Verdict — Newly Optimized For Offices And Retail
It may not look like so at first glance, but the Cubi NUC AI+ 3MG is a significant leap forward over last year's Cubi NUC AI+ 2MG. There are many changes under the hood including a revised motherboard with removable RAM, an even easier servicing procedure, more useful ports, and a much smaller, inconspicuous design to better blend into more professional environments. In fact, this is one of the smallest Intel mini PCs supporting up to 128 GB of removable memory. The significant CPU performance boost also addresses one of the main concerns about the older Lunar Lake model.
Conversely, the smaller size entails certain compromises. The MicroSD reader from last year's model is now gone and there is no integrated speaker which would have made the system more versatile. Core temperatures are generally warmer and fan noise isn't necessarily better or worse than before. The relatively large power brick, which is almost as heavy as the mini PC itself, can be difficult to tuck away depending on the environment. And finally, while the high RAM capacity is a standout feature, other aspects have taken a backseat including Local storage and GPU performance as competing models like the GMK EVO-X1 or Asus NUC 15 Pro+ are each faster machines with more storage options.
Pros
Cons
Price and Availability
Newegg is currently shipping the barebones MSI Cubi NUC AI+ 3MG with Core Ultra 9 CPU for $830. If configured with 32 GB of RAM, the price jumps up to $1220.
The Cubi NUC AI+ 3MG succeeds last year's Cubi NUC AI+ 2MG by swapping out the Intel Lunar Lake CPU for newer Panther Lake options. The external design has changed as well to be noticeably smaller and lighter than before.
Our test unit is a higher-end configuration with the Core Ultra 9 386H CPU for approximately $1100 if configured with RAM and storage. Thankfully, barebones options are available at lower starting prices. SKUs with entry-level Core Ultra 5 and Core Ultra 7 are also available.
The Cubi NUC AI+ 3MG competes with other business-centric and retail/office mini PCs like the GMK EVO series, Asus NUC 15 Pro+, or the HP EliteBoard G1a AI.
More MSI reviews:
Specifications
Case
The chassis utilizes the same black plastic materials for similar impressions as before albeit in a 17 percent lighter and 43 percent smaller volume. Perhaps more impressively, the smaller design doesn't entail more ventilation grilles or any major external revisions.
Connectivity
The most notable connectivity feature of the Cubi 3MG is its External Power Switch Pin on the side of the unit. The port allows for a cabled Power button to easily turn on or shut off the system should the mini PC be placed somewhere hard to reach like under an office desk or behind a table or monitor.
Port options have been significantly revised when compared to last year's model. There are now fewer USB-A ports and no more SD card reader or audio jack in favor of more HDMI ports and USB-C ports.
Keep in mind that only one of two Thunderbolt 4 ports supports Power Delivery. This port can be used to power the mini PC should you have a compatible USB-C AC adapter in lieu of the included barrel AC adapter.
Communication
WLAN module has been upgraded from Wi-Fi 6E to Wi-Fi 7 with Bluetooth 6.0 support.
| Networking | |
| iperf3 transmit AXE11000 | |
| MSI Cubi NUC AI+ 3MG | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 2MG | |
| GMK NucBox K16 | |
| Geekom A5 Pro | |
| MSI Cubi NUC 1M | |
| iperf3 receive AXE11000 | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 3MG | |
| MSI Cubi NUC AI+ 2MG | |
| Geekom A5 Pro | |
| GMK NucBox K16 | |
| MSI Cubi NUC 1M | |
| iperf3 transmit AXE11000 6GHz | |
| GMK NucBox K16 | |
| MSI Cubi NUC AI+ 3MG | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC 1M | |
| MSI Cubi NUC AI+ 2MG | |
| iperf3 receive AXE11000 6GHz | |
| GMK NucBox K16 | |
| MSI Cubi NUC AI+ 3MG | |
| MSI Cubi NUC AI+ 2MG | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC 1M | |
Maintenance
Accessories And Warranty
The retail box includes the external power switch for connecting to the power switch pin mentioned above. The external switch is not very long at just ~47 cm, however.
Performance
Testing Conditions
We set MSI Center to Performance mode prior to running the benchmarks below.
Processor
The Cubi AI 3MG ships with the same Core Ultra 9 386H CPU as found on the latest Razer Blade 16. However, raw multi-thread performance is about 18 percent slower on the MSI due to limited Turbo Boost potential. When running CineBench R15 xT in a loop, for example, the initial score of 2792 points would drop to 2380 points after just a couple of rounds as shown by our comparison graph below.
Raw performance is nevertheless much faster than last year's Cubi AI+ 2MG by 50 to 60 percent even after accounting for any throttling.
