Intel NUC11TNBi5 and Akasa Newton TN Fanless Case Review: Silencing the Tiger

Our review sample of the Intel NUC11TNKi5 came with all necessary components pre-installed, and we carried them over to the Newton TN build – we only had to load up the OS to start our evaluation process. Prior to that, we took some time to look into the BIOS interface. The video below presents the entire gamut of available options for the Tiger Canyon NUC.

Your browser does not support HTML video.

Intel’s mini-PCs have the most comprehensive and configurable UEFI / BIOS among all the ones that we have evaluated over the last several years. The Tiger Canyon NUC is no exception. The home screen shows the BIOS version, processor details, and installed memory capacity and speed right away – this is of great help in checking whether the installed SODIMMs are operating at the desired speeds. Under the ‘Advanced’ section, we have options to enable various SATA and M.2 NVMe ports, and disable the disk activity LED if needed. Different USB ports and headers can be selectively enabled or disabled. iSCSI volumes can be mounted in the boot stage, and the network settings can be configured for network boot when needed. The HDMI / DisplayPort audio can be disabled, if necessary. Thunderbolt support is also controllable via the BIOS, as are a bunch of other features.

One of the most useful features in the lineup of Intel NUCs targeting the professional / business use-cases is the display emulation option under ‘Advanced > Video’. This is useful for scenarios involving headless operation and digital signage installations, as it allows set up of a virtual display when no display is connected (virtual display emulation), or makes the OS think that a previously connected display is still connected to the display output (persistent display emulation).

The ‘Cooling’ section provides an overview of current temperatures and fan speeds, and allows fine grained control of the fan operation – including the sensors to which it responds, and the temperatures at which different actions need to trigger. In the ‘Performance’ section, users can control hyperthreading support, selectively enable or disable cores, and control various performance options like Turbo Boos and SpeedStep. The memory timings can also be finely controlled. Various password options including TCG Opal configuration can be activated from the ‘Security’ section. Virtualization support and VT-d are also enabled or disabled from this section.

The ‘Power’ section allows users to optimize the system for balanced performance, or low power operation, or for maximum performance. In the last case, Intel’s Dynamic Power Technology can also be configured for energy efficiency. Further customization is possible by allowing for different sustained and burst mode package power limits (PL1 and PL2), as well as the package power time window. Secondary power settings allow for configuration of LED behavior in different sleep states, as well as allowing support for waking up via different peripheral interfaces. PCIe ASPM support is also controllable from this section. The ‘Boot’ section allows configuration of secure boot, boot priorities, POST screen configuration, etc. A boot override option here is probably the only update we would like in the BIOS – a shortcut to rebooting and pressing F10 repeatedly in order to choose a temporary boot device.

Intel’s documentation provides a block diagram for the Tiger Canyon NUC board. It gives insights into the distribution of high-speed I/O lanes.

The PCIe 4.0 x4 lanes servicing the M.2 NVMe SSD and the two Thunderbolt 4 ports are directly off the processor die. A single PCIe 3.0 x1 lane from the Tiger Point-LP PCH is used to connect the Intel I225-LM controller. The AX201 WLAN controller hooks up via the CNVi interface in the PCH. The front USB 3.2 Gen 2 ports are from the processor, while the USB 2.0 and USB 3.2 Gen 2 ports in the rear are from the PCH.

In addition to the comparison between the actively-cooled and passively-cooled versions of the NUC11TNBi5, we also consider some of the other passively cooled PCs reviewed earlier, as well as a few other recent UCFF systems. In the table below, we have an overview of the various systems that we are comparing.

The NUC11TNKi5 and the NUC11TNBi5 (Akasa Newton TN) share the same configuration except for the pricing. The ASUS PN51-E1 slots in as a comparison point for what a mid-range Renoir can deliver (since we are comparing against a mid-range TGL-U part). It is a relevant comparison point for the NUC11TNKi5 from a price viewpoint, given they are both actively cooled. The ASUS PN50 and ASRock Industrial 4X4 BOX-4800U make up the high-end actively-cooled Renoir systems.

