Gemini Lake Pentium NUC Review (NUC7PJYH) – Performance (2/3)

In the second part of our Gemini Lake Pentium NUC review we run some popular benchmarks on it to get an idea how does it perform compared to the other recent NUCs. If you came here via a search engine, I’d suggest you take a look at the first part of the review first. We’re going to run our standard setup which means 3DMark, Cinebench R15 and Geekbench 3.

I installed a recent version of Windows 10 on the NUC. During the installation the WiFi adapter was not detected so I had to use the wired Ethernet connection for downloading the updates and drivers from Intel’s Download Center. Driver installation was painless this time. It seems that the CPU-Z utility in the above picture detects the CPU incorrectly – this will likely be fixed in future versions of CPU-Z. Thanks to the relatively powerful quad-core CPU the NUC7PJYH does quite well as a Windows 10 desktop PC. Multiple tabs or several applications open, it works better than any low-end NUC before.

Benchmark Results

Intel UHD Graphics driver version, Windows 10 and BIOS version 0037 was used for the following benchmark results. The unit was equipped with 2x4GB Kingston KVR24S17S6/4 memory modules and a Drevo 64 GB SSD drive.

Cinebench R15

Cinebench runs 3 separate benchmarks and gives figures that are comparable between systems. First a simple 3D car chase that measures mainly the GPU (OpenGL) performance. The result is in frames per second. After that there’s a rendering of 3D model with all cores. This stresses purely the CPU. Finally there’s another rendering of the same model, this time using just a single CPU core.

In the multi-core CPU tests the NUC7PJYH does pretty well. It actually beats both i3 NUCs from last 2 years. Both i3 models are dual-core with hyperthreading whereas NUC7PJYH has actually 4 cores, but no hyperthreading. When comparing to NUC7CJYH it’s worth noticing that single core performance is more or less identical, but multi-core performance approx. 100% better. Basically J5005 has 2 more identical cores compared to J4005.

In the OpenGL tests both previous i3 models pull ahead of the NUC7PJYH as the 18 EU GPU just cannot match the one in the i3 models. In this test the Pentium-powered NUC7PJYH beats the Celeron-based NUC7CJYH by 33%.


3DMark is a popular benchmarking suite that benchmarks video and gaming performance of the computer. I had some problems running the benchmarks as I got a black screen initially. I found out that if I plugged my screen in the HDMI connector 1 instead of HDMI 2 it worked just fine. Good to keep that in mind in case you face issues with the display.

In the 3DMark Cloud Gate test the NUC7PJYH is significantly faster than the lower end NUC7CJYH but significantly slower than the previous generation i3 models.

Sky Diver results tell a similar story.

Things don’t change much in Fire Strike either.


Below you can find the 64-bit Geekbench 3 score for NUC7PJYH. 6854 for multi-core and 2182 for single-core tests. Single-core performance is again identical when compared to NUC7CJYH but multi-core result is roughly twice as good.

Passmark CPU Benchmark

In the Passmark CPU test I got result of 2967. This is significantly higher than the NUC7CJYH score (1653).

Fan Noise

The default BIOS settings kept the fan running at very low levels even during the CPU intensive benchmarks. There’s also an option to completely turn off the fan when the temperatures are below a certain limit. This is enabled by default as well. I’d say that the box was almost totally silent.

Thoughts of the Performance

If the lower end Gemini Lake NUC left something to be desired I’d say that this Pentium Silver J5005 powered NUC surprised me nicely in the tests. The CPU itself showed pretty strong performance beating previous generation i3 models nicely. This is mainly due to being a quad-core CPU vs. dual-core with hyperthreading. Not all applications are able to benefit from this, but then again, having 4 cores should benefit desktop environment where you want to run multiple applications at the same time. The price difference between NUC7CJYH and NUC7PJYH is relatively small – about $50. I’d say the performance difference is worth that.

Read Further

Be sure to read also the earlier parts of this review!

