The Lenovo Yoga Mini is a 130mm aluminum puck, 48.6mm tall and about 600 grams, that fits Intel’s 16-core Core Ultra X7 358H, 32GB of LPDDR5X-8533, and Arc B390 graphics into 0.65 liters. That is the same Panther Lake silicon Intel has been shipping in this year’s thin-and-light notebooks, moved to a desk with two Thunderbolt 4 ports, 2.5GbE, and a mini-desktop cooling config. Our 91E60002UT review unit lists at $1,869.99 at Best Buy, while Lenovo’s US store sells the equivalent 91E60000US configuration, which adds Lenovo’s 800 wireless keyboard and mouse, at $1,711.99 on sale from $2,150.99.
Lenovo sells the Yoga Mini as a Copilot+ PC first, with a 50 TOPS NPU, a fingerprint reader on the power button, Wi-Fi Sensing presence detection, and an Adaptive Lighting ring on the underside that reacts to touch and notifications. The hardware underneath is what’s interesting: a laptop-class SoC in a chassis with potentially more thermal room than a notebook, which raises the question of whether the Yoga Mini sustains clocks a laptop with the same chip cannot. For this review, we ran it through the client suite we use for compact systems to find out.
Lenovo Yoga Mini Specifications
| Especificação | Lenovo Yoga Mini (91E60002UT, as tested) |
|---|---|
| Visão geral da plataforma | |
| Subcontratante | Intel Core Ultra X7 358H (Series 3) 16 cores (4 P-core, 8 E-core, 4 LP E-core), 16 threads P-core 1.9GHz base, up to 4.8GHz 18 MB Intel Smart Cache |
| Gráficos | Intel Arc B390 GPU (integrated, shared memory) Up to 122 TOPS, DirectX 12 Ultimate |
| NPU | Até 50 TOPS Copiloto+ PC |
| Memória | 32GB LPDDR5X-8533, dual-channel Soldered, no slots, not upgradable |
| Armazenamento | 1TB M.2 PCIe Gen4 NVMe SSD (Kioxia KBG7BZNV1T02 as tested) One drive, up to 2TB |
| Sistema Operacional | Windows 11 Home |
| Conectividade | |
| Portas traseiras | 1x Thunderbolt 4 1x USB-C (USB 10Gbps) with USB PD-in 1x USB-C (USB 10Gbps) 1x USB-A (USB 10Gbps) 1 x HDMI 2.1 TMDS 1x 2.5GbE Ethernet (RJ-45) |
| Left Ports | 1x Thunderbolt 4 1x 3.5mm headphone / microphone combo jack |
| Lado direito | Power button with fingerprint reader |
| Suporte de exibição | Até 4 monitores independentes 3840×2160 at 60Hz via HDMI, Thunderbolt 4, and USB-C with DP |
| Sem Fios | Wi-Fi 7 (802.11be, 2×2) Bluetooth 5.4 (Bluetooth 6.0 hardware ready) |
| Design e Poder | |
| Dimensões (W x D x H) | 130 x 130 x 48.6mm (5.12 x 5.12 x 1.91 polegadas) |
| Volume | 0.65 litros |
| Peso | Around 0.6kg (1.3 lb) |
| Chassi | Aluminum, Seashell |
| em áudio | Realtek ALC3341S codec Alto-falante único de 2 W Microfones duplos |
| Fonte de alimentação do laboratório | 100W USB-C adapter (90 percent efficiency) |
| Total | Firmware TPM 2.0 Leitor de impressão digital Wi-Fi Sensing human presence detection |
| Garantia | 1-year base, extensions to 3 years |
| Price (as tested) | $1,869.99 (Best Buy, 91E60002UT) $1,711.99 (Lenovo.com, 91E60000US with keyboard and mouse) |
Construir e projetar
The Yoga Mini is round, which is unusual in a mini PC category dominated by cubes and rectangles. The aluminum shell is finished in what Lenovo calls Seashell, and at 0.65 liters, it’s slightly smaller than the 0.7-liter ASUS Ascent QN10.
The right side carries the power button, which doubles as the fingerprint reader for Windows Hello. Lenovo puts a single 2W speaker inside the chassis, enough for system sounds and video calls through the dual microphones that flank the Yoga logo on the front, though anyone doing real audio work will reach for the 3.5mm jack or a Thunderbolt dock. Power comes in over USB-C from a 100W adapter.
Across the back, from left to right: 2.5GbE Ethernet, HDMI 2.1 TMDS, two USB-C 10Gbps ports (one accepts the 100W adapter over power delivery, and both carry DisplayPort), a USB-A 10Gbps port, and the first of two Thunderbolt 4 ports. The HDMI, both USB-C ports, and both Thunderbolt 4 ports all drive displays, and Lenovo rates the system for four independent monitors at 3840×2160 and 60Hz.
