Skip to the stories
Windows & PowerShell

MSI EdgeMesa N AI+ Puts RTX Spark N1X and 128GB on the Desk

MSI's 1.18-liter EdgeMesa N AI+ pairs the RTX Spark N1X with 128GB of memory. Its desktop cooler adds 4 to 11 percent in sustained CPU work over laptops.

MSI EdgeMesa N AI+ Puts RTX Spark N1X and 128GB on the Desk

The chip in MSI's EdgeMesa N AI+ isn't new to us. The 20-core Arm N1X with its 6,144-core Blackwell GPU has mostly shown up in laptops so far. What's new is the box. This 1.18-liter aluminum mini PC comes with 128GB of unified memory, a 4TB PCIe Gen5 SSD, 10GbE and Windows 11 Pro, and that makes it a useful test of what a desktop chassis and a 240W power adapter add to silicon that's already been measured in notebooks. My read is that the answer is more than you'd expect on sustained work, and less than you'd expect on everything else.

MLPerf Client v2.0 Token Rate chart in tokens per second. Llama 3.1 8B: MSI EdgeMesa N AI+ 56.3, ASUS ProArt GR1X 56.8, HP OmniBook Ultra 16 56.5, Dell XPS 16 Creator Edition 55.7; Phi-4-mini: MSI EdgeMesa N AI+ 96.8, ASUS ProArt GR1X 96.3, HP OmniBook Ultra 16 97, Dell XPS 16 Creator Edition 96.2
MSI EdgeMesa N AI+ three-quarter view showing the silver aluminum chassis and perforated front grille

What MSI Is Selling

MSI places the EdgeMesa N AI+ in its Edge AI PC category next to the EdgeXpert, a 1.2-liter desktop built on NVIDIA's DGX Spark platform. The EdgeXpert ships with DGX OS, a 200GbE ConnectX-7 port for clustering, and 64GB or 128GB of memory. The EdgeMesa takes the Windows side of that line. It runs the RTX Spark N1X under Windows 11 and offers a port set built around a desk with up to four displays. MSI showed it at Computex in June 2026 and pitches it as a compact desktop for AI developers, software engineers, system integrators and creators who want agents, RAG pipelines and coding assistants running locally.

MSI's product page claims up to 1 petaFLOP of FP4 AI performance, support for 120B-parameter LLMs with up to 1M tokens of context, and 1440p gaming above 100 fps. The page doesn't list the models, quantization levels or settings behind those numbers, so treat them as MSI's claims. The testing here covers token rates on models from 3B to 27B parameters and a 35B coding-agent model. The Windows client suite stops at that 35B agent, and 120B-class models only run in the Linux suite. The model-size claim therefore went untested on this hardware.

The configuration matters as much as the spec sheet. The LPDDR5X is unified memory shared by the CPU and GPU, so you pick your memory at purchase and can't upgrade it later. MSI lists up to 128GB and a single NVMe SSD of up to 4TB, which is what we tested. The US store carries two versions, under part numbers X-005US and X-006US. NVIDIA told us the 128GB unit carries a $7,499 MSRP and that RTX Spark desktops start at $4,399. It also said pricing varies by region, manufacturer and configuration, so check the exact SKU you're quoting.

Build, Ports and the Storage Trade-Off

The chassis is aluminum with an anti-fingerprint finish. It measures 152 x 147 x 53mm and weighs 1.32kg, which works out to 1.18 liters. That's a bit under the 1.3 liters MSI advertises and close to the EdgeXpert's 1.2 liters. The ASUS ProArt GR1X comes in at 1.15 liters, 150 x 150 x 51mm, so the two RTX Spark mini PCs take up roughly the same desk space. The front is a perforated grille under an MSI badge with no ports or controls. The base is perforated across most of its area, and power comes from a 240W adapter.

MSI EdgeMesa N AI+ front view with the MSI badge above the perforated grille in the StorageReview lab

Everything connects at the rear. The unit has the power button, four USB Type-C ports at 20Gbps, one HDMI 2.1b output, one 10GbE RJ45 port and a Kensington lock slot below a full-width vent. Three of the Type-C ports support DisplayPort 1.4 Alt Mode, and the fourth is the power input. With HDMI in use, that gives you up to four displays. MSI lists Wi-Fi 7 and Bluetooth 5.4. There's no USB Type-A port and no audio jack, so most desks will need a hub or dock.

