu-z

cpu-z-bench

The benchmark included in CPU-Z showed we had a single thread performance score of 1630 and a multi-thread performance score of 6203.

cipher-corei7

CipherShed v0.7.4.0 shows that the Intel NUC6i7KYK had a mean score of 3.9 GB/s on the AES Encryption test.

jetstream-corei7

We ran the JetStream v1.1 JavaSctipt on Google Chrome 50 and found an overall score of 195.77.

7zip-corei7

The 7-Zip 15.14 benchmark test showed that the Intel Core i7-6770HQ Processor in this system had a total rating of 20348 MIPS and handled compression tasks well.

cdm-i7

A quick run of CrystalDiskMark v5.1.2 showed the performance of the Samsung SSD 950 Pro 512GB storage drive was pretty solid. Sequential read speed to be 2550 MB/s and the sequential write speed was 1535 MB/s. The Random 4K read speed was 56 MB/s and the 4K random write speed was 215.4 MB/s.

wireless-performance

When it comes to wireless performance the we used LAN Speed Test to check the performance of the included the included Intel Dual Band Wireless-AC 8260 802.11ac PCIe solution. This is a dual-band 2×2 802.11ac Wi-Fi card with a theoretical peak speed rating of 867 Mbps. We used a desktop with Gigabit Ethernet to run LAN Speed Server that was hard connected to the ASUS RT-AC68U 802.11AC wireless router with Firmware version 3.0.0.4.378.9313. We tested the 2.4GHz and 5GHz performance with the router placed 10-feet away from the Intel NUC NUC6i7KYK and ran performance tests with 1MB and 100MB packet sizes. The results showed that we were averaging 250-300 Mbps (Up to 37.5 MB/s) with regards to writes and 470-480 Mbps (Up to 60 MB/s) for reads on the 5GHz band.

Video Playback

Since many Intel NUC  Kits will be used in the living room we thought we’d take a second and look at a number of video types on the device to see how it would perform as a HTPC as many people might be looking into buying this and running Kodi. We took a quick look at DXVA checker and found that the Skylake processor with Intel Iris 540 Graphics offers hardware acceleration for MPEG-2, VC1, H.264, HEVC, VP9 (8-bit only) and WMV9. When it comes to audio bitstreaming the device supports DTS, Dolby Digital and Dolby Digital Plus.

 dxva-corei7

UPDATE 5/23/2016: A reader in the comment section requested that we run Intel SGEMM on the NUC6i7KYK, so we loaded up Microsoft VS2013 Update 5 and looked at the kernel performance on the 6thGeneration Intel Core i7-6700HQ Processor with Intel Iris Pro Graphics 580, which contains 72 EUs running at 950MHz. With 8  shader cores per EU it means that there are a total of 576 shader cores. Intel says that the compute performance of the Iris Pro Graphics 580 is 1.1 TFlops and that it is up to 75% faster than the previous generation Broadwell Iris Pro 6200 performance (GT3e with 48 EUs / 384 shader cores). The block_read_32x2_1x8 results for a 2048×2048 matrix show a 772  GFLOPS  for the peak kernal performance.  That is just over 70% compute efficiency.

SGEMM

Let’s take a look at power consumption and CPU temperatures on the Intel NUC NUC6i7KYK.

Print
Jump To:
  • Intel says it supports DDR4 2133+, but mostly it talks about simply 2133. Are there benefits from using 2400?

  • Joe Schmoe

    Still waiting on the 6770HQ to be released for 3rd parties… especially for the triple-fan Z Canvas. At least that thing would keep this from ever throttling.

  • B Brad

    Not terrible, but the GPU isn’t as fast as I expected. Anyone seen numbers to compare the Skylake integrated graphics (Intel Pro 540/580) to the previous best integrated GPU in the broadwell (i5-5675c and i7-5775c)? LegitReviews has a i5-5775c review, but don’t cover any of the same games for an apples to apples comparison.

    • Nathan Kirsch

      What would you like to see exactly? We have an office PC running a 5775C, so can run some benchmarks for you if you’d like.

      • B Brad

        Recreating any of the above published game benchmarks would be great. I just want a general idea of how the new skylake iris pros compare to the previous generation equivalents with intel’s in package video memory. From what I can tell the new iris pro 580 isn’t delivering the expected performance increase that you would expect from the published specifications.

        • Mariano Ruiz

          Did you find any direct comparison between the Iris 580 and the 6200?

          I’ve done some research but couldn’t find any direct comparison. I would say they are very even, even though the Iris 580 has more EUs.

  • KurtKrampmeier

    Can Undervolting achieve significantly better thermals and less
    cpu throttling? And if so, by how much? I want to use this as a 24/7
    load and very small and light portable cpu package. Thank you!

  • Charles Borner

    “This is the first Intel NUC to feature a mighty Intel Core i7”

    Uh. No. No it isn’t.

    The 5th Gen BOXNUC5I7RYH came with an i7…

    • Nathan Kirsch

      You are correct… I forgot to put in there ‘quad-core’ after Core i7! Thanks!

      • Charles Borner

        Ah HAAAAAA!!!!

        Don’t ever let it happen again!

        =)

  • Paul

    I wouldn’t recommend these Intel NUC’s. Filled with Bios bugs and hardware failures. Intel should just stick to what they are good at. Making cpu’s.

  • Can you try running the decent and non-naive Intel SGEMM example on the Skull Canyon? It’s a quick compile.

    https://software.intel.com/en-us/articles/sgemm-for-intel-processor-graphics#_Toc419731380

    Results for 64-bit VS2013 build and .4444 drivers

    A Core i7-4790 HD4600 peaks at ~280 GFLOPS which is ~75% of theoretical (160 FMA @ 1.2 GHz)

    A NUC 5i5RYH HD6000 peaks at ~480 GFLOPS which is ~66% of theoretical (384 FMA @ 950 MHz)

    • Nathan Kirsch

      Got some KISS (keep it stupid simple) instructions as it looks like there many options you can run and don’t see any examples online on how to run it. I’ve installed VS2013 Update 5 and am running .4444 drivers.

      • You just run it with no arguments (“C:…> GEMM.exe”) and it will iterate through all of the possible kernels. The article identifies the kernels that are expected to be more performant. For example, The “block_read” kernels perform more efficient loads using an Intel OpenCL extension.

        The best result for each matrix size is probably a fair measure as it seems pretty consistent.

        • Nathan Kirsch

          send me an e-mail to nate at the sites url if you could

        • Nathan Kirsch

          Here are the results that I worked out with Allan for SGEMM! Will add these to the review for all that are interested in compiler performance of Intel Iris Pro Graphics 580!

        • Achieving over 70% of theoretical (1094.4 GFLOPS) and nearly reaching 800 GFLOPS with such a clean OpenCL implementation is a very respectable result.

          One of the neat things about the HD580 is that it can also perform lower resolution 16-bit floating point operations with doubled throughput over 32-bit floats.

          Perhaps Intel will update their benchmark to reflect this as fp16x2 support is quite interesting to some people.