Saturday 15 October 2022

REVIEW: Beelink Mini S (Celeron N5095A 4C/4T CPU, 8GB DDR4 RAM, 256GB SSD) Mini-PC. (And E1DA Cosmos ADC follow-up: RIAA EQ & 768kHz.)

In the image above, we see the Beelink Mini S computer. I was sent this product from one of the distributors (thanks to MinixPC - see here for the current price) for an honest opinion in the context of the fact that over the last while, I had already explored the small MeLE Quieter2Q and Quieter3Q as well as the more powerful AMD-based Beelink SER4 Ryzen 7 machines. 

Now, unlike the MeLE Quieter machines, this one is not fanless, but it is indeed quiet and I suspect the fan noise will be no problem in the vast majority of settings. On balance however, what you get is a machine that is less expensive (you should be able to find one for around US$180 or better) and the active cooling will keep it cooler especially if you do end up placing it in more cramped environments. As usual, make sure there is still air flow of course!

Let's have a closer look at the item and performance...

Above we see the box contents, the computer unit itself is small, about 4.5" x 4" x 1.6" (11.5 x 10 x 4cm). It's mostly plastic so the weight is only around 270g. There's the 12V, 3A power supply, a couple of HDMI cables (short 25cm, and longer 100cm lengths), manual, and wall/monitor mount metal bracket with screws.

A closer look at the computer itself:

As you can see, we have a total of 4 USB3 ports (2 front, 2 rear), front headphone jack and power button, gigabit RJ45 ethernet port, 2 HDMI 2.0a (yes this can do HDR, but as we'll see later, limited), and the power jack. Oh yeah, there's also a front CMOS CLR recessed button if you mess up the BIOS and need to reset, and Kingston security slot.

Notice from the pictures that clearly the air flow comes in from the sides and is vented out the rear just above the ethernet port and HDMI. Needless to say, make sure airflow isn't obstructed to these key regions.

At this price point and this machine marketed as a "Mini Office Computer", it's not surprising that we don't see anything fancy. No USB-C, the built-in WiFi is 802.11b/g/n/ac, and Bluetooth 4.0 specs. The recent 10nm Gen11 Intel Celeron N5095A CPU is a 4-core/4-thread device running at base 2.0GHz, with turbo up to 2.9GHz. This is the same generation as the MeLE Quieter3Q's N5105 but the main difference is the graphics subsystem. The N5095 UHD Graphics iGPU runs at 450-750MHz stock and has 16 execution units compared to the N5105 with 24. While this is better than the MeLE Quieter2Q's J4125 processor with 12 execution units, and should be fine when it comes to office-type applications and general web streaming (like YouTube), be mindful of tasks that demand a bit more from the graphics system like 3D or higher resolution 4K work. We'll talk about this later.

Let's now take a quick peek inside the box. Note that because the base is flush with the sides, after taking off the 4 small Phillips-head screws holding the bottom down, I had to put a screw in one of the wall-plate mounts so I can pull the bottom off, it would have been nice if they had some kind of "lip" to allow the user to pull the base apart easier. In comparison, it was easier to open up the case of the Beelink SER4 4700U.

As for the guts of the machines, we can see the enclosure for the 2.5" HDD/SSD if you need extra storage attached to the machine's base. Inside the main box, there's the AZW-branded SODIMM which is a single 8GB DDR4-2666 stick. The M.2 drive is a SATA3 2280 256GB. While the SATA3 M.2 drive should be reasonably fast (maximum 6Gbps or 600MB/s), it would not be as quick as a NVMe M.2 SSD.

We can't see the heatsink, fan and CPU located on the other side of the mainboard.

The machine is pre-installed with Windows 11; just stick in your USB keyboard, an HDMI monitor, and it's basically plug-and-play from there.

Let's have a deeper look at the hardware data using HWiNFO64:

As you can see, the N5095A has a 15W TDP which is higher than the Quieter3Q's N5105 (10W) even though the CPU speed is about the same and the N5095A's iGPU performance is actually less powerful. We see some data for the DDR4-2666 RAM, and confirmation of the 16 EU iGPU. As with other Celeron processors, these lack some of the advanced instructions like AVX2.