Cinebench R15 Multi Loop
Cinebench R23: Multi Core | Single Core
Cinebench R20: CPU (Multi Core) | CPU (Single Core)
Cinebench R15: CPU Multi 64Bit | CPU Single 64Bit
Blender: v2.79 BMW27 CPU
7-Zip 18.03: 7z b 4 | 7z b 4 -mmt1
Geekbench 6.6: Multi-Core | Single-Core
Geekbench 5.5: Multi-Core | Single-Core
HWBOT x265 Benchmark v2.2: 4k Preset
LibreOffice : 20 Documents To PDF
R Benchmark 2.5: Overall mean
| CPU Performance rating | |
| Minisforum AI X1 Pro, HX 470 | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| Peladn HO5, AI 9 HX 370 | |
| Average Intel Core Ultra 9 386H | |
| MSI Cubi NUC AI+ 3MG | |
| GMK NucBox K13 | |
| HP EliteBoard G1a AI | |
| MSI Cubi NUC AI+ 2MG | |
| GMK NucBox K16 | |
| MSI Cubi NUC 1M | |
| Minix Elite EU512-AI -2! | |
| Geekom A5 Pro | |
| Cinebench R23 / Multi Core | |
| Minisforum AI X1 Pro, HX 470 | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| Peladn HO5, AI 9 HX 370 | |
| Average Intel Core Ultra 9 386H (12144 - 21100, n=6) | |
| MSI Cubi NUC AI+ 3MG | |
| Minix Elite EU512-AI | |
| GMK NucBox K16 | |
| HP EliteBoard G1a AI | |
| GMK NucBox K13 | |
| MSI Cubi NUC AI+ 2MG | |
| MSI Cubi NUC 1M | |
| Geekom A5 Pro | |
| Cinebench R23 / Single Core | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| Minisforum AI X1 Pro, HX 470 | |
| Average Intel Core Ultra 9 386H (2010 - 2101, n=6) | |
| MSI Cubi NUC AI+ 3MG | |
| Peladn HO5, AI 9 HX 370 | |
| MSI Cubi NUC AI+ 2MG | |
| GMK NucBox K13 | |
| MSI Cubi NUC 1M | |
| HP EliteBoard G1a AI | |
| Minix Elite EU512-AI | |
| GMK NucBox K16 | |
| Geekom A5 Pro | |
| Cinebench R20 / CPU (Multi Core) | |
| Minisforum AI X1 Pro, HX 470 | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| Peladn HO5, AI 9 HX 370 | |
| Average Intel Core Ultra 9 386H (4757 - 8299, n=6) | |
| MSI Cubi NUC AI+ 3MG | |
| Minix Elite EU512-AI | |
| GMK NucBox K16 | |
| HP EliteBoard G1a AI | |
| GMK NucBox K13 | |
| MSI Cubi NUC AI+ 2MG | |
| MSI Cubi NUC 1M | |
| Geekom A5 Pro | |
| Cinebench R20 / CPU (Single Core) | |
| Asus NUC 15 Pro+ | |
| Minisforum AI X1 Pro, HX 470 | |
| GMK EVO-T1 | |
| Average Intel Core Ultra 9 386H (767 - 796, n=6) | |
| MSI Cubi NUC AI+ 3MG | |
| MSI Cubi NUC AI+ 2MG | |
| Peladn HO5, AI 9 HX 370 | |
| GMK NucBox K13 | |
| MSI Cubi NUC 1M | |
| HP EliteBoard G1a AI | |
| Minix Elite EU512-AI | |
| GMK NucBox K16 | |
| Geekom A5 Pro | |
| Cinebench R15 / CPU Multi 64Bit | |
| Minisforum AI X1 Pro, HX 470 | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| Peladn HO5, AI 9 HX 370 | |
| Average Intel Core Ultra 9 386H (2197 - 3303, n=6) | |
| MSI Cubi NUC AI+ 3MG | |
| Minix Elite EU512-AI | |
| GMK NucBox K16 | |
| HP EliteBoard G1a AI | |
| GMK NucBox K13 | |
| MSI Cubi NUC AI+ 2MG | |
| Geekom A5 Pro | |
| MSI Cubi NUC 1M | |
| Cinebench R15 / CPU Single 64Bit | |
| Asus NUC 15 Pro+ | |
| Minisforum AI X1 Pro, HX 470 | |
| GMK EVO-T1 | |
| Peladn HO5, AI 9 HX 370 | |
| MSI Cubi NUC AI+ 2MG | |
| Average Intel Core Ultra 9 386H (249 - 314, n=6) | |
| GMK NucBox K13 | |
| HP EliteBoard G1a AI | |
| MSI Cubi NUC AI+ 3MG | |
| MSI Cubi NUC 1M | |
| GMK NucBox K16 | |
| Minix Elite EU512-AI | |
| Geekom A5 Pro | |
| Blender / v2.79 BMW27 CPU | |
| MSI Cubi NUC 1M | |
| Geekom A5 Pro | |
| MSI Cubi NUC AI+ 2MG | |
| GMK NucBox K13 | |
| HP EliteBoard G1a AI | |
| GMK NucBox K16 | |
| Minix Elite EU512-AI | |
| MSI Cubi NUC AI+ 3MG | |
| Average Intel Core Ultra 9 386H (156 - 299, n=6) | |
| GMK EVO-T1 | |
| Asus NUC 15 Pro+ | |
| Peladn HO5, AI 9 HX 370 | |
| Minisforum AI X1 Pro, HX 470 | |
| 7-Zip 18.03 / 7z b 4 -mmt1 | |
| Minisforum AI X1 Pro, HX 470 | |
| Average Intel Core Ultra 9 386H (5488 - 6493, n=6) | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| Peladn HO5, AI 9 HX 370 | |
| MSI Cubi NUC 1M | |
| MSI Cubi NUC AI+ 2MG | |
| GMK NucBox K13 | |
| HP EliteBoard G1a AI | |
| MSI Cubi NUC AI+ 3MG | |
| Geekom A5 Pro | |
| GMK NucBox K16 | |
| Minix Elite EU512-AI | |
| Geekbench 6.6 / Multi-Core | |
| Asus NUC 15 Pro+ | |