The left side adds the second Thunderbolt 4 port and the 3.5mm combo jack. Thunderbolt 4 on both sides means a dock or a fast external NVMe drive can be plugged in without reaching around the back. Wireless is Wi-Fi 7 at 2×2 with Bluetooth 5.4 on the M.2 2230 card inside, with Bluetooth 6.0 support in the hardware pending a firmware update.
The underside is where the Yoga Mini gets interesting. A ring of radial vents surrounds the intake, and inside that ring sits a translucent LED strip for Lenovo’s Adaptive Lighting, which glows when the system wakes, pulses with media, and blinks for notifications. A tap on the chassis toggles the lighting, and a second tap is meant to trigger Lenovo Smart Connect for cross-device sharing once an over-the-air update enables it. The fan exhausts through the same ring, so the unit needs clearance on all sides to regulate thermals.
Serviceability and Upgradability
Lenovo documents top and base cover removal in its service manual, and we had the Yoga Mini open in a few minutes. With the center foot and base cover off, the motherboard sits exposed with the SSD facing up, and that is the only serviceable part of the system. The 32GB of LPDDR5X is soldered to the board, so the memory configuration chosen at purchase is permanent.
The M.2 2280 SSD sits under a finned aluminum heatsink with the M.2 2230 Wi-Fi card beside it, and each is held by a single Phillips screw once the cover is off. The slot is PCIe Gen4, ours shipped with the 1TB Kioxia KBG7BZNV1T02, and Lenovo lists support for drives up to 2TB, though we see no reason a larger 2280 drive would not work.
Pulling the board assembly the rest of the way out shows how Lenovo packed the system. The rear I/O lives on a daughterboard of its own, where most mini PCs solder the ports directly to the motherboard, and both boards ride on the system board bracket, which also carries the heatsink and fan. There is little reason to disassemble at this depth, but the layout is tidy for a compact 0.65-liter system.
Lenovo documents the full teardown for replacing every component on its support site. One detail we found curious: the official procedure for removing the top and bottom covers calls for a suction cup, since neither is screwed down.
Lenovo Yoga Mini Performance
Our Lenovo Yoga Mini review unit is the 91E60002UT configuration with an Intel Core Ultra X7 358H, 32GB of LPDDR5x-8533 memory, Intel Arc B390 integrated graphics, a 1TB Kioxia KBG7BZNV1T02 NVMe SSD, and Windows 11 Home. All benchmarks were run on the Lenovo Default power plan.
Comparables for a compact desktop built on a laptop platform come from both categories. The ASUS Ascent QN10 is the most direct rival, a mini PC of similar size and price built on the Snapdragon X2 Elite, which makes it the check on how Intel and Qualcomm compare in this form factor. The ASUS NUC14 Pro is two generations back, a Core Ultra Series 1 system from the Meteor Lake era that ASUS has since followed with the Arrow Lake NUC 15 Pro, and it shows how far the x86 mini PC has moved in two years.
O Dell Pro 5 14 Intel is a laptop, but it carries the Core Ultra X7 368H and the same Arc B390 graphics, so it shows what the same silicon family delivers in a notebook chassis and whether the Yoga Mini gains anything from its cooling layout.
Not every test in our suite was available on every comparison system. Where a system was not run on a given benchmark, we leave it out of that table, since older results would not be comparable, and where it was run on only part of a test, we mark the gap as N/A.
PCMark 10
PCMark 10 is UL Solutions’ general-purpose productivity benchmark. It measures the kind of work most people do all day: app start times, web browsing, and video conferencing in the Essentials group; spreadsheets and document editing in the Productivity group; and photo editing, rendering and visualization, and video editing in the Digital Content Creation group. The three group scores roll up into an overall score. Only the Dell Pro 5 14 Intel had PCMark results in our records among the comparables.
| PCMark 10 (quanto maior, melhor) | Lenovo Yoga Mini | Dell Pro 5 14 Intel |
|---|---|---|
| Pontuação geral | 10,369 | 7,945 |
| Essenciais | 11,965 | 10,751 |
| Produtividade | 20,073 | 13,821 |
| Criação de conteúdo digital | 12,597 | 9,158 |
The Yoga Mini posts 10,369 overall against 7,945 for the Dell Pro 5 14, a 31 percent lead, and it leads every group: Essentials at 11,965 against 10,751, Productivity at 20,073 against 13,821, and Digital Content Creation at 12,597 against 9,158. The widest margin is Productivity at 45 percent, where the spreadsheet and writing workloads reward the 358H’s sustained clocks over a laptop part; the narrowest is Essentials at 11 percent, where app launches and web browsing are closer to a wash.