The Jasper Lake USB controller is capable of USB3.1 (10Gbps) and as you can see above, I have confirmation that my portable SSD drive is connecting at "SuperSpeedPlus" plugged into one of the front USB ports. Note that the Bluetooth is based on Intel's Bluetooth 4.0 module.

WiFi is the Intel Dual Band AC 3165 (WiFi5), not their latest WiFi6 AX modules.

Other than some alphanumeric codes, I don't see a brand name or clear model number for the M.2 SATA3 SSD.

So how fast is this machine (stock speed)?

With knowledge of the above specs, we can surmise that this box would be faster than the eMMC drives in both the MeLE Quieter machines, will have GPU performance somewhere between the Quieter2Q and Quieter3Q, and raw CPU speed would be around the same as the Quieter3Q. Let's see how those predictions pan out with some benchmarks!

Here's the AIDA64 Extreme 6.7 suite of memory, integer and FPU benchmarks with comparisons to some of my other machines:

Memory Tests:           Read (MB/s)    Write (MB/s)   Copy (MB/s)    Latency (ns)
Beelink Mini S N5095     15910           17470          15860          60.8
MeLE Quieter3Q N5105     16959           22453          19042          66.9
MeLE Quieter2Q J4125     10409           10631          12860          93.5
Beelink Ryzen 7 4700U    36173           35510          29360          99.8
Intel NUC 6i5SYH         32161           45606          36301          66.1
Intel i7-7700K Server    31323           33913          31512          61.9
AMD Ryzen 9 3900X        49044           45780          46165          77.1

CPU Speed (Int): CPU Queen  PhotoWorxx(MPx/s)   ZLib(MB/s)  AES(MB/s)  SHA3(MB/s)
Beelink Mini S    22293       7383               188.0        19420        515
MeLE Q3Q N5105    20157       8640               144.5        15822        431
MeLE Q2Q J4125    19156       6771               120.4        9301         365
Beelink 4700U     49409       20336              422.8        48864        1534
NUC 6i5SYH        18092       15463              119.3        6101         553
i7-7700K Server   54240       20647              392.4        20509        1874
Ryzen 9 3900X     124348      29694              1191.3       107024       3705

CPU Speed (FP):   Julia    Mandel    SinJulia    Ray-Trace(kRay/s) FP64 Ray-Trace
Beelink Mini S    7462     3967      1003        1244                 677
MeLE Q3Q N5105    5604     2974      1003        1034                 533
MeLE Q2Q J4125    5835     3179      1112        933                  499
Beelink 4700U     50212    27457     8003        8750                 4870
NUC 6i5SYH        11358    6126      1603        2417                 1352
i7-7700K Server   38236    20577     5422        8160                 4505
Ryzen 9 3900X     113102   59539     20537       21413                11413

While it's clearly not a fair comparison to pit a machine like this to an i5, Ryzen 7 or Ryzen 9, among its Celeron peers, the Beelink Mini S performs quite well. In fact, other than PhotoWorxx (integer) where it lagged behind, and SinJulia (floating point) where it tied the MeLE Q3Q, the Mini S performed on the whole a bit faster than the fanless N5105 machine on these CPU-intensive tasks. It manages to edge out the MeLE Quieter3Q by 10-20% on a number of tests which is noteworthy.

With UserBenchmark, we can compare the performance to other machines out there with similar components:

Given the combination of parts in the machine, it's good that this is scoring in the upper half of expectations at 66%'ile. This is where we also see the relative weakness of the iGPU in the Celeron N5095 chip clocking in at 2.6% of all tested machines out there. Significantly lower than the N5105 of the MeLE Quieter3Q scoring at 4.2% (not that this is impressive).