| Average Intel Core Ultra 9 386H (13776 - 17405, n=8) | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 3MG | |
| Peladn HO5, AI 9 HX 370 | |
| Minisforum AI X1 Pro, HX 470 | |
| HP EliteBoard G1a AI | |
| GMK NucBox K13 | |
| MSI Cubi NUC AI+ 2MG | |
| MSI Cubi NUC 1M | |
| Geekom A5 Pro | |
| GMK NucBox K16 | |
| Geekbench 6.6 / Single-Core | |
| Minisforum AI X1 Pro, HX 470 | |
| GMK EVO-T1 | |
| Average Intel Core Ultra 9 386H (2778 - 2919, n=8) | |
| Peladn HO5, AI 9 HX 370 | |
| MSI Cubi NUC AI+ 3MG | |
| GMK NucBox K13 | |
| MSI Cubi NUC AI+ 2MG | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC 1M | |
| HP EliteBoard G1a AI | |
| Geekom A5 Pro | |
| GMK NucBox K16 | |
| Geekbench 5.5 / Multi-Core | |
| Average Intel Core Ultra 9 386H (13098 - 17557, n=6) | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| Peladn HO5, AI 9 HX 370 | |
| MSI Cubi NUC AI+ 3MG | |
| Minisforum AI X1 Pro, HX 470 | |
| HP EliteBoard G1a AI | |
| Minix Elite EU512-AI | |
| GMK NucBox K13 | |
| MSI Cubi NUC 1M | |
| MSI Cubi NUC AI+ 2MG | |
| GMK NucBox K16 | |
| Geekom A5 Pro | |
| Geekbench 5.5 / Single-Core | |
| Average Intel Core Ultra 9 386H (1992 - 2162, n=6) | |
| Peladn HO5, AI 9 HX 370 | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 2MG | |
| GMK NucBox K13 | |
| MSI Cubi NUC AI+ 3MG | |
| Minisforum AI X1 Pro, HX 470 | |
| HP EliteBoard G1a AI | |
| MSI Cubi NUC 1M | |
| Asus NUC 15 Pro+ | |
| Minix Elite EU512-AI | |
| Geekom A5 Pro | |
| GMK NucBox K16 | |
| HWBOT x265 Benchmark v2.2 / 4k Preset | |
| Asus NUC 15 Pro+ | |
| Peladn HO5, AI 9 HX 370 | |
| Minisforum AI X1 Pro, HX 470 | |
| GMK EVO-T1 | |
| Average Intel Core Ultra 9 386H (13.2 - 24.8, n=6) | |
| MSI Cubi NUC AI+ 3MG | |
| HP EliteBoard G1a AI | |
| Minix Elite EU512-AI | |
| GMK NucBox K13 | |
| MSI Cubi NUC AI+ 2MG | |
| GMK NucBox K16 | |
| Geekom A5 Pro | |
| MSI Cubi NUC 1M | |
| LibreOffice / 20 Documents To PDF | |
| Minix Elite EU512-AI | |
| HP EliteBoard G1a AI | |
| Minisforum AI X1 Pro, HX 470 | |
| GMK NucBox K16 | |
| Geekom A5 Pro | |
| Peladn HO5, AI 9 HX 370 | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 3MG | |
| GMK NucBox K13 | |
| Average Intel Core Ultra 9 386H (38.5 - 56.1, n=6) | |
| MSI Cubi NUC AI+ 2MG | |
| MSI Cubi NUC 1M | |
| R Benchmark 2.5 / Overall mean | |
| Minix Elite EU512-AI | |
| GMK NucBox K16 | |
| Geekom A5 Pro | |
| MSI Cubi NUC AI+ 2MG | |
| HP EliteBoard G1a AI | |
| GMK NucBox K13 | |
| MSI Cubi NUC 1M | |
| GMK EVO-T1 | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC AI+ 3MG | |
| Average Intel Core Ultra 9 386H (0.4055 - 0.4384, n=6) | |
| Peladn HO5, AI 9 HX 370 | |
| Minisforum AI X1 Pro, HX 470 | |
Cinebench R23: Multi Core | Single Core
Cinebench R20: CPU (Multi Core) | CPU (Single Core)
Cinebench R15: CPU Multi 64Bit | CPU Single 64Bit
Blender: v2.79 BMW27 CPU
7-Zip 18.03: 7z b 4 | 7z b 4 -mmt1
Geekbench 6.6: Multi-Core | Single-Core
Geekbench 5.5: Multi-Core | Single-Core
HWBOT x265 Benchmark v2.2: 4k Preset
LibreOffice : 20 Documents To PDF
R Benchmark 2.5: Overall mean
* ... smaller is better
AIDA64: FP32 Ray-Trace | FPU Julia | CPU SHA3 | CPU Queen | FPU SinJulia | FPU Mandel | CPU AES | CPU ZLib | FP64 Ray-Trace | CPU PhotoWorxx | CPU CheckMate
| Performance rating | |
| Asus NUC 15 Pro+ -1! | |
| Average Intel Core Ultra 9 386H -1! | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 3MG -1! | |
| GMK NucBox K16 -1! | |
| MSI Cubi NUC AI+ 2MG -1! | |
| Geekom A5 Pro | |
| MSI Cubi NUC 1M -1! | |
| AIDA64 / FP32 Ray-Trace | |
| Asus NUC 15 Pro+ (v6.92.6600) | |
| GMK EVO-T1 (v6.92.6600) | |
| GMK EVO-T1 (v8.20.8100) | |
| Average Intel Core Ultra 9 386H (10001 - 21727, n=6) | |
| GMK NucBox K16 (v8.20.8100) | |
| MSI Cubi NUC AI+ 3MG (v8.20.8100) | |
| MSI Cubi NUC AI+ 2MG (v6.92.6600) | |
| Geekom A5 Pro (v8.20.8100) | |
| Geekom A5 Pro (v6.92.6600) | |
| MSI Cubi NUC 1M (v6.92.6600) | |
| AIDA64 / FPU Julia | |
| Asus NUC 15 Pro+ (v6.92.6600) | |
| GMK EVO-T1 (v6.92.6600) | |
| GMK EVO-T1 (v8.20.8100) | |