We also have results from a recent re-benchmarking session of the ASRock Industriall NUC BOX-1165G7. It uses the latest BIOS (P1.40), but modifies the default options to better reflect usage in a home / business scenario (ASRock Industrial stated that they had optimized the BIOS defaults for industrial deployment). ‘Advanced > CPU Configuration > CPU Operation Mode’ was changed from default ‘Normal Mode’ (PL1 = 28W / PL2 = 28W) to ‘Performance Mode’ (PL1 = 38W / PL2 = 55W), ‘Advanced > CPU Configuration > CPU C-States Support’ was changed from default ‘Disabled’ to ‘Enabled’, ‘CPU C States Support > Enhanced Halt State (C1E)’ set to ‘Enabled’, ‘CPU C States Support > Package C State Support’ set to ‘Auto’, and ‘CPU C States Support > CFG Lock’ left at ‘Disabled’. 

The OnLogic HX500 is a representation for Intel’s previous generation in a fanless avatar. Even though it utilizes a desktop CPU with a 35W TDP, it makes sense to include it in the comparisons as Tiger Lake-U includes fine-grained power management with boosts up to 64W to make it competitive against real desktop CPUs.

The Akasa Turing / Bean Canyon configuration is also included in the benchmarks to identify improvements that consumers can see for similarly priced systems when upgrading after a couple of generations. Results from a Core i5 Comet Lake-U system (the Minisforum UM850) are also included. Finally, we have Supermicro’s SM-E100-12T-H – another TGL-U representation, albeit at a much lower package power limit.

Our 2022 test suite for Windows 11-based systems carries over some of the standard benchmarks we have been using over the last several years, including UL’s PCMark and BAPCo’s SYSmark. Starting this year, we are also including BAPCo’s CrossMark multi-platform benchmarking tool.

PCMark 10

UL’s PCMark 10 evaluates computing systems for various usage scenarios (generic / essential tasks such as web browsing and starting up applications, productivity tasks such as editing spreadsheets and documents, gaming, and digital content creation). We benchmarked select PCs with the PCMark 10 Extended profile and recorded the scores for various scenarios. These scores are heavily influenced by the CPU and GPU in the system, though the RAM and storage device also play a part. The power plan was set to Balanced for all the PCs while processing the PCMark 10 benchmark.

The ‘Essentials’ workload score is dictated by single-threaded burst performance, and the Tiger Lake-U systems take up the top three spots. However, the ‘Productivity’ and ‘Digital Content Creation’ workloads seem to appreciate larger number of cores. The octa-core Renoir systems come out on top, and even the hexa-core equipped ASUS PN51-E1 leapfrogs the Tiger Lake-U systems in the ‘Productivity’ workload. For the ‘Gaming’ workload, the TGL-U systems with the new Iris Xe iGPU is able to surpass Renoir. The iGPU in the TGL-U Core i7 in the ASRock Industrial NUC BOX-1165G7 enjoys a healthy lead over the competition. Overall, though, the octa-core Renoir-equipped UCFF PCs get a slight edge over the mid-range TGL-U configuration in the NUC11TNBi5. Another key takeaway here is that the scores for the actively-cooled and passively-cooled versions are pretty much similar, lending credence to the possibility of Akasa having pulled off another excellent offering in the Newton TN.

BAPCo SYSmark 25

BAPCo’s SYSmark 25 is an application-based benchmark that uses real-world applications to replay usage patterns of business users in the areas of productivity, creativity, and responsiveness. The ‘Productivity Scenario’ covers office-centric activities including word processing, spreadsheet usage, financial analysis, software development, application installation, file compression, and e-mail management. The ‘Creativity Scenario’ represents media-centric activities such as digital photo processing, AI and ML for face recognition in photos and videos for the purpose of content creation, etc. The ‘Responsiveness Scenario’ evaluates the ability of the system to react in a quick manner to user inputs in areas such as application and file launches, web browsing, and multi-tasking.

Scores are meant to be compared against a reference desktop (the SYSmark 25 calibration system, a Lenovo Thinkcenter M720q with a Core i5-8500T and 8GB of DDR4 memory to go with a 256GB M.2 NVMe SSD). The calibration system scores 1000 in each of the scenarios. A score of, say, 2000, would imply that the system under test is twice as fast as the reference system.