Here's the M.2 SATA3 SSD drive speed using CrystalDiskMark for a better look:

Yes, definitely a much faster performing part compared to the eMMC of the MeLE Quieter mini PCs (by about 2-3x). This speed advantage is noticeable in usage as boot-up and application launching feels quite fast; certainly important if the machine is to be used in an office-type setting. In fact, that 550MB/s read speed is basically the limit of SATA3 (ideal conditions up to 600MB/s).

Examining the Graphics / Audio potential as HTPC...

Obviously, with the low integrated GPU speed noted on UserBenchmark, the Mini S is not going to be a gamer's machine! Having said that, I ran it downstairs in my sound room on the 4K TV to see the limitations if one were to try it as a home theater box.

Since I found issues with the MeLE Quieter3Q previously for 4K video and HDR, it stands to reason that I would generally not recommend this machine for HTPC applications if you want 4K/60fps/HDR. It does perform fairly well for 4K/60fps/SDR streaming off YouTube. For example, here's the 4K/60fps Costa Rica video with stats:

44 dropped frames out of 10,251 in 4K/60fps/SDR is not noticeable and looks great still.

I mentioned above that the HDMI output is capable of HDR, but from what I can tell, it's only limited to the 4K/30fps display setting.

And it will do a fair job (using K-Lite Codec Pack) with playing back common 4K/24fps/HDR10 movies encoded in HEVC; here's Edge of Tomorrow:

If you have a look at the stats bottom left, only 5 dropped frames in 5800. I'm also bitstreaming the Dolby TrueHD-Atmos audio through HDMI with no problem. The machine is certainly usable when you're in a pinch on movie nights on a 4K screen.

As with the other Mini PC computer boxes I've looked at through the year, we can tell Windows to use multichannel HDMI audio layout as below (I typically use 24/96 resolution):

Notice the options for DTS:X and Dolby Atmos which would bring you to the Microsoft store to download/purchase these as add-ons. Anyone out there using these? Any opinions on how useful?

Since this machine will not allow 4K/60fps/HDR to be turned on, obviously don't expect much from any of the few 60fps/HDR UHD BluRay rips you may have. I'd still recommend a purpose-built device like the AppleTV or nVidia Shield for 4K/HDR media room playback rather than a low-power PC at this point.

Going beyond the media room and basic office applications, even an inexpensive box like this will be fine with lighter duty media applications like photo and audio editing. As a "real life" trial, I ran Lightroom Classic 9 on this machine supplying it with some Sony 24MP raw files I took back in 2019 (at the final RMAF):

Pictures from the last Rocky Mountain Audio Fest. Images from the Tekton and Wilson rooms.

The Beelink Mini S was certainly fast enough with Lightroom, no issues with the usual routine like exposure, color balance, resize/rotation, lens distortion compensation, etc. Resizing, sharpening and exports to JPEG likewise did not take long. I found the machine just as fast as my usual i5 or i7 laptops when at the shows (like the recent PAF'22) as I edit pictures before posting them to the blog in the evenings.

I'm no gamer but clearly the iGPU is not suited for anything demanding. 2D and light 3D games in 1080P (like Minecraft?) should be alright, anything more likely will need you to push resolution lower for reasonable framerates (reviews have mentioned DotA 2 as an example of a game that one should drop to 720P).

The wireless connectivity works well even if it's not supporting the latest-and-greatest revisions. The Beelink's WiFi 5 (802.11ac) was able to communicate with my ASUS ROG GT-AX11000 in an adjacent room transferring files at around 20-25MB/s, quite reasonable compared to other machines I've had here at home. The Bluetooth 4.0 module doesn't have bells and whistles like the latest high speed, low-energy, or extended range BT 4.2 or 5.x versions. However, I connected some recently-acquired QCY T18 wireless AirPod-like earbuds to the machine for music playback. (At some point guys, I'll have to start discussing measurements of Bluetooth earphones since I think this has been neglected/dismissed by traditional audiophiles for way too long!)

As you can see, I'm playing Robbie Williams' latest XXV, connected to the QCY headphones over Bluetooth, and the Bluetooth Tweaker app is telling us that Windows 11 has elected to send the audio as AAC 48kHz, 285kbps which sounds very good. Range was more than 20' here at home line of sight, and would vary depending on walls and other obstructions.