| Average Intel Core Ultra 9 386H (53370 - 113595, n=6) | |
| GMK NucBox K16 (v8.20.8100) | |
| MSI Cubi NUC AI+ 3MG (v8.20.8100) | |
| MSI Cubi NUC AI+ 2MG (v6.92.6600) | |
| Geekom A5 Pro (v8.20.8100) | |
| Geekom A5 Pro (v6.92.6600) | |
| MSI Cubi NUC 1M (v6.92.6600) | |
| AIDA64 / CPU SHA3 | |
| Asus NUC 15 Pro+ (v6.92.6600) | |
| GMK EVO-T1 (v6.92.6600) | |
| GMK EVO-T1 (v8.20.8100) | |
| Average Intel Core Ultra 9 386H (2776 - 5604, n=6) | |
| MSI Cubi NUC AI+ 3MG (v8.20.8100) | |
| GMK NucBox K16 (v8.20.8100) | |
| MSI Cubi NUC AI+ 2MG (v6.92.6600) | |
| MSI Cubi NUC 1M (v6.92.6600) | |
| Geekom A5 Pro (v6.92.6600) | |
| Geekom A5 Pro (v8.20.8100) | |
| AIDA64 / CPU Queen | |
| Asus NUC 15 Pro+ (v6.92.6600) | |
| Geekom A5 Pro (v6.92.6600) | |
| MSI Cubi NUC 1M (v6.92.6600) | |
| GMK EVO-T1 (v6.92.6600) | |
| MSI Cubi NUC AI+ 2MG (v6.92.6600) | |
| AIDA64 / FPU SinJulia | |
| GMK NucBox K16 (v8.20.8100) | |
| Asus NUC 15 Pro+ (v6.92.6600) | |
| GMK EVO-T1 (v6.92.6600) | |
| GMK EVO-T1 (v8.20.8100) | |
| Geekom A5 Pro (v6.92.6600) | |
| Geekom A5 Pro (v8.20.8100) | |
| Average Intel Core Ultra 9 386H (5314 - 7332, n=6) | |
| MSI Cubi NUC AI+ 3MG (v8.20.8100) | |
| MSI Cubi NUC AI+ 2MG (v6.92.6600) | |
| MSI Cubi NUC 1M (v6.92.6600) | |
| AIDA64 / FPU Mandel | |
| GMK EVO-T1 (v6.92.6600) | |
| GMK EVO-T1 (v8.20.8100) | |
| Asus NUC 15 Pro+ (v6.92.6600) | |
| GMK NucBox K16 (v8.20.8100) | |
| Average Intel Core Ultra 9 386H (28414 - 61578, n=6) | |
| MSI Cubi NUC AI+ 3MG (v8.20.8100) | |
| MSI Cubi NUC AI+ 2MG (v6.92.6600) | |
| Geekom A5 Pro (v6.92.6600) | |
| Geekom A5 Pro (v8.20.8100) | |
| MSI Cubi NUC 1M (v6.92.6600) | |
| AIDA64 / CPU AES | |
| Asus NUC 15 Pro+ (v6.92.6600) | |
| GMK EVO-T1 (v6.92.6600) | |
| GMK EVO-T1 (v8.20.8100) | |
| Average Intel Core Ultra 9 386H (46362 - 72112, n=6) | |
| GMK NucBox K16 (v8.20.8100) | |
| MSI Cubi NUC AI+ 3MG (v8.20.8100) | |
| MSI Cubi NUC AI+ 2MG (v6.92.6600) | |
| Geekom A5 Pro (v6.92.6600) | |
| Geekom A5 Pro (v8.20.8100) | |
| MSI Cubi NUC 1M (v6.92.6600) | |
| AIDA64 / CPU ZLib | |
| Asus NUC 15 Pro+ (v6.92.6600) | |
| GMK EVO-T1 (v6.92.6600) | |
| GMK EVO-T1 (v8.20.8100) | |
| Average Intel Core Ultra 9 386H (669 - 1271, n=6) | |
| MSI Cubi NUC AI+ 3MG (v8.20.8100) | |
| GMK NucBox K16 (v8.20.8100) | |
| MSI Cubi NUC AI+ 2MG (v6.92.6600) | |
| MSI Cubi NUC 1M (v6.92.6600) | |
| Geekom A5 Pro (v6.92.6600) | |
| Geekom A5 Pro (v8.20.8100) | |
| AIDA64 / FP64 Ray-Trace | |
| Asus NUC 15 Pro+ (v6.92.6600) | |
| GMK EVO-T1 (v6.92.6600) | |
| GMK EVO-T1 (v8.20.8100) | |
| Average Intel Core Ultra 9 386H (5138 - 11627, n=6) | |
| GMK NucBox K16 (v8.20.8100) | |
| MSI Cubi NUC AI+ 3MG (v8.20.8100) | |
| MSI Cubi NUC AI+ 2MG (v6.92.6600) | |
| Geekom A5 Pro (v8.20.8100) | |
| Geekom A5 Pro (v6.92.6600) | |
| MSI Cubi NUC 1M (v6.92.6600) | |
| AIDA64 / CPU PhotoWorxx | |
| Average Intel Core Ultra 9 386H (54042 - 89891, n=6) | |
| GMK EVO-T1 (v8.20.8100) | |
| GMK EVO-T1 (v6.92.6600) | |
| Asus NUC 15 Pro+ (v6.92.6600) | |
| MSI Cubi NUC AI+ 2MG (v6.92.6600) | |
| MSI Cubi NUC AI+ 3MG (v8.20.8100) | |
| MSI Cubi NUC 1M (v6.92.6600) | |
| Geekom A5 Pro (v6.92.6600) | |
| Geekom A5 Pro (v8.20.8100) | |
| GMK NucBox K16 (v8.20.8100) | |
| AIDA64 / CPU CheckMate | |
| Average Intel Core Ultra 9 386H (4051 - 4826, n=5) | |
| MSI Cubi NUC AI+ 3MG (v8.20.8100) | |
| GMK EVO-T1 (v8.20.8100) | |
| GMK NucBox K16 (v8.20.8100) | |
| Geekom A5 Pro (v8.20.8100) | |
System Performance
Geekbench AI: Quantized GPU 1.7 | Half Precision GPU 1.7 | Single Precision GPU 1.7
PCMark 10: Score | Essentials | Productivity | Digital Content Creation
CrossMark: Overall | Productivity | Creativity | Responsiveness
WebXPRT 3: Overall
WebXPRT 4: Overall
Mozilla Kraken 1.1: Total
| Performance rating | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 3MG | |
| MSI Cubi NUC AI+ 2MG | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC 1M -3! | |
| GMK NucBox K16 -3! | |
| Geekom A5 Pro | |
| Geekbench AI / Quantized GPU 1.7 | |