MSI EdgeMesa N AI+ rear panel with power button, four USB Type-C ports, HDMI, 10GbE, and Kensington lock slot below the vent

MSI calls the cooling design the AI Pulse Cooling System. It's a vapor chamber with two fans, rated by MSI to sustain 140W. With the outer shell off, a fin stack runs the full width of the chassis behind the rear vent, above the port row. For comparison, the Dell XPS 16 Creator Edition fits the same N1X into a 17.8mm laptop chassis. A mini PC should separate itself on cooling, so the sustained-load and thermal results below are where this design has to prove its worth.

MSI EdgeMesa N AI+ with the outer shell removed showing the heatsink fin stack above the rear ports

Storage is a single M.2 NVMe slot inside the chassis. Getting to it takes some effort, because the inner sled slides out of the outer case. The drive sits under a thermal pad in a cutout of the inner frame. The 4TB Samsung module in our unit is the short M.2 2242 format. That narrows replacement options to 2242 drives, a smaller retail selection than the common 2280 length, and it makes the factory capacity choice more important. The GR1X takes the other route, with two M.2 2280 slots in a 1.15-liter chassis. If you expect to add storage later, that's a real difference between the two RTX Spark desktops.

Underside of the MSI EdgeMesa N AI+ with the base plate removed showing the M.2 SSD under a thermal pad

Test Setup: Memory Allocation and What We Compared

The review unit is the MS-B215 with the RTX Spark N1X, 128GB of LPDDR5X-9400 and the 4TB Samsung SSD. It runs Windows 11 Pro 26H1 on Arm in the Best performance power mode and was connected to a 4K external display. Out of the box, Windows presented 62.5GB of the unified memory to the GPU as dedicated memory, with another 47.9GB shared.

That's one physical pool split by policy. The NVIDIA app has a slider that sets the GPU allocation. The default sits close to an even split on both the 64GB and 128GB systems here, and the app warns that minimum and maximum settings may cause issues. NVIDIA advises keeping the default unless a workload needs more, since a bigger GPU share leaves less memory for Windows and other apps, and it recommends keeping at least 16GB for the CPU. NVIDIA also says agents and LLMs see little performance loss when they spill into system memory, and it suggests raising the allocation for image and video generation models that need more than the default gives them.

We tested every system at its default. Our suite sizes its model tiers to the dedicated amount, so a model that gets skipped on the 64GB systems might still run there with spillover or a larger allocation. Keep that in mind when you read the "not run" entries below.

The results come from sr-client 1.13, the Windows client suite covering local AI, content creation, CPU, graphics and storage. The AI tests use llama.cpp on CUDA for LLM serving, MLPerf Client and Ollama on CUDA for model loading, the coding agent and RAG, and stable-diffusion.cpp on Vulkan for image generation. Each figure is the median of repeated passes. Where a test ran more than once, the newest valid result is reported, and anything the suite flagged for run-to-run variance or a failed pass carries an asterisk. The two laptops were tested earlier on sr-client 1.11.

The comparison set starts with the ASUS ProArt GR1X, the other RTX Spark mini PC in the lab. It's a 1.15-liter desktop with 64GB of memory and a 1TB Micron Gen4 SSD, and it ran the same sr-client 1.13 release. The two fastest RTX Spark laptops at each memory size round out the set: the HP OmniBook Ultra 16, the 128GB system with a 2TB Sandisk Gen5 SSD, and the Dell XPS 16 Creator Edition, the 64GB system with a 1TB SK hynix drive.

Configuration CPU Memory GPU memory Boot drive
MSI EdgeMesa N AI+ RTX Spark N1X, 20 cores 128GB LPDDR5X-9400 62.5GB dedicated 4TB Samsung MZALC4T0HBL1, PCIe Gen5
ASUS ProArt GR1X RTX Spark N1X, 20 cores 64GB LPDDR5X-9400 30.5GB dedicated 1TB Micron MTFDKBA1T0QGN, PCIe Gen4
HP OmniBook Ultra 16 RTX Spark N1X, 20 cores 128GB LPDDR5X-9400 62.5GB dedicated 2TB Sandisk SN8050S, PCIe Gen5
Dell XPS 16 Creator Edition RTX Spark N1X, 20 cores 64GB LPDDR5X-9400 30.5GB dedicated 1TB SK hynix PVC10

All four machines run the same 20-core N1X. The two desktops run it at a higher power limit with larger coolers, so any differences come down to power, cooling, memory capacity and storage. An x64 build of Cinebench run under emulation crashed on all four, so there are no emulation results to report.

Local AI Performance

LLM Serving

The serving tests push llama.cpp with one, four and eight simultaneous requests, which is roughly how a small team sharing one box would load it. With a single request, the four systems land within 6 percent of each other on every model. The EdgeMesa managed 58.1 tokens per second on Ministral 3 3B, 42.5 on the 8B model and 13.0 on Qwen3.8 27B. One request leaves most of the chip idle, so cooling barely enters the picture.