Geekbench 7
Geekbench 7 is a cross-platform benchmark that runs a series of short workloads modeled on real applications, including file compression, HTML5 rendering, PDF rendering, photo filtering, ray tracing, and machine learning inference, and produces single-core and multi-core CPU scores. The GPU compute tests run the same kind of image processing, particle physics, and machine learning kernels through the OpenCL and Vulkan APIs, which makes them a useful look at how much general-purpose compute an integrated GPU can provide outside of gaming. Because Geekbench runs natively on both Arm and x86, it is one of the few tests where the Snapdragon-based Ascent QN10 and the Intel systems can be compared directly.
| Geekbench 7 (quanto maior, melhor) | Lenovo Yoga Mini (Arc B390) | ASUS Ascent QN10 (Adreno) | ASUS NUC 14 Pro (Arc) | Dell Pro 5 14 Intel (Arc B390) |
|---|---|---|---|---|
| CPU de núcleo único | 2,583 | 3,272 | 2,291 | 2,701 |
| CPU multinúcleo | 16,811 | 24,049 | 13,911 | 18,737 |
| GPU OpenCL | 51,571 | 30,982 | 27,790 | 54,664 |
| GPU Vulkan | 44,450 | 40,042 | 25,088 | 47,233 |
The CPU results put the Yoga Mini in the middle of the field. Its single-core score of 2,583 trails the Ascent QN10 at 3,272 and the Dell Pro 5 14 at 2,701, while its multi-core score of 16,811 lands well ahead of the NUC 14 Pro at 13,911 but behind the Dell at 18,737 and a long way behind the QN10 at 24,049. The Dell carries the higher-clocked X7 368H, so its lead on single-core work is expected, but the multi-core gap suggests Lenovo has tuned the Yoga Mini for quiet operation over maximum sustained clocks.
The GPU result is the one that separates the Yoga Mini from the field. An OpenCL score of 51,571 is 66 percent higher than the QN10 and 86 percent higher than the NUC 14 Pro, and only a few points behind the Dell at 54,664, while the Vulkan score of 44,450 clears the QN10 by 11 percent. For anyone who cares about GPU compute in a tiny box, the Arc B390 gives this system an advantage over the Snapdragon alternative.
Cinebench 2026
Cinebench 2026 is based on Maxon’s Redshift render engine and measures how quickly a system can render a complex scene using the CPU. The single-thread test isolates per-core performance and is a good proxy for how responsive a system feels in lightly threaded applications, while the multi-thread test loads every core and thread and rewards systems that can sustain high clocks across the whole package for the duration of the run. The MP ratio is the multi-thread score divided by the single-thread score, which shows how effectively a system scales across its cores.
| Cinebench 2026 (quanto maior, melhor) | Lenovo Yoga Mini | ASUS Ascent QN10 | ASUS NUC14 Pro | Dell Pro 5 14 Intel |
|---|---|---|---|---|
| Thread único da CPU | 509 | 639 | 444 | 530 |
| CPU com múltiplos núcleos | 4,473 | 6,476 | 3,684 | 3,656 |
| Taxa de MP | 8.79x | 10.13x | 8.30x | 6.90x |
The single-thread result of 509 is within 4 percent of the Dell Pro 5 14 and comfortably ahead of the NUC 14 Pro, though the QN10 at 639 is the clear leader in per-core throughput.
The multi-thread result says more about the chassis than the chip. At 4,473, the Yoga Mini is 22 percent ahead of the Dell Pro 5 14 despite the Dell carrying the higher-binned processor, and its 8.79x scaling ratio is much healthier than the 6.90x the Dell managed. The QN10 remains well ahead at 6,476, which is consistent with the Snapdragon X2 Elite’s extra performance cores, and the NUC 14 Pro trails the field at 3,684.
Perfil de CPU 3DMark
The 3DMark CPU Profile benchmark runs the same physics and simulation workload at six different thread counts, from a single thread up to the maximum the processor supports. Reading across the results shows how a processor scales as more cores are engaged and where a system stops gaining from additional threads, which is useful for understanding how it will behave in applications that fall somewhere between purely single-threaded and fully parallel.
| Perfil de CPU 3DMark (quanto maior, melhor) | Lenovo Yoga Mini | ASUS Ascent QN10 | ASUS NUC14 Pro | Dell Pro 5 14 Intel |
|---|---|---|---|---|
| Tópicos Máximos | 9,004 | 9,544 | 7,680 | 10,370 |
| Tópicos 16 | 8,929 | 9,118 | 6,899 | 10,355 |
| Tópicos 8 | 6,185 | 5,327 | 5,617 | 6,669 |
| Tópicos 4 | 4,184 | 3,209 | 3,582 | 4,256 |
| Tópicos 2 | 2,244 | 1,702 | 1,914 | 2,269 |
| 1 Tópico | 1,159 | 857 | 1,002 | 1,181 |
The Yoga Mini scales cleanly from 1,159 at a single thread to 9,004 at max threads. Its single-thread and two-thread results are the second-best in the group and well ahead of the QN10, which manages only 857 and 1,702 in this test, and the NUC 14 Pro at 1,002 and 1,914.
The Dell Pro 5 14 leads at every thread count, finishing at 10,370, which shows the X7 368H does have more short-burst headroom than the 358H in this configuration. The QN10 overtakes the Yoga Mini once 16 or more threads are engaged, but the gap at the top is only 6 percent, and the Yoga Mini is 17 percent ahead of the NUC 14 Pro at max threads.