| MSI Cubi NUC AI+ 2MG | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 3MG | |
| Geekom A5 Pro | |
| Geekbench AI / Half Precision GPU 1.7 | |
| MSI Cubi NUC AI+ 2MG | |
| MSI Cubi NUC AI+ 3MG | |
| GMK EVO-T1 | |
| Asus NUC 15 Pro+ | |
| Geekom A5 Pro | |
| Geekbench AI / Single Precision GPU 1.7 | |
| MSI Cubi NUC AI+ 2MG | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 3MG | |
| Geekom A5 Pro | |
| PCMark 10 / Score | |
| MSI Cubi NUC AI+ 3MG | |
| GMK EVO-T1 | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC AI+ 2MG | |
| GMK NucBox K16 | |
| Geekom A5 Pro | |
| MSI Cubi NUC 1M | |
| PCMark 10 / Essentials | |
| MSI Cubi NUC 1M | |
| MSI Cubi NUC AI+ 3MG | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 2MG | |
| Asus NUC 15 Pro+ | |
| Geekom A5 Pro | |
| GMK NucBox K16 | |
| PCMark 10 / Productivity | |
| MSI Cubi NUC AI+ 3MG | |
| Geekom A5 Pro | |
| GMK NucBox K16 | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 2MG | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC 1M | |
| PCMark 10 / Digital Content Creation | |
| GMK EVO-T1 | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC AI+ 3MG | |
| MSI Cubi NUC AI+ 2MG | |
| GMK NucBox K16 | |
| MSI Cubi NUC 1M | |
| Geekom A5 Pro | |
| CrossMark / Overall | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 3MG | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC 1M | |
| MSI Cubi NUC AI+ 2MG | |
| GMK NucBox K16 | |
| Geekom A5 Pro | |
| CrossMark / Productivity | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 3MG | |
| MSI Cubi NUC 1M | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC AI+ 2MG | |
| GMK NucBox K16 | |
| Geekom A5 Pro | |
| CrossMark / Creativity | |
| MSI Cubi NUC AI+ 3MG | |
| GMK EVO-T1 | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC AI+ 2MG | |
| MSI Cubi NUC 1M | |
| GMK NucBox K16 | |
| Geekom A5 Pro | |
| CrossMark / Responsiveness | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 3MG | |
| MSI Cubi NUC 1M | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC AI+ 2MG | |
| GMK NucBox K16 | |
| Geekom A5 Pro | |
| WebXPRT 3 / Overall | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 2MG | |
| MSI Cubi NUC 1M | |
| MSI Cubi NUC AI+ 3MG | |
| Geekom A5 Pro | |
| Asus NUC 15 Pro+ | |
| GMK NucBox K16 | |
| WebXPRT 4 / Overall | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 3MG | |
| MSI Cubi NUC AI+ 2MG | |
| MSI Cubi NUC 1M | |
| Asus NUC 15 Pro+ | |
| Geekom A5 Pro | |
| GMK NucBox K16 | |
| Mozilla Kraken 1.1 / Total | |
| GMK NucBox K16 | |
| Geekom A5 Pro | |
| MSI Cubi NUC 1M | |
| MSI Cubi NUC AI+ 2MG | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 3MG | |
* ... smaller is better
| PCMark 10 Score | 8325 points | |
Help | ||
| AIDA64 / Memory Copy | |
| Average Intel Core Ultra 9 386H (75437 - 129946, n=6) | |
| MSI Cubi NUC AI+ 2MG | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 3MG | |
| MSI Cubi NUC 1M | |
| Geekom A5 Pro | |
| Geekom A5 Pro | |
| GMK NucBox K16 | |
| AIDA64 / Memory Read | |
| Average Intel Core Ultra 9 386H (81021 - 118858, n=6) | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC AI+ 2MG | |
| MSI Cubi NUC AI+ 3MG | |
| GMK EVO-T1 | |
| GMK EVO-T1 | |
| MSI Cubi NUC 1M | |
| Geekom A5 Pro | |
| Geekom A5 Pro | |
| GMK NucBox K16 | |
| AIDA64 / Memory Write | |
| MSI Cubi NUC AI+ 2MG | |
| Average Intel Core Ultra 9 386H (75959 - 135096, n=6) | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 3MG | |
| MSI Cubi NUC 1M | |
| GMK NucBox K16 | |
| Geekom A5 Pro | |
| Geekom A5 Pro | |
| AIDA64 / Memory Latency | |
| GMK EVO-T1 | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 3MG | |
| GMK NucBox K16 | |
| Asus NUC 15 Pro+ | |
| Average Intel Core Ultra 9 386H (87.2 - 116.9, n=6) | |
| Geekom A5 Pro | |
| Geekom A5 Pro | |
| MSI Cubi NUC AI+ 2MG | |
| MSI Cubi NUC 1M | |
* ... smaller is better
DPC Latency
Storage Devices
The system ships with the same Phison 1 TB ESR01TBTCCZ PCIe4 NVMe SSD as found on last year's Cubi 2MG.