SYSmark 25 also adds energy measurement to the mix. A high score in the SYSmark benchmarks might be nice to have, but potential customers also need to determine the balance between power consumption and the performance of the system. For example, in the average office scenario, it might not be worth purchasing a noisy and power-hungry PC just because it ends up with a 2000 score in the SYSmark 25 benchmarks. In order to provide a balanced perspective, SYSmark 25 also allows vendors and decision makers to track the energy consumption during each workload. In the graphs below, we find the total energy consumed by the PC under test for a single iteration of each SYSmark 25 workload. For reference, the calibration system consumes 8.88 Wh for productivity, 10.81 Wh for creativity, and 19.69 Wh overall.

The scores for the NUC11TNKi5 and the Akasa Newton TN build are approximately the same (within the realm of run to run variations), and there is a sizable different in the energy consumption in favor of the passive build (probably due to the absence of the fan’s power consumption). On the raw scores side, the mid-range TGL-U systems make it to the top half dominated by Core i7 offerings with higher package power budgets.

BAPCo CrossMark 1.0.1.86

BAPCo’s CrossMark aims to simplify benchmark processing while still delivering scores that roughly tally with SYSmark. The main advantage is the cross-platform nature of the tool – allowing it to be run on smartphones and tablets as well.

The relative performance seen in SYSmark 25 translate to CrossMark also, as expected. The responsiveness ratings point to the first sign of hiccups for the Akasa Newton TN build. While the SSD has a significant impact on this sub-component score, our monitoring software didn’t report any dangerous temperatures for the SSD. As such, we trust SYSmark 25 more than CrossMark because it operates with real workloads and without idle time compression, rather than a simulation of the activity.

Standardized benchmarks such as UL’s PCMark 10 and BAPCo’s SYSmark take a holistic view of the system and process a wide range of workloads to arrive at a single score. Some systems are required to excel at specific tasks – so it is often helpful to see how a computer performs in specific scenarios such as rendering, transcoding, JavaScript execution (web browsing), etc. This section presents focused benchmark numbers for specific application scenarios.

3D Rendering – CINEBENCH R23

We use CINEBENCH R23 for 3D rendering evaluation. R23 provides two benchmark modes – single threaded and multi-threaded. Evaluation of different PC configurations in both supported modes provided us the following results.

In the single-threaded case, Tiger Lake’s Core i7 with a 38W PL1 takes the lead, as expected. The Core i5 NUC configurations come right behind, hobbled a bit by their 28W PL1 setting. In the multi-threaded case, the sheer number of cores help the octa-core Renoir mini-PCs emerge on top, followed by the 35W Comet Lake-equipped OnLogic HX500 well behind. Again, the important takeaway here is that the relative difference between the active and passively-cooled builds is minuscule.

Transcoding: Handbrake 1.5.1

Handbrake is one of the most user-friendly open source transcoding front-ends in the market. It allows users to opt for either software-based higher quality processing or hardware-based fast processing in their transcoding jobs. Our new test suite uses the ‘Tears of Steel’ 4K AVC video as input and transcodes it with a quality setting of 19 to create a 720p AVC stream and a 1080p HEVC stream.

Software transcoding rates heavily depend on the number of cores, followed by the performance potential of each core. The high-end Renoir-based systems with eight cores have a significant lead that the quad-core TGL-U systems are unable to match up to.

While the Renoir systems support VCE, we are not including results from those transcodes because the end results are not the same – and it wouldn’t be an apples-to-apples comparison. The rate here is a function of the speed of the GPU clock feeding into the QuickSync block. All TGL-U systems end up with approximately the same numbers.

Archiving: 7-Zip 21.7

The 7-Zip benchmark is carried over from our previous test suite with an update to the latest version of the open source compression / decompression software.

7-Zip is again a multi-threaded workload where Renoir’s octa-core SKUs grab the top spots, and Tiger Lake’s single-threaded performance advantage is not quite relevant.

Web Browsing: JetStream, Speedometer, and Principled Technologies WebXPRT4

Web browser-based workloads have emerged as a major component of the typical home and business PC usage scenarios. For headless systems, many applications based on JavaScript are becoming relevant too. In order to evaluate systems for their JavaScript execution efficiency, we are carrying over the browser-focused benchmarks from the WebKit developers used in our notebook reviews. Hosted at BrowserBench, JetStream 2.0 benchmarks JavaScript and WebAssembly performance, while Speedometer measures web application responsiveness.