Compressão 7-Zip
The 7-Zip benchmark uses the built-in test in the popular open-source archiving utility to measure compression and decompression throughput, reported in GIPS (giga instructions per second). Compression is sensitive to memory latency and cache behavior, while decompression scales more directly with core count and clock speed. It is a practical benchmark because file archiving is a common real-world task that loads the cores, cache, and memory together.
| 7-Zip 24.09 (GIPS, quanto maior, melhor) | Lenovo Yoga Mini | ASUS Ascent QN10 | Dell Pro 5 14 Intel |
|---|---|---|---|
| Compressão | 93.603 | 96.729 | 93.417 |
| Descomprimindo | 81.927 | 120.205 | 89.181 |
| Avaliação Total | 87.765 | 108.467 | 91.299 |
Compression throughput is essentially a three-way tie, with the Yoga Mini at 93.603 GIPS, the Dell at 93.417, and the QN10 slightly ahead at 96.729. Decompression is where the systems separate, as the Yoga Mini posted 81.927 GIPS against 89.181 for the Dell and 120.205 for the QN10, which pulls its total rating down to 87.765, about 4 percent behind the Dell and 19 percent behind the QN10. The NUC 14 Pro was not run in this test.
triturador de y
y-cruncher is a multithreaded program that computes the digits of Pi and other mathematical constants. It is a demanding test of both CPU throughput and memory bandwidth, since the larger digit counts stream enormous working sets through the memory subsystem. We run the standard Pi computation at 1 billion, 2.5 billion, and 5 billion digits, along with the BBP digit-extraction formula at 1, 10, and 100 billion, and report the elapsed time in seconds, so lower numbers are better. The 5 billion digit run was not completed on the Yoga Mini or the mini PC comparables, and we include it only for the Dell.
| y-cruncher (segundos, quanto menor, melhor) | Lenovo Yoga Mini | ASUS Ascent QN10 | ASUS NUC14 Pro | Dell Pro 5 14 Intel |
|---|---|---|---|---|
| Pi 1B | 28.755 | 71.531 | 36.561 | 29.405 |
| Pi 2.5B | 80.831 | 218.332 | N/D | 90.719 |
| Pi 5B | N/D | N/D | N/D | 204.685 |
| Pi BBP 1B | 1.799 | 5.844 | 2.306 | 1.718 |
| Pi BBP 10B | 19.580 | 67.701 | 25.144 | 20.212 |
| Pi BBP 100B | 223.731 | 778.722 | 296.970 | 256.416 |
This is the Yoga Mini’s best showing of the CPU suite. Its 28.755 seconds for 1 billion digits is the fastest time in the group, edging the Dell Pro 5 14 at 29.405 and leaving the NUC 14 Pro at 36.561 and the QN10 at 71.531 far behind. At 2.5 billion digits, the lead over the Dell grows to 11 percent, 80.831 seconds against 90.719, and the QN10 needed 218.332 seconds, nearly three times as long.
The BBP results follow the same pattern, with the Dell taking the 1 billion and 10 billion runs by a fraction of a second and the Yoga Mini pulling ahead on the 100 billion run by 13 percent. The LPDDR5x-8533 memory and the desktop thermal envelope are both helping here, since y-cruncher punishes any system that has to throttle midway through a long run.
Blender 5.2
Blender is the open-source 3D creation suite, and its benchmark renders three standard scenes, Monster, Junkshop, and Classroom, using the Cycles engine. We run the benchmark twice, once on the CPU and once on the GPU, and report samples per minute, so higher is better. The GPU runs use Intel’s oneAPI backend on the Arc B390. Only the Intel systems with Arc graphics have GPU results, since the Blender GPU path is not available on the Snapdragon-based QN10, and the NUC 14 Pro was run on the CPU only.
| Blender 5.2 (amostras/min, quanto maior, melhor) | Lenovo Yoga Mini (Arc B390) | ASUS Ascent QN10 | ASUS NUC14 Pro | Dell Pro 5 14 Intel (Arc B390) |
|---|---|---|---|---|
| GPU | ||||
| Monstro | 544.64 | N/D | N/D | 560.28 |
| sucata | 372.30 | N/D | N/D | 466.36 |
| Sala de aula | 393.19 | N/D | N/D | 414.57 |
| CPU | ||||
| Monstro | 122.87 | 188.04 | 109.16 | 129.61 |
| sucata | 89.19 | 133.21 | 81.08 | 90.52 |
| Sala de aula | 63.34 | 93.93 | 57.41 | 62.20 |
On the GPU, the Yoga Mini and the Dell Pro 5 14 are close on Monster, 544.64 against 560.28 samples per minute, and on Classroom, 393.19 against 414.57, but the Dell opens up a 25 percent lead on the memory-hungry Junkshop scene.