| Drive Performance rating - Percent | |
| GMK EVO-T1 | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC AI+ 3MG | |
| MSI Cubi NUC 1M | |
| Geekom A5 Pro | |
| MSI Cubi NUC AI+ 2MG | |
| GMK NucBox K16 | |
* ... smaller is better
Disk Throttling: DiskSpd Read Loop, Queue Depth 8
GPU Performance
The integrated Graphics 4 Xe3 performs most similarly to last year's Arc 140V if not slightly faster by a few percentage points on average. The Core Ultra 9 386H does not feature the elite Arc B390 of the Panther Lake X7 series and so graphics performance is not a primary feature of the MSI Cubi.
| 3DMark 11 Performance | 7599 points | |
| 3DMark Cloud Gate Standard Score | 30367 points | |
| 3DMark Fire Strike Score | 6190 points | |
| 3DMark Time Spy Score | 3237 points | |
Help | ||
* ... smaller is better
| Performance rating - Percent | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 -1! | |
| MSI Cubi NUC AI+ 2MG | |
| MSI Cubi NUC AI+ 3MG | |
| GMK NucBox K16 | |
| MSI Cubi NUC 1M -2! | |
| Geekom A5 Pro -2! | |
| Cyberpunk 2077 - 1920x1080 Ultra Preset (FSR off) | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 3MG | |
| MSI Cubi NUC AI+ 2MG | |
| GMK NucBox K16 | |
| MSI Cubi NUC 1M | |
| Geekom A5 Pro | |
| Baldur's Gate 3 - 1920x1080 Ultra Preset AA:T | |
| MSI Cubi NUC AI+ 2MG | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 3MG | |
| GMK NucBox K16 | |
| GTA V - 1920x1080 Highest AA:4xMSAA + FX AF:16x | |
| MSI Cubi NUC AI+ 2MG | |
| MSI Cubi NUC AI+ 3MG | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| GMK NucBox K16 | |
| MSI Cubi NUC 1M | |
| Geekom A5 Pro | |
| Final Fantasy XV Benchmark - 1920x1080 High Quality | |
| GMK EVO-T1 | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC AI+ 2MG | |
| MSI Cubi NUC AI+ 3MG | |
| MSI Cubi NUC 1M | |
| GMK NucBox K16 | |
| Geekom A5 Pro | |
| Strange Brigade - 1920x1080 ultra AA:ultra AF:16 | |
| MSI Cubi NUC AI+ 2MG | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC AI+ 3MG | |
| GMK NucBox K16 | |
| MSI Cubi NUC 1M | |
| Dota 2 Reborn - 1920x1080 ultra (3/3) best looking | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 3MG | |
| MSI Cubi NUC AI+ 2MG | |
| MSI Cubi NUC 1M | |
| GMK NucBox K16 | |
| Geekom A5 Pro | |
| X-Plane 11.11 - 1920x1080 high (fps_test=3) | |
| Asus NUC 15 Pro+ | |
| GMK EVO-T1 | |
| MSI Cubi NUC AI+ 2MG | |
| GMK NucBox K16 | |
| MSI Cubi NUC AI+ 3MG | |
| Geekom A5 Pro | |
| F1 24 | |
| 1920x1080 Low Preset AA:T AF:16x | |
| GMK EVO-T1 | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC AI+ 2MG | |
| MSI Cubi NUC AI+ 3MG | |
| Average Intel Graphics 4 Xe3 Panther Lake iGPU (55.6 - 67.2, n=5) | |
| GMK NucBox K16 | |
| Geekom A5 Pro | |
| 1920x1080 Medium Preset AA:T AF:16x | |
| GMK EVO-T1 | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC AI+ 3MG | |
| Average Intel Graphics 4 Xe3 Panther Lake iGPU (52.7 - 64.3, n=5) | |
| GMK NucBox K16 | |
| Geekom A5 Pro | |
| 1920x1080 High Preset AA:T AF:16x | |
| GMK EVO-T1 | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC AI+ 2MG | |
| MSI Cubi NUC AI+ 3MG | |
| Average Intel Graphics 4 Xe3 Panther Lake iGPU (40.8 - 46.5, n=5) | |
| GMK NucBox K16 | |
| Geekom A5 Pro | |
| 1920x1080 Ultra High Preset AA:T AF:16x | |
| GMK EVO-T1 | |
| Asus NUC 15 Pro+ | |
| MSI Cubi NUC AI+ 2MG | |
| MSI Cubi NUC AI+ 3MG | |
| Average Intel Graphics 4 Xe3 Panther Lake iGPU (13.2 - 14.8, n=5) | |
| GMK NucBox K16 | |
Cyberpunk 2077 ultra FPS Chart
| low | med. | high | ultra | |
|---|---|---|---|---|
| GTA V (2015) | 157.9 | 149.5 | 51.1 | 20.4 |
| Dota 2 Reborn (2015) | 104.1 | 80.3 | 81.2 | 79.9 |
| Final Fantasy XV Benchmark (2018) | 64.3 | 34.6 | 25.1 | |
| X-Plane 11.11 (2018) | 41.7 | 33.2 | 28.6 | |
| Strange Brigade (2018) | 172.6 | 68 | 56.9 | 48.5 |
| Baldur's Gate 3 (2023) | 31.6 | 26.3 | 23.1 | 22.8 |
| Cyberpunk 2077 (2023) | 37.1 | 28.4 | 24.5 | 21.1 |
| F1 24 (2024) | 66.9 | 61.5 | 45.8 | 14.2 |
Emissions
System Noise
The internal fan is always active no matter the onscreen load much like what we observed on older Cubi models. It's quiet enough with no major throttling issues to be unnoticeable in an office or retail setting, but it can ramp up to the mid 40 dB(A) range when running more demanding loads for extended periods to be as loud as some gaming laptops.