From a real-life workload perspective, we also process WebXPRT4 from Principled Technologies. WebXPRT4 benchmarks the performance of some popular JavaScript libraries that are widely used in websites.

Browsers enjoy multiple cores, but it is single-threaded performance that ends up as a performance limiter. Here, Tiger Lake shines across the board – particularly in the real-world Principled Technologies WebXPRT4 workload.

Application Startup: GIMP 2.10.30

A new addition to our systems test suite is AppTimer – a benchmark that loads up a program and determines how long it takes for it to accept user inputs. We use GIMP 2.10.30 with a 50MB multi-layered xcf file as input. What we test here is the first run as well as the cached run – normally on the first time a user loads the GIMP package from a fresh install, the system has to configure a few dozen files that remain optimized on subsequent opening. For our test we delete those configured optimized files in order to force a fresh load every second time the software is run.

As it turns out, GIMP does optimizations for every CPU thread in the system, which requires that higher thread-count processors take a lot longer to run. So the test runs quick on systems with fewer threads, however fast cores are also needed. This combination works out well for Tiger Lake, and we see the four TGL-U systems take up the top spots.

Tiger Lake’s integrated GPU based on the reworked scalable Xe microarchitecture provides a significant increase in performance over the previous Intel iGPUs. In mini-PCs, the GPUs are generally used for casual gaming and multimedia workloads (including shader-enabled video postprocessing). GPU performance evaluation typically involves gaming workloads, and for select PCs, GPU compute.

The Intel Iris Xe Graphics in the NUC11TNBi5’s Core i5-1135G7 has 80 EUs clocked between 0.4 GHz and 1.3 GHz (not 96, as shown in the screenshot above by GPU-Z). The performance of this iGPU is miles ahead of previous iGPUs from both Intel and AMD, as the benchmarks below show.

GFXBench

The DirectX 12-based GFXBench tests from Kishonti are cross-platform, and available all the way down to smartphones. As such, they are not very taxing for discrete GPUs and modern integrated GPUs. We processed the offscreen versions of the ‘Aztec Ruins’ benchmark.

The extra EUs in the Core i7-1165G7, coupled with the higher PL1 limits enable the ASRock Industrial NUC BOX-1165G7 to grab the top spot here. However, the NUC11TNBi5 configurations take up the next two spots. The high-end Renoir SKUs reach around 80% of the performance of the NUC11TNBi5’s iGPU.

UL 3DMark

Four different workload sets were processed in 3DMark – Fire Strike, Time Spy, Night Raid, and Wild Life.

3DMark Fire Strike

The Fire Strike benchmark has three workloads. The base version is meant for high-performance gaming PCs. It uses DirectX 11 (feature level 11) to render frames at 1920 x 1080. The Extreme version targets 1440p gaming requirements, while the Ultra version targets 4K gaming system, and renders at 3840 x 2160. The graph below presents the overall score for the Fire Strike Extreme and Fire Strike Ultra benchmark across all the systems that are being compared.

The relative difference observed in GFXBench translate to both Fire Strike workloads, with the NUC11 configurations taking up the second and third spots behind the NUC BOX-1165G7.

3DMark Time Spy

The Time Spy workload has two levels with different complexities. Both use DirectX 12 (feature level 11). However, the plain version targets high-performance gaming PCs with a 2560 x 1440 render resolution, while the Extreme version renders at 3840 x 2160 resolution. The graphs below present both numbers for all the systems that are being compared in this review.

The advantage for the TGL-U’s iGPU runs thin in the Time Spy workloads, with the high-end Renoir systems catching up to the NUC11’s Core i5-1135G7 at 4K resolution, and only slightly behind at 1080p.

3DMark Wild Life

The Wild Life workload was initially introduced as a cross-platform GPU benchmark in 2020. It renders at a 2560 x 1440 resolution using Vulkan 1.1 APIs on Windows. It is a relatively short-running test, reflective of mobile GPU usage. In mid-2021, UL released the Wild Life Extreme workload that was a more demanding version that renders at 3840 x 2160 and runs for a much longer duration reflective of typical desktop gaming usage.

The GFXBench story and analysis repeats for the Wild Life workload also.