By the way, if you look at the "CODEC#4 supported by the device" line, we see that the QCY T18 is capable of "aptX Adaptive" (Qualcomm QCC3050 chipset internally) which includes 16/44.1 aptX Lossless depending on your source device support. In recent days, I see that Qualcomm is using the term "aptX" less and just calling it "Snapdragon Sound" for consumer recognition on smartphones.


Well, in these uncertain, inflationary economic times where fiscal prudence is becoming more important, it's good that at a low price point of around US$180 for the 8GB RAM / 256GB SSD version, there are computers like the Beelink Mini S. It's a no-nonsense box that will get general light to moderate 2D computing tasks done (office, school work, surfing web...). At my workplace, this would be a fine machine in the office for a secretary or research assistant to compose mail, data gathering, basic data entry, moderate-sized database management, small-business spreadsheets, medical point-of-care access, etc. Maybe as a fileserver for a small group project. For these kinds of niches, the price point, no-frills feature set, and fast SSD speed are really all that are needed.

Relative speed as measured with benchmark suites, comparatively, it's great that the Beelink Mini S CPU performance was about 10-20% faster than the more expensive but fanless MeLE Quieter3Q although the Q3Q's N5105 has faster graphics though still not to the point of being a recommended 4K/60fps/HDR media player. In principle, the two CPUs should be about the same apart from the graphics so it's possible there are BIOS tweaks that could be done for the Q3Q, another possibility is there's thermal throttling happening in the fanless MeLE design.

For less than US$200, in 2022, it doesn't take a lot of money to get "core" computing tasks done! Compare this price to even a Raspberry Pi 4 8GB which is a way less capable machine (significantly less than 50% the computing power, doesn't include the 256GB SSD) at around the same money if not more and I think we can get a sense of the relative value per dollar.

[Thanks again to MinixPC for sending this box along for the review. You can use the code "ARCHMINIS" to get another US$10 off on their site.]


BIOS Options: Power Limiting & Fan noise...

I mentioned at the beginning that I think the fan noise would not be an issue - it's quiet. For the discerning sound room and home theater hobbyist, it can be audible however. Thankfully, the noise is primarily the low hum of air blowing out the back of the unit although pitch will vary with fan speed.

Up to this point, I have been talking about the stock machine, its capabilities, and benchmarks which I think would be what the vast majority of customers would be using this for. As you know, here at The Musings, I like to see if I can optimize the product to target a certain role that would be useful as an audio hobbyist; usually this involves a very quiet, ideally silent machine!

Let's look under the BIOS hood and see what I can do about that fan noise.

Yes, the BIOS does have the ability to underclock the machine and drop fan speed. We can do this by changing the machine's "power limit" parameters to restrict the amount of watts it's allowed. For example, let me show you my "almost silent" settings:

Looking in Advanced → CPU - Power Management Control → View/Configure Turbo Options, I see that the stock BIOS setting has "Power Limit 1" up at 15000 (15W) which is the CPU's TDP. Let's curtail PL1, and activate PL2 to reconfigure the total watts. In the screenshot above, PL1 is now at 8.5W which is the average power to aim for. PL2 at 10W means that in turbo mode, for a few seconds (usually around 10 seconds) the CPU can use up a bit more juice to get things started. Once the burst activity has settled, it has to at least go back down to PL1 for 20 seconds (PL1 Time Window) until an idle period, and then it's allowed to "turbo" again up to 10W briefly.

In this way we keep the total CPU power limited generally around 8.5W.

Further power savings can be had by turning down the speed of the iGPU (Advanced → GT - Power Management Control):

I dialed it down to 350MHz maximum (normally up to 750MHz). "RC6 (Render Standby)" reduces iGPU power use during sleep.