Noise level
| Idle |
| 28 / 33.2 / 33.5 dB(A) |
| Load |
| 39.1 / 45.2 dB(A) |
![]() | ||
30 dB silent 40 dB(A) audible 50 dB(A) loud |
||
min: | ||
| MSI Cubi NUC AI+ 3MG Ultra 9 386H, Intel Graphics 4 Xe3 PTL | GMK NucBox K16 R7 7735HS, Radeon 680M | GMK EVO-T1 Ultra 9 285H, Arc 140T | Geekom A5 Pro R5 7530U, Vega 7 | Asus NUC 15 Pro+ Ultra 9 285H, Arc 140T | MSI Cubi NUC AI+ 2MG Core Ultra 7 258V, Arc 140V | MSI Cubi NUC 1M Core 7 150U, Iris Xe G7 96EUs | |
|---|---|---|---|---|---|---|---|
| Noise | -6% | -20% | 15% | -3% | -2% | 6% | |
| off / environment * (dB) | 23.6 | 23.6 -0% | 24.7 -5% | 25 -6% | 25.3 -7% | 24.8 -5% | 23.5 -0% |
| Idle Minimum * (dB) | 28 | 30.9 -10% | 38.7 -38% | 26.6 5% | 25.4 9% | 28 -0% | 29.9 -7% |
| Idle Average * (dB) | 33.2 | 31.6 5% | 38.8 -17% | 26.6 20% | 25.5 23% | 36.1 -9% | 29.9 10% |
| Idle Maximum * (dB) | 33.5 | 31.6 6% | 38.9 -16% | 26.7 20% | 37.3 -11% | 36.1 -8% | 32 4% |
| Load Average * (dB) | 39.1 | 45.3 -16% | 51.6 -32% | 33.5 14% | 44.4 -14% | 38.9 1% | 33.5 14% |
| Cyberpunk 2077 ultra * (dB) | 44.4 | 49.2 -11% | 52 -17% | 31.2 30% | 46.2 -4% | 37 17% | 39.1 12% |
| Load Maximum * (dB) | 45.2 | 51.7 -14% | 51.6 -14% | 33.8 25% | 52.7 -17% | 49.2 -9% | 40.5 10% |
| Witcher 3 ultra * (dB) | 38.7 | 31.2 | 47.7 |
* ... smaller is better
Temperature
Surface temperatures are relatively cool partly due to the plastic chassis and efficient CPU. The rear becomes slightly warmer than the front as to be expected.
Ventilation grilles are kept to a minimum unlike on some alternatives like the GMK EVO series.
(+) The maximum temperature on the upper side is 36.2 °C / 97 F, compared to the average of 35.4 °C / 96 F, ranging from 20.6 to 69.5 °C for the class Mini PC.
(+) The bottom heats up to a maximum of 34.6 °C / 94 F, compared to the average of 36.1 °C / 97 F
(+) In idle usage, the average temperature for the upper side is 22.8 °C / 73 F, compared to the device average of 32.1 °C / 90 F.
| MSI Cubi NUC AI+ 3MG Ultra 9 386H, Intel Graphics 4 Xe3 PTL | GMK NucBox K16 R7 7735HS, Radeon 680M | GMK EVO-T1 Ultra 9 285H, Arc 140T | Geekom A5 Pro R5 7530U, Vega 7 | Asus NUC 15 Pro+ Ultra 9 285H, Arc 140T | MSI Cubi NUC AI+ 2MG Core Ultra 7 258V, Arc 140V | |
|---|---|---|---|---|---|---|
| Heat | 8% | -26% | -8% | -25% | 3% | |
| Maximum Upper Side * (°C) | 36.2 | 28.4 22% | 36.4 -1% | 36.7 -1% | 42.2 -17% | 30.8 15% |
| Maximum Bottom * (°C) | 34.6 | 29.4 15% | 48.5 -40% | 40.5 -17% | 31.3 10% | 26.4 24% |
| Idle Upper Side * (°C) | 23.2 | 23.4 -1% | 29.6 -28% | 23.1 -0% | 34.9 -50% | 26.4 -14% |
| Idle Bottom * (°C) | 22.4 | 23.6 -5% | 30 -34% | 25.1 -12% | 32.2 -44% | 25.4 -13% |
* ... smaller is better
Stress Test
The CPU would boost to 3 GHz, 61 W, and 92 C when initiating Prime95 stress. After 20 seconds, however, it would dip down to just 2.1 GHz and 30 W in order to maintain a core temperature of 88 C. The limited Turbo Boost potential supports our power consumption and CineBench R15 observations.
Core temperature would stabilize at 80 C when running Cyberpunk 2077 compared to only 63 C on last year's NUC 2MG.
| Average CPU Clock (GHz) | GPU Clock (MHz) | Average CPU Temperature (°C) | |
| System Idle | -- | -- | 37 |
| Prime95 Stress | 2.1 | -- | 88 |
| Prime95 + FurMark Stress | 1.4 | 2000 | 79 |
| Cyberpunk 2077 Stress | 0.7 | 2500 | ~80 |
Energy Management
Power Consumption
The Cubi AI+ 3MG is noticeably more efficient than last year's Cubi AI+ 2MG especially in terms of performance-per-watt. When running Prime95, for example, consumption would stabilize at around 52 W compared to 49 W on the older model despite the massive 60 to 65 percent increase in multi-thread performance. Graphics performance-per-watt has increased as well albeit by a much smaller degree.