3DMark Night Raid

The Night Raid workload is a DirectX 12 benchmark test. It is less demanding than Time Spy, and is optimized for integrated graphics. The graph below presents the overall score in this workload for different system configurations.

The ‘Night Raid’ workload sees the gap between the high-end Renoir and mid-range TGL-U iGPU narrow down, but the advantage still rests with TGL-U.

One of the key drivers of advancements in computing systems is multi-tasking. On mobile devices, this is quite lightweight – cases such as background email checks while the user is playing a mobile game are quite common. Towards optimizing user experience in those types of scenarios, mobile SoC manufacturers started integrating heterogeneous CPU cores – some with high performance for demanding workloads, while others were frugal in terms of both power consumption / die area and performance. This trend is now slowly making its way into the desktop PC space.

Multi-tasking in typical PC usage is much more demanding compared to phones and tablets. Desktop OSes allow users to launch and utilize a large number of demanding programs simultaneously. Responsiveness is dictated largely by the OS scheduler allowing different tasks to move to the background. Intel’s Alder Lake processors work closely with the Windows 11 thread scheduler to optimize performance in these cases. Keeping these aspects in mind, the evaluation of multi-tasking performance is an interesting subject to tackle.

We have augmented our systems benchmarking suite to quantitatively analyze the multi-tasking performance of various platforms. The evaluation involves triggering a VLC transcoding task to transform 1716 3840×1714 frames encoded as a 24fps AVC video (Blender Project’s ‘Tears of Steel’ 4K version) into a 1080p HEVC version in a loop. VLC internally uses the x265 encoder, and the settings are configured to allow the CPU usage to be saturated across all cores. The transcoding rate is monitored continuously. One complete transcoding pass is allowed to complete before starting the first multi-tasking workload – the PCMark 10 Extended bench suite. A comparative view of the PCMark 10 scores for various scenarios is presented in the graphs below. Also available for concurrent viewing are scores in the normal case where the benchmark was processed without any concurrent load, and a graph presenting the loss in performance.

Except for the gaming workload which is pretty much unaffected by the CPU loading, the sheer number of cores in the Renoir-based systems help it in salvaging good scores in the presence of concurrent loading.

Following the completion of the PCMark 10 benchmark, a short delay is introduced prior to the processing of Principled Technologies WebXPRT4 on MS Edge. Similar to the PCMark 10 results presentation, the graph below show the scores recorded with the transcoding load active. Available for comparison are the dedicated CPU power scores and a measure of the performance loss.

The Tiger Lake-U systems are the top performers here, both in terms of raw scores and minizing performance loss.

The final workload tested as part of the multitasking evaluation routine is CINEBENCH R23.

The large number of cores in the Renoir systems allow it to shuffle around the tasks in such a way that the performance loss for rendering workloads is minimized.

After the completion of all the workloads, we let the transcoding routine run to completion. The monitored transcoding rate throughout the above evaluation routine (in terms of frames per second) for select systems are tabulated below.

The transcoding rates in different systems drop down with simultaneous loading, as expected. The key numbers to note in the above table are the first and second encoding passes for the Newton build. With the numbers actually showing a slight advantage for the second pass, it is clear that there is no throttling at play for this particular workload and duration.

The HTPC-related sections in previous SFF PC reviews covered a range of aspects. Display refresh rate stability (particularly, the ability to drive 23.976 Hz for stutter-free playback of cinema content), OTT streaming efficiency (YouTube and Netflix), and local media playback performance and efficiency evaluation were some of them. While such a detailed study may still make sense for dedicated HTPC reviews, we have decided to pare down the evaluated aspects for system reviews. Workloads were processed on the Intel NUC11TNBi5 in Akasa’s Newton TN fanless case for the results in this section.

YouTube Streaming Efficiency

4K video streaming has become ubiquitous enough for its support to be a necessity even for secondary HTPCs. HDR has also become affordable – in fact, one of the key changes for HTPCs in Jasper Lake is the enabling of HDR over HDMI, a feature not available in Gemini Lake-based systems. Keeping these aspects in mind, we have chosen Mystery Box’s Peru 8K HDR 60FPS video as our test sample moving forward. On PCs running Windows, it is recommended that HDR streaming videos be viewed using the Microsoft Edge browser after putting the desktop in HDR mode.