On the CPU side, the Yoga Mini lands within 5 percent of the Dell on Monster, within 2 percent on Junkshop, and slightly ahead on Classroom, and it is 10 to 13 percent ahead of the NUC 14 Pro on every scene. The QN10 is the CPU rendering leader by a wide margin at 188.04, 133.21, and 93.93, roughly 50 percent ahead of the Yoga Mini.
Lux Mark
LuxMark is an OpenCL benchmark built on the LuxCoreRender physically based render engine. It renders two scenes of increasing complexity, Food and Hall, on the GPU and reports a score based on samples per second. Because it runs through OpenCL and no vendor-specific API, it is a good measure of how well an integrated GPU handles general compute-heavy rendering outside of a game engine.
| LuxMark v4 (quanto maior, melhor) | Lenovo Yoga Mini | Dell Pro 5 14 Intel |
|---|---|---|
| salão | 3,430 | 3,287 |
| Alimentícia | 1,663 | 1,585 |
The Yoga Mini scored 3,430 on Hall and 1,663 on Food, both a few percent ahead of the Dell Pro 5 14 with the same Arc B390 at 3,287 and 1,585. This is the pattern we expect from the same GPU with a bit more thermal room, and it mirrors what we saw in the Geekbench OpenCL results.
V-Ray
The V-Ray benchmark from Chaos measures rendering throughput in the V-Ray render engine. The CPU test reports vsamples and the GPU test reports vpaths, and the two are not comparable with each other. It is a short, intense render that favors sustained clocks on both the CPU and the GPU.
| V-Ray (quanto maior, melhor) | Lenovo Yoga Mini | Dell Pro 5 14 Intel |
|---|---|---|
| GPU (vpaths) | 918 | 919 |
| CPU (vsamples) | 19,450 | N/D |
The GPU result of 918 vpaths is a dead heat with the Dell Pro 5 14 at 919, which is what two systems with the same Arc B390 should produce when the workload is not thermally limited. The CPU test returned 19,450 vsamples on the Yoga Mini, and we do not have CPU results for the comparables to put that in context.
Teste de velocidade Blackmagic RAW
The Blackmagic RAW Speed Test decodes 8K Blackmagic RAW footage at a 12:1 compression ratio and reports how many frames per second the system can process on the CPU and on the GPU. It is a direct look at how a system will handle the kind of high-resolution camera footage that video editors work with every day, and it is one of the tests where integrated graphics can take real work off the CPU.
| Teste de velocidade Blackmagic RAW (fps, quanto maior, melhor) | Lenovo Yoga Mini | Dell Pro 5 14 Intel |
|---|---|---|
| CPU 8K 12:1 | 67 | 76 |
| GPU 8K 12:1 | 81 | 86 |
The Yoga Mini decoded 8K 12:1 footage at 67 frames per second on the CPU and 81 frames per second on the GPU. Both trail the Dell Pro 5 14, which posted 76 and 86, by 12 percent and 6 percent, respectively.
Vídeo Topázio AI
Topaz Video AI is a widely used application for AI-based video upscaling, denoising, and frame interpolation. We run its built-in benchmark, which times each of the Topaz models at 1X, 2X, and 4X scaling factors along with the Apollo and Aion slow-motion models, and reports frames per second, so higher is better. The models lean on the GPU and, where supported, the NPU, and the benchmark reflects a commercial application people pay for, which a synthetic inference loop does not. The 16X Aion slow-motion model did not finish on either Intel system.
| Topaz Video AI (fps, quanto maior, melhor) | Lenovo Yoga Mini | Dell Pro 5 14 Intel |
|---|---|---|
| Artemis 1X / 2X / 4X | 5.89 / 5.19 / 1.87 | 6.27 / 5.23 / 1.88 |
| Íris 1X / 2X / 4X | 4.84 / 2.97 / 0.90 | 5.06 / 3.10 / 0.93 |
| Proteus 1X / 2X / 4X | 6.04 / 5.94 / 2.42 | 6.37 / 6.03 / 2.42 |
| Gaia 1X / 2X / 4X | 3.18 / 2.15 / 1.45 | 3.27 / 2.24 / 1.47 |
| Nyx 1X / 2X | 1.50 / 1.46 | 1.56 / 1.52 |
| Hyperion HDR 1X | 3.06 | 3.13 |
| 4X Slowmo Apollo / APFast | 8.08 / 22.91 | 8.52 / 24.27 |
| Aion em câmera lenta 16X | DNF | DNF |
Results across the Topaz models are remarkably consistent between the two Arc B390 systems. The Yoga Mini trails the Dell Pro 5 14 by roughly 3 to 6 percent on every model at every scaling factor, which lines up with the clock speed difference between the X7 358H and the X7 368H and shows no shortfall in the Yoga Mini’s thermal design. The 4X upscaling results below 2.5 frames per second on both systems are a reminder that integrated graphics remain a long way from a discrete GPU for this kind of work, but for 1X and 2X passes the Yoga Mini is a usable Topaz box.