Given that the computer heat sink and fan were designed for 15W performance, this lower performance level at 8.5-10W means we can slow down the fan quite substantially. In the BIOS, we can turn on thermal functions and drop the active set-points for the fan (see Advanced → Thermal Configuration):

Then go to Platform Thermal Configuration to specifically adjust fan speed and make sure the Active Trip Points are enabled:

These settings dropped the maximum "Point 0" to 60% (at 71°C) and 30% (up to 55°C) fan speed (default was 100% and 50% I think). Now we reset the machine and it should be nice and quiet, basically inaudible unless I literally put my ear to the box when it's running heavy benchmarks.

In stock configuration, total machine power draw when idle is around 8W, maximum ~24W when pushed 100%. By dropping the power level and slowing the fan, the total power draw now goes down to 5-6W idle, maximum 17W under heavy load, and quiet. With these changes, using LinPack Xtreme, I'm seeing more than 23GFLOPS of processing power available. That's still a fair amount of potential.

So you might be wondering: "What's that Archimago guy doing with those underclocked/powered settings?"

Let's talk about that next time, friends! ;-)


To end, for those of you who have the E1DA Cosmos ADC, Ivan (aka IVX) has released the latest firmware which adds 768kHz capability. Just flash it with the ComTrue USB ISP firmware Tool:

Select the new "CosmosAdc_V13_RIAA_768.hex" firmware file and press "Auto" to get the flash done and verified.

After this, you can then use the ComTrue ASIO driver as input in REW, increase the samplerate to 768kHz, and get measurements like this:

As you can see in the graphs above, I'm measuring the same RME ADI-2 Pro FS playing a 0dBFS 1kHz (24/96)  signal at +13dBu (~3.5Vrms) into the E1DA Cosmos ADC. Notice the high frequency noise floor modulation >200kHz when running at 768kHz. There are also noise spikes up there which could be artifactual. Finally, between 20Hz to 20kHz, the noise floor is elevated when running at 768kHz resulting in higher N and THD+N results. Ivan noted that to improve the noise would likely require a hardware update of the ComTrue CT7601C to CT7601PR part; probably not worth the extra expense. Personally, I'm happy enough with 384kHz so 768kHz is "icing on the cake". I can imagine using the higher bandwidth just as a quick screen to make sure a device doesn't create too much noise above 200kHz beyond around -110dB with such a 0dBFS fundamental. One should not use the 768kHz sample mode for detailed resolution measurements.

This firmware also has a DSP curve setting included for those of you who like ripping vinyl directly into the Cosmos ADC acting as a phono preamp, applying very accurate RIAA equalization. Use the ADC volume setting in Windows or Cosmos Tweak to tell the ADC what mode you want:

It's generally more convenient to just use the Windows volume settings. Put the volume level between -45 to -55dB in Windows to turn on the RIAA EQ.

Here's what the RIAA EQ looks like in action when sent a flat sine sweep:

If it has been awhile since you've examined the RIAA EQ, there it is in all its "glory". About 40dB variation has been applied between 20Hz to 20kHz! Depending on the quality of a phono preamp, the accuracy of that compensation curve would be another source of potential error in vinyl playback (assuming that when the LP was created, the "recording" EQ was accurate to begin with).

Notice in particular the bass boost on playback (and concomitant bass reduction when creating the LP) which reduces the groove excursion needed when cutting the LP so that the length of each side can be a reasonable duration, typically 20-25 minutes per side. A potential side effect with this kind of system is that low-frequency distortions like turntable rumble also gets boosted during playback, a reason why a turntable with good vibration isolation is important beyond time precision (ie. exactness of 33 1/3 or 45 rpm).

Both the 768kHz sample rate and RIAA DSP are cool features and could come in handy at some point. Great to see ongoing developments and optimizations.

Okay everyone, hope you're all doing well and having a great time enjoying the music!


  1. B-link makes excellent products. I have a B-link GT-King Pro
    It's a mediaplayer based on android with 4 GB ram and 64 GB storage
    It has not one but two ESS Hi-Fi ES9018 audio decoding chips (DAC) and two RiCore RT6862 Op-Amp chips. Impressive right?