Intel’s Xe Graphics in TGL-U supports hardware decoding for VP9 Profile 2. Taking advantage of this feature, MS Edge automatically fetches the 4Kp60 VP9 Profile 2 encode from the YouTube servers. The playback was flawless in both builds, with the few dropped frames happening in the beginning during playback resolution / fullscreen switching.

D3D usage makes its appearance for upscaling purposes (when the streamed / decoded video doesn’t match the native display resolution), but soon disappears after the 4Kp60 stream starts coming in. The at-wall power consumption is around 20W on an average and the decoder loading is south of 40%, with the video processor load coming in around 55%. With smooth playback and no dropped frames worth analyzing, the graph analysis turns out to be pretty uneventful.

Hardware-Accelerated Encoding and Decoding

The transcoding benchmarks in the systems performance section presented results from evaluating the QuickSync encoder within Handbrake’s framework. The iGPU in the systems support hardware encode for AVC, JPEG, HEVC (8b and 10b, 4:2:0 and 4:4:4), and VP9 (8b and 10b, 4:2:0 and 4:4:4). The capabilities of the decoder engine are brought out by DXVAChecker.

The decoder engine in Tiger Lake-U supports the latest and greated that Intel has to offer (except for AV1 encode, which is present only in the Arc dGPUs for now).

Local Media Playback

Evaluation of local media playback and video processing is done by playing back files encompassing a range of relevant codecs, containers, resolutions, and frame rates. A note of the efficiency is also made by tracking GPU usage and power consumption of the system at the wall. Users have their own preference for the playback software / decoder / renderer, and our aim is to have numbers representative of commonly encountered scenarios. Considering the target market for Tiger Lake-U systems, we played back the test streams using the following install-and-forget combinations:

• VLC 3.0.17.4
• Kodi 19.4

The fourteen test streams (each of 90s duration) were played back from the local disk with an interval of 30 seconds in-between. Various metrics including GPU usage and at-wall power consumption were recorded during the course of this playback. Based on the DXVAChecker report presented previously, the GPU should be able to play back all codecs with hardware acceleration.

All our playback tests were done with the desktop HDR setting turned on. It is possible for certain system configurations to automatically turn on/off the HDR capabilities prior to the playback of a HDR video, but, we didn’t take advantage of that in our testing.

VLC is the playback software of choice for the average PC user who doesn’t need a ten-foot UI. Its install-and-play simplicity has made it extremely popular. Over the years, the software has gained the ability to take advantage of various hardware acceleration options. Kodi, on the other hand, has a ten-foot UI making it the perfect open-source software for dedicated HTPCs. Support for add-ons make it very extensible and capable of customization. We played back our test files using the default VLC and Kodi configurations, and recorded the following metrics.

Playback is perfectly smooth in both VLC and Kodi for all streams except the 8Kp60 AV1 clip. Power consumption is below 20W for all streams decoded in hardware. While hardware decoding support for AV1 exists, this is a case of the software infrastructure yet to catch up with the hardware capabilities. With alpha releases of Kodi supporting AV1 hardware decoding, this scenario for the 8Kp60 AV1 clip is bound to change for the better in the coming months.

The power consumption at the wall was measured with a 4K display being driven through one of the HDMI ports in the system. In the graph below, we compare the idle and load power of the Intel NUC11TNBi5 (Akasa Newton TN) with other systems evaluated before. For load power consumption, we ran the AnandTech System Stress Test (a combination of Prime95, Furmark, and the AIDA64 System Stability Test with various stress components) and noted the peak as well as idling power consumption at the wall.

The numbers are consistent with the TDP and configured PL1 / PL2 values for the processors in the systems, and do not come as any surprise. The absence of the fan in the Newton TN build provides some idle power saving compared to the regular NUC11TNKi5.