Note that consumption can temporarily spike to as high as 98 W before falling back down due to CPU Turbo Boost limitations. The 120 W AC adapter (~13 x 6.9 x 2.5 cm) is more than sufficient in this regard as the mini PC itself barely averages more than 50 W to 60 W when under load.
| Off / Standby | |
| Idle | |
| Load |
|
Key:
min: | |
| MSI Cubi NUC AI+ 3MG Ultra 9 386H, Intel Graphics 4 Xe3 PTL, Phison 1 TB ESR01TBTCCZ-27J, , x, | GMK NucBox K16 R7 7735HS, Radeon 680M, TWSC TSC3AN1T0-F6Q10S, , x, | GMK EVO-T1 Ultra 9 285H, Arc 140T, Crucial P3 Plus CT1000P3PSSD8, , x, | Geekom A5 Pro R5 7530U, Vega 7, Wodposit WPBSN4M8-1TGP, , x, | Asus NUC 15 Pro+ Ultra 9 285H, Arc 140T, Micron 3500 1TB MTFDKBA1T0TGD, , x, | MSI Cubi NUC AI+ 2MG Core Ultra 7 258V, Arc 140V, Phison 1 TB ESR01TBTCCZ-27J, , x, | MSI Cubi NUC 1M Core 7 150U, Iris Xe G7 96EUs, WD PC SN560 SDDPNQE-1T00, , x, | |
|---|---|---|---|---|---|---|---|
| Power Consumption | -10% | -27% | 42% | -12% | 20% | 8% | |
| Idle Minimum * (Watt) | 6.2 | 7.4 -19% | 6.1 2% | 3.2 48% | 3.9 37% | 2.9 53% | 6.4 -3% |
| Idle Average * (Watt) | 7.9 | 7.7 3% | 7.9 -0% | 4.3 46% | 6.1 23% | 5.7 28% | 8.3 -5% |
| Idle Maximum * (Watt) | 8.5 | 7.7 9% | 13.2 -55% | 5.1 40% | 8.2 4% | 10.2 -20% | 10.4 -22% |
| Load Average * (Watt) | 41.4 | 52 -26% | 60.7 -47% | 32.1 22% | 56.1 -36% | 38.7 7% | 38.5 7% |
| Cyberpunk 2077 ultra external monitor * (Watt) | 47.6 | 68 -43% | 67.4 -42% | 29.4 38% | 63.9 -34% | 41.1 14% | 36.1 24% |
| Load Maximum * (Watt) | 98 | 80.1 18% | 117.1 -19% | 43.5 56% | 160.6 -64% | 60 39% | 52.5 46% |
* ... smaller is better
Power Consumption Cyberpunk / Stress Test
Power Consumption external Monitor
Notebookcheck Rating
The MSI Cubi NUC AI+ 3MG is one of the smallest mini PCs you can find with a Panther Lake Core Ultra 9 CPU. However, a couple of missing features prevent it from being the ultimate retail or office solution.
MSI Cubi NUC AI+ 3MG
- 06/03/2026 v8
Allen Ngo
Potential Competitors in Comparison
Image | Model / Review | Price | Weight | Height | Display |
|---|---|---|---|---|---|
| MSI Cubi NUC AI+ 3MG Intel Core Ultra 9 386H ⎘ Intel Graphics 4 Xe3 Panther Lake iGPU ⎘ 32 GB Memory, 1000 GB SSD | Amazon: List Price: 830 USD | 552 g | 37.5 mm | x | |
| GMK NucBox K16 AMD Ryzen 7 7735HS ⎘ AMD Radeon 680M ⎘ 32 GB Memory, 1024 GB SSD | Amazon: List Price: 650 USD | 567 g | 63 mm | x | |
| GMK EVO-T1 Intel Core Ultra 9 285H ⎘ Intel Arc Graphics 140T ⎘ 64 GB Memory, 1024 GB SSD | Amazon: List Price: 999€ | 946 g | 73.6 mm | x | |
| Geekom A5 Pro AMD Ryzen 5 7530U ⎘ AMD Radeon RX Vega 7 ⎘ 16 GB Memory, 1024 GB SSD | Amazon: List Price: 619€ | 423 g | 37 mm | x | |
| Asus NUC 15 Pro+ Intel Core Ultra 9 285H ⎘ Intel Arc Graphics 140T ⎘ 32 GB Memory, 1024 GB SSD | Amazon: $1,698.96 List Price: 991€ | 690 g | 42 mm | x | |
| MSI Cubi NUC AI+ 2MG Intel Core Ultra 7 258V ⎘ Intel Arc Graphics 140V ⎘ 32 GB Memory, 1000 GB SSD | Amazon: List Price: 900 USD | 669 g | 50.1 mm | x | |
| MSI Cubi NUC 1M Intel Core 7 150U ⎘ Intel Iris Xe Graphics G7 96EUs ⎘ 16 GB Memory, 1024 GB SSD | Amazon: List Price: 950 USD | 660 g | 50.1 mm | x |
Transparency
The selection of devices to be reviewed is made by our editorial team. The test sample was provided to the author as a loan by the manufacturer or retailer for the purpose of this review. The lender had no influence on this review, nor did the manufacturer receive a copy of this review before publication. There was no obligation to publish this review. As an independent media company, Notebookcheck is not subjected to the authority of manufacturers, retailers or publishers.
This is how Notebookcheck is testing
Every year, Notebookcheck independently reviews hundreds of laptops and smartphones using standardized procedures to ensure that all results are comparable. We have continuously developed our test methods for around 20 years and set industry standards in the process. In our test labs, high-quality measuring equipment is utilized by experienced technicians and editors. These tests involve a multi-stage validation process. Our complex rating system is based on hundreds of well-founded measurements and benchmarks, which maintains objectivity. Further information on our test methods can be found here.












