Geração de imagens por IA Procyon
The Procyon AI Image Generation benchmark from UL Solutions runs Stable Diffusion 1.5 and Stable Diffusion XL inference through the vendor inference engine best suited to the hardware and reports a score based on how quickly images are generated. The FP16 tests run at full precision, and the INT8 tests use quantized models that run on either the GPU or the NPU. Scores are only comparable when the same engine and the same device are used, so we label the engine and device for every result.
The Yoga Mini and the Dell Pro 5 14 use Intel’s OpenVINO engine, the NUC 14 uses OpenVINO on its older Arc iGPU, and the Ascent QN10 runs the INT8 test on its Hexagon NPU through Qualcomm’s QNN engine. The NPU result on the Yoga Mini is listed separately so that it is not confused with the GPU run.
| Geração de imagens por IA Procyon (quanto maior, melhor) | Lenovo Yoga Mini | ASUS Ascent QN10 | ASUS NUC14 Pro | Dell Pro 5 14 Intel |
|---|---|---|---|---|
| SD 1.5 FP16 | 613 (OpenVINO, GPU) | N/D | N/D | 635 (OpenVINO, GPU) |
| SD 1.5 INT8 | 7,119 (OpenVINO, GPU) | 5,457 (QNN, NPU) | 1,004 (OpenVINO, GPU) | 7,693 (OpenVINO, GPU) |
| SD 1.5 INT8 (NPU) | 2,936 (OpenVINO, NPU) | N/D | N/D | N/D |
| SDXL FP16 | 727 (OpenVINO, GPU) | N/D | N/D | 646 (OpenVINO, GPU) |
On the FP16 test, the Yoga Mini scored 613, within 4 percent of the Dell Pro 5 14 at 635. The INT8 GPU test tells the same story at 7,119 against 7,693, and both dwarf the 1,004 that the NUC 14 Pro managed on the same engine, a reminder of how far Intel’s Arc graphics have come. The QN10’s 5,457 came from its NPU on a different engine and is not directly comparable, but it does show the Snapdragon system is competitive in image generation when the workload is steered to the right accelerator.
The Yoga Mini’s own NPU scored 2,936 on the INT8 test, less than half its GPU result, which is why we would expect most Stable Diffusion workloads on this system to run on the Arc B390. On the heavier SDXL test, the Yoga Mini pulled ahead of the Dell at 727 against 646, a 13 percent lead that again points to the desktop chassis sustaining GPU clocks through a long run.
Visão Computacional Procyon AI
The Procyon AI Computer Vision benchmark runs a set of common vision models, including MobileNet, ResNet-50, Inception, DeepLab, YOLO, and Real-ESRGAN, and reports a score that reflects inference throughput across all of them. It can target the CPU, GPU, or NPU through several inference engines, and as with image generation, the results are only comparable across systems when the same engine and device are used.
We ran the Yoga Mini through both the Windows ML engine, which gives us a like-for-like comparison with the NUC 14 Pro and the Dell Pro 5 14, and Intel’s OpenVINO engine, which is the only path that exposes the NPU on this platform. The Ascent QN10 was tested on Qualcomm’s SNPE engine, and its results are not comparable with the WinML numbers here, so it is omitted from this table.
| Procyon AI Computer Vision (higher is better) | Lenovo Yoga Mini | ASUS NUC14 Pro | Dell Pro 5 14 Intel |
|---|---|---|---|
| WinML | |||
| CPU | 139 | 107 | 119 |
| GPU | 373 | 135 | 398 |
| OpenVINO | |||
| CPU | 118 | N/D | N/D |
| GPU | 550 | N/D | N/D |
| NPU | 1,205 | N/D | N/D |
Through Windows ML the Yoga Mini posted 139 on the CPU and 373 on the GPU. The CPU result is the best in the group, 17 percent ahead of the Dell and 30 percent ahead of the NUC 14 Pro, while the GPU result trails the Dell at 398 by 6 percent and is nearly triple the NUC 14 Pro’s 135.
Switching to OpenVINO moved the GPU score up to 550 and exposed the NPU, which scored 1,205, more than triple the best GPU result through WinML. For a vision workload that can use the Intel NPU through OpenVINO, the Yoga Mini has a great deal more inference throughput available than the WinML column suggests.
3DMark Gaming Benchmarks
3DMark’s gaming tests put the Arc B390 through a range of rendering workloads at different resolutions and feature levels. Night Raid and Wild Life are the lightest, built for integrated graphics and mobile-class GPUs; Fire Strike is a DirectX 11 test at 1080p; Time Spy moves to DirectX 12 at 1440p; Solar Bay adds ray tracing on a lighter workload; and Steel Nomad, Port Royal, and Speed Way are the heavy modern DirectX 12 and ray-tracing tests built for discrete cards.