Stress Testing

Our thermal stress routine is a combination of Prime95, Furmark, and Finalwire’s AIDA64 System Stability Test. The following 9-step sequence is followed, starting with the system at idle:

• Start with the Prime95 stress test configured for maximum power consumption
• After 30 minutes, add Furmark GPU stress workload
• After 30 minutes, terminate the Prime95 workload
• After 30 minutes, terminate the Furmark workload and let the system idle
• After 30 minutes of idling, start the AIDA64 System Stress Test (SST) with CPU, caches, and RAM activated
• After 30 minutes, terminate the previous AIDA64 SST and start a new one with the GPU, CPU, caches, and RAM activated
• After 30 minutes, terminate the previous AIDA64 SST and start a new one with only the GPU activated
• After 30 minutes, terminate the previous AIDA64 SST and start a new one with the CPU, GPU, caches, RAM, and SSD activated
• After 30 minutes, terminate the AIDA64 SST and let the system idle for 30 minutes

Traditionally, this test used to record the clock frequencies – however, with the increasing number of cores in modern processors and fine-grained clock control, frequency information makes the graphs cluttered and doesn’t contribute much to understanding the thermal performance of the system. The focus is now on the power consumption and temperature profiles to determine if throttling is in play. All the thermal stress tests reported in this section were conducted with the ambient temperature between 78F and 82F and no explicit airflow across the system under test.

The first set of graphs is for the actively-cooled Intel NUC11TNKi5. Starting off with the temperatures and fan speed, we see that the package briefly reaches 92C during the initial PL2 bursts. However, the fan kicks up to 3500 RPM and soon cools it down to around 70C. For regular PL1 loading, the fan stays at around 3150 RPM. The maximum sustained temperature for the package was recorded during the AIDA64 CPU, FPU, caches, and RAM stress – around 83C. The maximum SSD temperature was recorded to be around 65C during the AIDA64 SSD stress component.

Based on the temperatures recorded above, we do not expect to see any throttling in the picture. This is confirmed by the power numbers reported for the internal components, as well as the at-wall power consumption number.

In regions of pure CPU stress, the package power consumption is 28W right through. Addition of workload components like GPU and SSD stressing reduces this number somewhat as CPU cycles are spent controlling those peripherals also. For the GPU in particular, the power balance budgeting between the CPU and the GPU comes into play and prevents the full 28W capacity from being utilized. The maximum allowed GPU power consumption appears to be around 18W.

A similar stress sequence was applied to the Newton TN build. Analyzing the temperature graph first, a spike to 90C+ is seen in the beginning. However, the ramp up of the temperature during the Prime95 component is much steeper than the actively cooled version. The temperature during the Furmark-only segment reaches around 94C, but was still short of the junction temperature (105C). On the other hand, during first AIDA64 SST stress segment, the temperatures seemed to touch 102C and was still ramping up when the 30 minutes allotted to the segment had elapsed. While the VRM circuitry remained under safe limits (a maximum of 92C and stabilized), the SSD hit 86C during the stress test.

The temperatures recorded above appear to get stabilized within 30 minutes for all workloads except the AIDA64 SST’s first segment. In this context, the package power numbers would be interesting to analyze. We look at the collected power numbers below.

The power numbers reveal interesting aspects. The package power numbers for simultaneous Prime95 and Furmark remains solid at 28W, but the package temperature had started flattening around 86C. On the other hand, the first segment of AIDA64 SST also has a package power of 28W, but the temperature peak was 102C, and was still ramping up after 30 minutes. It appears that the case would eventually throttle the CPU assuming an extended duration for this stress segment. To analyze this further, we configured the segment to run alone for 4 hours while recording the same parameters.

As expected, the junction temperature (105C) was hit around 1 hour into the stress test and throttling kicked in to drop the package power and keep the package temperature around 100C. The package temperature oscillated between 25W and 26W during this duration. As an additional experiment, we repeated the same stress test after setting the PL1 limit to 25W in the BIOS.

Unfortunately, even with a PL1 of 25W, the junction temperature was still hit around 165 minutes into the test. This indicates that the problem is not one of handling a 28W TDP – rather it is an issue of drawing away the heat from a particular hot spot on the package that is triggered explicitly by the AIDA64 CPU, FPU, Caches, and RAM stress test.

Thermal Profile

One of the key aspects of fanless systems is the thermal profile under load. Our stress test saw the internal package temperature go as high as 105C. However, even with the package temperature oscillating around that mark, the highest temperature we recorded deep inside one of the ridges was just 70C. A FLIR One Pro thermal camera was used for this purpose.

The gallery below presents some of the other thermal pictures gathered during the course of the four hour stress test which managed to hit the junction temperature. The numbers in the pictures show that the chassis is not heat saturated.