    Even though I bought an Cosmos ADC recently I was too lazy to measure it. I have no regret buying the Cosmos.ADC not even for a second.

    Turns out that a large proportion of my expensive audiophile stuff wasn't that good objectively speaking. Looks can be deceiving...

    Bring forth the Guillotine.

    But in reality most content (streaming) watch it with are encoded wiih AC3 5 640 kbps.(yikes). Personally i prefer DTS because it sounds a little better. But that's subjective. Not really evidence based. If i had to undergo a double blind test i would probably fail. miserably. I'm not good with tests haha

    If i find a TV Show or Films exceptionally i honor the creators by buying the
    blu-ray. Yes i know you probably thinking 4K HDR is much better. (visually)

    But if something is that good it transcends he medium.
    If i watched the show on worn out VHS tapes i probably would still be rating it as one of the best shows in history. (subjective)

    Both of those codecs dont do justice

    1. Hi FastFreddy! It's insightful to learn about Beelink. Could you please confirm—does GT-King Pro still run Android Pie, or were there any updates, and whether it's actually rooted as these folks say:

      If it's true I would be really concerned about connecting this box to any network. Usually hackers go a long way in order to root a device, while this device comes pre-rooted :)

    2. Thanks for the note Freddy,
      I had no idea that they made a box with the higher end DACs like the Sabre ES9018, much less two of them in there!

      Yes, the truth is out as you say:
      "Turns out that a large proportion of my expensive audiophile stuff wasn't that good objectively speaking. Looks can be deceiving..."

      Over time, as consumers recognize this more, I think it will upend the "High End" world and their claim to "high fidelity". ;-) Have fun with the Cosmos ADC as that will shine the light into the performance of many of our devices!

  2. i was speaking of Breaking Bad. I'm grateful that blogspot does not have an edit button. Amen. (I'm an agnostic. aka the easy way out). Science will win in the end.

    1. Agree Freddy,
      Ultimately "truth" I think will win out! Science being just the mechanism we have to empirically test to reach for truth...

  3. Aloha Arch!

    Nice review! Agree with FastFreddy! Beelink makes excellent products. As you know, based on your posts here, I picked up the SER4 4800U with 32 gigs of RAM. I use it as a Visual Studio C++ development computer for my DSP audio projects. It's fast and quiet. Even though I try and sneak it on my Hisense 75 U9DG 4K TV from time to time :-)

    I remember being over at your place listening to tunes. You do have a quiet room and I imagine at night, dead quiet. You have been to my place before, I have this setup in the living room and I can't hear the fan at all, even under heavy compilation. I have to put my ear right up to the rear of the SER4 to hear the fan, even though it is like two feet away from me.

    Man, it must be quiet there! Do you have an SLM? Or you could even use REW to measure the ambient noise in your sound room with the computer off and then on measured 2 feet away, like if it were on your desk. I wonder if there is a measurable difference? :-) I know you are hypersensitive to fan noise, but my place must be too noisy as I can't hear that or the Apple mini M1 fan either... Or maybe I am going deaf!

    I use Dolby Atmos for headphones and surprised how well it works! I don't know how it is possible for an Atmos system using Windows audio...?

    Hope you are enjoying the great weather man! Wow, Lesley and I were mountain biking today in T-shirt/shorts and sweating! Unheard for the middle of October here, can only be climate change I am afraid, so short lived.

    Keep up the great writings Arch! Cheers!


    1. Hey there Mitch,
      Lovely having you out here a few years back in the midst of the Pandemic. ;-)

      Yeah, I've tested using a sound level meter (as well as calibrated REW) the ambient noise level a few years ago:

      On a quiet evening, the room is just below 30dB(A) SPL. It's the quietest room in the house away from stuff like fridges and other things that can turn on intermittently. In the basement so I don't hear the family upstairs, and isolated enough so kids can still sleep at night even if watching a movie with the sub going. The man-cave "oasis" to de-stress after a long day. ;-)

      Alas, I do get quite sensitive to fan noise so going fanless (and making sure no audible coil whine and other sources of noise!) is important to me. A small background "hush" I suppose isn't too bad but I hate that ramping up in frequency when fans go faster during heavier loads.