For an integrated GPU, the heavy tests mostly show how far it is from discrete-class hardware, while the lighter tests show what it can run at playable rates. We run the gaming suite only on the mini PCs, so the Dell Pro 5 14 isn’t in this table, and the NUC 14 Pro was not run on Wild Life or Night Raid.
| 3DMark (quanto maior, melhor) | Lenovo Yoga Mini | ASUS Ascent QN10 | ASUS NUC14 Pro |
|---|---|---|---|
| Night Raid | 45,900 | N/D | N/D |
| Vida selvagem | 42,152 | 19,996 | N/D |
| Ataque de fogo | 14,076 | 9,949 | 6,591 |
| Fire Strike Graphics / Physics | 16,841 / 26,944 | 10,615 / 25,253 | 7,137 / 28,289 |
| Espião do tempo | 7,308 | 4,014 | 3,643 |
| Time Spy Graphics / CPU | 6,774 / 13,225 | 3,590 / 12,141 | 3,269 / 10,403 |
| Baía Solar | 27,236 | 21,561 | 11,630 |
| Nômade de Aço | 1,595 | 1,042 | 611 |
| Port Royal | 4,091 | 1,520 | 1,429 |
| Caminho da velocidade | 975 | 324 | 335 |
On the graphics side, the Arc B390 is the best integrated GPU in this group by a wide margin. The Yoga Mini’s Time Spy graphics score of 6,774 is 89 percent ahead of the QN10’s 3,590 and more than double the NUC 14 Pro’s 3,269, and Fire Strike graphics at 16,841 is 59 percent ahead of the QN10 and 2.4x the NUC. Wild Life at 42,152 is more than twice the QN10, and Solar Bay at 27,236 is 26 percent ahead of the Snapdragon system and 2.3x the older Arc.
The ray-tracing results are where the gap is widest: Port Royal at 4,091 is 2.7x the QN10 and 2.9x the NUC 14 Pro, and Speed Way at 975 is triple either, which is the Battlemage-generation RT hardware showing up. Steel Nomad at 1,595 and Speed Way at 975 are still a long way from a discrete card, so this is a 1080p system for modern titles, but Night Raid at 45,900 and Wild Life at 42,152 say lighter games and older titles will run comfortably at the 1440p the system was tested at.
The CPU subscores line up with the rest of the review. Fire Strike Physics at 26,944 sits between the QN10’s 25,253 and the NUC 14 Pro’s 28,289, and the Time Spy CPU score of 13,225 leads both, 12,141 and 10,403.
SPECviewperf 15
SPECviewperf 15 is the industry standard for measuring professional graphics performance. It replays real workloads captured from applications including 3ds Max, Blender, CATIA, Creo, Maya, Siemens NX, SolidWorks, Unreal Engine, and Enscape, along with energy and medical visualization viewsets, and reports frames per second for each. We run the benchmark at 1080p on this class of system, and results can only be compared against other runs at the same resolution. Neither mini PC comparable has SPECviewperf results, so the Dell Pro 5 14, which was also run at 1080p, is the only point of comparison.
| SPECviewperf 15 (FHD, quanto maior, melhor) | Lenovo Yoga Mini | Dell Pro 5 14 Intel |
|---|---|---|
| 3dsmax-08 | 18.44 | 20.21 |
| liquidificador-01 | 20.48 | 21.19 |
| catia-07 | 9.86 | 10.27 |
| creo-04 | 30.80 | 32.53 |
| energy-04 | 8.60 | 10.78 |
| escape-01 | 14.46 | 14.28 |
| Maya-07 | 80.28 | 82.42 |
| médico-04 | 20.79 | 22.96 |
| snx-05 | 43.26 | 46.22 |
| solidworks-08 | 22.88 | 23.60 |
| unreal_engine-01 | 37.47 | 38.77 |
Across the eleven viewsets, the Yoga Mini lands a few percent behind the Dell Pro 5 14 on most, with the two systems within 3 percent on maya-07, solidworks-08, and unreal_engine-01, and the Yoga Mini slightly ahead on enscape-01 at 14.46 against 14.28. The widest gaps are on energy-04, where the Yoga Mini’s 8.60 is 20 percent behind, and medical-04 at 9 percent behind, both of which are volumetric viewsets that lean on memory bandwidth and clock speed. The 80.28 on maya-07 and 43.26 on snx-05 are strong results for integrated graphics, and this is a system that can drive light CAD and DCC viewport work.
Estação de trabalho SPEC 4
SPECworkstation 4 is the broadest test in our suite. It runs more than 50 workloads drawn from real professional applications and organizes the results into hardware subsystem scores for CPU, graphics, accelerator, and storage, along with industry vertical scores for AI and machine learning, energy, financial services, life sciences, media and entertainment, product design, and productivity and development. Each score is a ratio against a reference workstation, so higher is better, and the verticals are a good shorthand for how a system will fare in a specific line of work.