11 Responses

  1. Tomm says:

    Multicore results are quite nice, single core pretty much same with CJYH (dual core vs PJYH quad core)

  2. Pawel says:

    I am in market for one of these and miss direct comparison to my PPYH. in your review.
    I mean what do I gain (besides new instructions/codecs built into chip)?
    It would be nice if you put 3200 sticks in 7PJYH. Bruce (from NUC team) told us new Skull (2018) can easilly take 3800.
    Why not push it (review) to its limits? Put Samsung 850 Pro, G.Skill 4000/18 x2 and show us how fast it can really be.
    …. other sites! You/WE should ask Intel for a demo unit for you, ask G.Skill for fastest sticks, ask Samsung for 850 Pro and then do the testing. nucblog is the nuchub not others.
    Your aproach feels more like illustrated technical / white papers.

    As for Linux:
    I’d like to know (b/c I need to know some things before I recomend unit to my friends clients) how fast it boots, how fast LibreOffice starts…ShotCut… Gimp… Dark Table. Make some projects and repeat the procedure with every NUC you test. We nedd to move forward with these reviews.

    I appreciate your work, dont get me wrong, but we need standarisation of these reviews.
    Make some petition for us to sign so you could send it to Intel.

    Ps. Before buying Galactico I’m buying 4000/18 for 7KYK just to max it out with 960Pro x2.
    Pps. How much faster is browsing Internet (Windows 10Firefox w/uBlock) using 7PJYH compared to PPYH?
    Whats the screen refresh/redraw experience @4K in say Photoshop compared to any OpenGL card (Ati W2100 to be the cheapest 10 bits one).

    Again: not rude but begging for a change

    • Olli says:

      Hi Pawel,

      I know you as a long time commentator here, so by no means I take offense on your constructive feedback! It’s a long feedback you wrote and I need to think about some of the things you say here. It’s just a hobby for me to write the blog and while it has given me a lot of interesting contacts and made me look into various products more than I’ve ever done before, it’s still just a hobby for me.

      I get very little hardware from the vendors (mostly they just ignore me), so trying out all kinds of different memory and SSD drives would require me to actually invest something into a collection of hardware that probably will be obsolete in a year or two. Also, Samsung for example borrowed me the 960 drives a year ago, but they wanted them back after a couple of weeks. Most of the things you see on review here, I actually buy and sell later on when the novelty has worn out (or I just need the money for something else :) ). I know many people would like to see a dozen of games tested on the more powerful NUCs, but the hard fact is that as long as each one costs €50/$50 I probably will not buying them just for the sake of the review.

      I do agree that sometimes reading my own text I feel like reading a white paper and I guess this is where my engineering background shows, as well as the fact that English is not my native language.

      I guess what I’m trying to say is that it requires a lot of work as it is now and I do have a daytime job that currently requires me to take a flight to another country once a week. I do recognize the needs for improvement and growth that you describe, but I may not have the resources for all this at the moment. However, I’ll put your words in the back of my mind and keep thinking about them. In the mean time, if any of you have any concrete improvement proposals (let’s say: “here’s a script that you can run on freshly installed Ubuntu 18.04 desktop that prints out the information I’d like to have from each NUC.”) I’d be happy to welcome them.

      In any case, thank you for taking the time to send me feedback – really appreciate it!

    • Tomm says:

      CL 16 seem to be lowest RAM official support on Gemini Lake NUCs, 860 Pro for full power, but not really point. Evo is enough and noticeable cheaper.

      KVR ram is CL 17, so not at all difference with CL 16, but G.Skill DS RAM kit cheaper.

  3. Gcoates says:

    Does the Bios have an option to power on after a power outage? You note the bios is limited in options. That is a buy/no buy for me.

    • Olli says:

      Hi there! Luckily it’s not that limited. Under power options you’ll find the item “After Power Failure” and the options are “Stay Off”, “Last state” and “Power on”.

  4. I need ram for this that is confirmed to run in dual channel. Any ideas?

    I have tried 2 sets of RAM atm and both not working in dual channel.

  5. Vishwesh says:

    Deal Olli, can you please confirm if PJYH supports Dual Channel or not? I asked on Intel forum about particular ram kit, but Intel support hasn’t bothered replying even after 8 days. One of the seniors member there says it doesn’t support dual channel. So can you please clear the doubt?

  6. INTC says:

    CPU-Z 1.86 is out, does it still misreport J5005 as J4105?

  7. nucci says:

    cpu-z show me dual chanel: no, but hw monitor : yes. i use these memory:

    Crucial CT4G4SFS824A 4 Gt (DDR4, 2400 M / s, PC4-19200, Single Rank x8, SODIMM, 260-Pin)

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.