      I think for Windows audio and Atmos, what they do is encode as E-AC3 stream rather than discreet lossless HDMI, lossy like Apple Spatial Audio. I suspect this is what the XBOX game machines also do when it's not passing through TrueHD-Atmos. It'll certainly keep the bitrate much lower. Someone correct me if I'm wrong on this!

      Indeed, enjoy the weather Mitch. I heard it's drought conditions up at the Sunshine Coast though, so we do need some rain up... Just not too much and let's see if we can keep some sun a bit longer here on the West Coast. ;-)


    2. Aloha Arch! We finally got some rain, I hope it continues, but like you say, not too much :-)

      Nice ambient noise level! The quietest room in my house is below the range of 30dB(A) and (C) on my calibrated GalaxyAudio CM-140 SLM. But I live out in the sticks. Unfortunately, the sound system isn't in that room...

      I measure 33dB(A) and 40 dB(C) in my listening room, which has the kitchen off to the side as you have seen. Fridge noise mostly.

      If I place the mic 6 inches away from the fans on my M1 and the Beelink, I measure 35dB(A) and 42 dB(C). An increase of 2 dB SPL which is barely noticeable. If I place my ear where the mic is, I can barely hear the fan noise. If I am 2 ft away, I read 33dB(A) and 40 dB(C), which is the same as ambient noise of the room. That's why I can't here them 2ft away. Maybe what I do simply does not push the fan up, so it is tolerable for me as I can't hear the fan on either the Beelink or the M1 at 2ft. Maybe you can recommend a test I can run on either to ramp those fans up...

      I have the Dolby Atmos for Headphones and watching movies on Netflix with it engaged is a real treat for the ears compared without it. If I click on Dolby Atmos, it asks me to turn on my Dolby device and connect it via HDMI to my PC.

      Hope you are having a great start to the weekend. Arch!

      Kind regards,

    3. Hey Mitch,
      Nice tests of ambient noise and the computer levels! I think we don't talk enough about ambient noise level... I've often for example wondered about the sound rooms that audio reviewers use and how soft or loud the ambient noise level is. Especially for the guys who review stuff in the city! My feeling is that this is a pre-requisite for all audio reviewers to measure along with a discussion of the room they're using and their own hearing ability before talking about how they can hear the effect of cables, ethernet switches, and other Golden Ear claims. ;-)

      Try running LinPack Xtreme:

      For say 10 minutes in stress test mode. That should ramp the fan up to the max pretty quick and let you know if you hear it. I use that these days quite a bit as it will bomb out with flaky memory or overclocked CPUs. I know my Beelink SER4 is also quite quiet when I use it for measurement acquisition running realtime FFTs and the like and works really well!

      Speaking of "hearing ability" as an audiophile, I was out at BC Place last night for Elton John's "farewell" concert. Great to see the 75 year old play likely one last time live here in Vancouver. No breaks for 2.5 hours of playing and singing. Anyhow, I think I'll need a couple of days for my hearing to return to normal this weekend. ;-)

      I know you go to concerts not infrequently these days, but man, I'm glad I don't go to too many rock/pop (mostly classical) for the sake of maintaining one's ears hopefully well into retirement...

    4. Hey Arch,

      Cool, thanks for the LinPack Xtreme. Wow, it tool much longer than I thought to wind the fan up to it's rated 46 dB(A) max at 6 inches. I can hear that. But under normal operations, I can't hear it or maybe I tune it out as I sleep using a noise generator.

      I visit a buddy of mine in downtown Toronto on Young and Bloor and aside from the sirens blaring every 10 minutes, the apartments are "pressurised" with quit a bit of air flow. One can hear it as the door closes. No low level detailed listening there.