On the Yoga Mini, the Octave workload failed repeatedly, which means the CPU subsystem score and the Productivity and Development vertical that depends on it could not be computed. The Dell Pro 5 14 is the only comparable system with SPECworkstation 4 results, and its media and entertainment vertical did not finish.
| Estação de trabalho SPEC 4 (proporção SPEC, quanto maior, melhor) | Lenovo Yoga Mini | Dell Pro 5 14 Intel |
|---|---|---|
| Subsistemas de hardware | ||
| CPU | DNF | 1.19 |
| Gráficos | 1.60 | 1.68 |
| Acelerador | 2.25 | N/D |
| Armazenamento | 1.11 | 1.76 |
| Indústrias Verticais | ||
| IA e aprendizado de máquina | 1.44 | 1.36 |
| Energia | 1.34 | 1.18 |
| Serviços Financeiros | 1.01 | 0.78 |
| Ciências da Vida | 1.42 | 1.34 |
| Mídia e Entretenimento | 1.40 | DNF |
| Produto Design | 1.52 | 1.64 |
| Produtividade e Desenvolvimento | DNF | 1.10 |
Where the two systems overlap, the Yoga Mini is ahead on the AI and Machine Learning vertical at 1.44 against 1.36, on Energy at 1.34 against 1.18, on Financial Services at 1.01 against 0.78, and on Life Sciences at 1.42 against 1.34, all of which lean on sustained multi-core throughput. The Dell takes Graphics at 1.68 against 1.60, Product Design at 1.64 against 1.52, and Storage by a wide margin at 1.76 against 1.11, which is a drive difference and not a platform one.
Desempenho de Armazenamento
We measure storage two ways, starting with the 3DMark Storage benchmark, which replays real-world traces of game loading, installing, saving, and recording gameplay, and produces a single score that reflects how the drive behaves under the mixed, bursty access patterns of everyday use. The Blackmagic Disk Speed Test is a simpler sequential test that reports peak read and write throughput in megabytes per second, the kind of number that matters when moving large video files. The Yoga Mini ships with a 1TB Kioxia KBG7BZNV1T02 NVMe SSD.
| Armazenamento | Lenovo Yoga Mini | ASUS Ascent QN10 | ASUS NUC14 Pro | Dell Pro 5 14 Intel |
|---|---|---|---|---|
| Armazenamento 3DMark | 2,540 | 2,393 | 1,887 | 3,144 |
| Gravação Blackmagic (MB/s) | 5,151.1 | N/D | N/D | 8,747.3 |
| Leitura Blackmagic (MB/s) | 5,276.6 | N/D | N/D | 8,609.6 |
The Kioxia drive scored 2,540 in 3DMark Storage, which puts it ahead of both mini PC comparables, 6 percent over the QN10 at 2,393 and 35 percent over the NUC 14 Pro at 1,887, but behind the Dell Pro 5 14 at 3,144.
The sequential numbers follow the same order, with the Yoga Mini at 5,151 MB/s write and 5,277 MB/s read, solidly in PCIe Gen4 territory, while the Dell’s Gen5 drive is closer to 8,700 MB/s in both directions.
Conclusão
We’re normally dubious when it comes to systems that prioritize form over all else, but the little Lenovo Yoga Mini did quite a bit better than we expected, as it manages to be aesthetically pleasing and performs well for its class.
Overall, it’s well appointed with Intel’s Core Ultra X7 358H and Arc B390 graphics in a 0.65-liter aluminum puck with two Thunderbolt 4 ports, 2.5GbE, and 32GB of LPDDR5X-8533. On PCMark 10, the Yoga Mini scored 10,369 overall, 31 percent ahead of the Dell Pro 5 14 Intel with the same graphics and the higher-binned X7 368H, and its Cinebench 2026 multi-thread score of 4,473 was 22 percent ahead of that laptop, with a better 8.79x scaling ratio.
The GPU is the reason to pick this box over the Snapdragon-based ASUS Ascent QN10. The Arc B390 posted 51,571 in Geekbench OpenCL, 66 percent ahead of the QN10, and on the 3DMark gaming suite it led the QN10 by 59 percent on Fire Strike graphics, 89 percent on Time Spy graphics, and 2.7x to 3x on the ray-tracing tests. In Procyon, the Yoga Mini’s OpenVINO path reached 550 on the GPU and 1,205 on the NPU for computer vision, and its 7,119 on Stable Diffusion 1.5 INT8 is seven times what the NUC 14 Pro managed.
The compromises start with multi-core CPU throughput, where the QN10 wins every test, by 43 percent in Geekbench multi-core and by about 50 percent in Blender CPU rendering, so a buyer whose work is CPU-bound and runs on Arm might look there. Also, the memory is soldered, and the Kioxia SSD is a mid-pack Gen4 drive at 5,277MB/s read against the 8,610MB/s in the Gen5 Dell.
For a desk where space (and design) matters and the work leans on the internals Intel provides in this build, the Yoga Mini is a very capable mini system that ended up punching well above its 0.65-liter weight.
Amazon