      Speaking of quiet, one studio I worked at was pretty well damped, back in the day where dead was in. Had to sleep overnight, and did not go well. It is so quiet one can not only hear the heart beat (loudly!) but also the blood pumping though the veins. Gah! Every gurgle, creak, muscle movement sounded like it was amplified to max volume. Slept in the control room with music playing.

      Very cool you got to see Elton! 2.5 hrs, total pro. Good times and good memories.

      It used to be club/concert level was around ~95 dB(C) at the mix position, which is quite loud, twice as loud as our "audiophile" 83 dB SPL (C) weighting we use for critical listening and mixing/mastering music . At the Commodore Club a few years back, I measured one of the bands there averaging 107 dB to peak of 110 dB SPL (C). So loud that your entire body was part of the sound. We left. I wear earplugs anytime the level goes over 90 dB SPL, so that means every concert :-)

      Cool again that you got to see a pro like Elton for the last time.

  4. On the hand the Celeron N5095A wipes its ass with the Amlogic S922X in the GT-King Pro (hexacore) in terms of performance. The Celeron is 125% faster In single core, 128% in multi-core. The price is a slightly higher power consumption.
    the Mini-S idles at 7 or 8 watts and when completely stressed the system will draw 17 watts. The Amlogic CPU idles at 4 watts and with normal use about 6 watts. The maximum power draw is about 10 watts when fully stressed.
    The GT-King Pro is fanless.

    Both processors can of course decode H264, H265 (aka HEVC 8/10 bit), AVC, VP8, VP9 and VC1 in hardware. Unfortunately AV1 is not supported in hardware so the CPU has to do a lot of work. you can still play software-decoded AV1 but because the CPU cores are relatively slow frame drops will start to occur at AV1 1080p / 30 fps. (1)

    AV1 is a codec currently used by Netflix and other streaming services. The reason is simple: It saves bandwidth (money). Tests from University of Waterloo in 2020 found that when using a mean opinion score (MOS) for 2160p (4K) video AV1 had bitrate saving of 9.5% compared to HEVC and 16.4% compared to VP9.
    The encoding process takes about twice as long time to encode H265 but this is a small price to pay in comparison with the bandwidth savings.


    Note: The U59 being tested is almost identical to the mini-S and not much more expensive. The U59 has two dimm slots which means the memory operates in ""dual channel"" memory which in some cases can make a tiny difference.

    I bought my GT-King Pro for € 121,97 (but this was pre-covid/inflation)
    You can buy the Mini S for $150. The U59 will set you back $212.40.
    but for a few dollars more you have U59 pro with Celeron N5105. It's a fraction faster but more importantly It's being more energy efficient (10 W TDP vs N5095's 15 W) and features a faster iGPU model (24 EU UHD Graphics versus 16 EU UHD Graphics, with a higher clock rate to boot. The 16GB RAM 500GB SSD version is listed on Aliexpress for $245.

    In theory the GT-King-Pro is the audiophile choice due two the dual ESS9018 but I didn't measured it yet because I'm wasting my time reading blogs on the internet.

    1. LOL Freddy,
      Stop wasting time reading blogs on the Internet. ;-)

      I think when it comes to low-power media machines, there are a lot of options out there and those TV boxes like the Beelink GT King Pro with its Amlogic S922X SoC can do really well as a decoder! As a general machine, something like this Mini S or the faster and more expensive U59 (Pro) models have their roles as inexpensive work, application, and online "workhorses".

      The GT King Pro can take on nVidia's Shield and Apple TV... Would never expect the Mini S to compete in that space however!

  5. I'm curious what the common use cases are for a device like this. I appreciate it as a GPU but wondering about entertainment purposes. Is this going to perform better or easier than the various video streaming devices out there like a Roku ro Fire TV?

    1. Hi Doug,
      Personally, for video streaming, no, I see this more as a work/general processing machine. I'll happily stick with my Apple TV 4K for video playback, HDR/Dolby Vision streaming.

      I would have no problem replacing a small tower computer at work with one of these for the secretary doing Word, E-mails, medical billings, etc.

      Of course, this weekend you'll see what I would use the power-limited, quiet settings for! ;-)