LLM Serving, Qwen3.8 27B Output Throughput chart in output tokens per second. 1 request: MSI EdgeMesa N AI+ 13, ASUS ProArt GR1X 13.1, HP OmniBook Ultra 16 12.6, Dell XPS 16 Creator Edition 12.4; 4 requests: MSI EdgeMesa N AI+ 31.1, ASUS ProArt GR1X 31.4, HP OmniBook Ultra 16 28.1, Dell XPS 16 Creator Edition 27.5; 8 requests: MSI EdgeMesa N AI+ 39.8, ASUS ProArt GR1X 40, HP OmniBook Ultra 16 33.8, Dell XPS 16 Creator Edition 31.9

The gap opens as requests stack up. At four concurrent requests on Qwen3.8 27B, the EdgeMesa produced 31.1 tokens per second, against 28.1 on the HP and 27.5 on the XPS 16. At eight requests it reached 39.8, against 33.8 and 31.9. Those are leads of 18 and 25 percent over the laptops. Time to first token on the 27B model was 8.7 seconds on the MSI and 10.8 to 10.9 seconds on the laptops. Ministral 3 8B at four requests showed a smaller version of the same pattern, at 107.5 tokens per second against 104.1 and 96.3.

LLM Serving, Ministral 3 8B Output Throughput chart in output tokens per second. 1 request: MSI EdgeMesa N AI+ 42.5, ASUS ProArt GR1X 42.7, HP OmniBook Ultra 16 41.5, Dell XPS 16 Creator Edition 40.6; 4 requests: MSI EdgeMesa N AI+ 107.5, ASUS ProArt GR1X 109.3, HP OmniBook Ultra 16 104.1, Dell XPS 16 Creator Edition 96.3

The GR1X tracked the EdgeMesa at every step. It hit 31.4 tokens per second at four requests and 40.0 at eight on Qwen3.8 27B, 109.3 on Ministral 3 8B at four requests, and an 8.6-second first token on the 27B model. The two desktops finished within 2 percent of each other across the serving table, even though the GR1X has half the memory. Every model in that table fits in 64GB.

The serving table stops at the 27B class by design. Our Windows client suite doesn't include gpt-oss 120B or other models of that size, and those run in the Linux suite on larger unified-memory desktops and GPU workstations. MSI's 120B claim for this system is one we didn't test.

MLPerf Client and the Single-User Case

MLPerf Client sends one prompt at a time, and here the four systems effectively tie. The EdgeMesa generated 56.3 tokens per second on Llama 3.1 8B and 96.8 on Phi-4-mini. The GR1X managed 56.8 and 96.3, the HP 56.5 and 97.0, and the XPS 16 55.7 and 96.2. Time to first token favored the desktops slightly, at 0.25 seconds on the MSI and 0.24 on the GR1X for Llama 3.1 8B, against 0.28 and 0.31 on the laptops. For one person chatting with a small model, the mini PCs and the laptops feel the same.

Model Loading and the Coding Agent

Cold loads track the SSD and the memory path more than the chassis. The EdgeMesa took gpt-oss:20b from a cold start to first token in 7.00 seconds, between the HP at 6.70 and the XPS 16 at 7.73. The GR1X needed 10.94 seconds on its Gen4 drive. On the 21GB qwen3.6:35b model, the MSI loaded cold in 8.32 seconds, against 9.33 on the HP. The suite doesn't run the 35B model on the 64GB systems. Warm loads of gpt-oss:20b were within a few tenths of a second across all four, with the EdgeMesa at 3.96 seconds and the others at 3.68 to 3.69.

The coding agent asks a model to write and test code over several rounds, three times per configuration. On qwen3.6:35b, the EdgeMesa passed all three attempts in 145 seconds at 86.8 tokens per second. The 128GB HP passed one of three, flagged by the suite, and took 1,783 seconds. On gpt-oss:20b the picture is less clean. The MSI passed two of three attempts (flagged), the GR1X passed one (flagged), the HP passed two (flagged) and the XPS 16 passed all three in 360 seconds. Pass counts on that model vary from run to run, so don't read too much into any single figure there.

Running Speedometer alongside the agent showed how much the agent load hurts interactive work. The browser slowed by 1.78x on the EdgeMesa and 1.79x on the GR1X, against 2.35x on the HP and 2.79x on the XPS 16. That's the clearest sign so far that a desktop's cooling and power budget help when the machine is doing AI work in the background while someone is still using it.

Local AI, key results EdgeMesa N AI+ (128GB) GR1X (64GB) OmniBook Ultra 16 (128GB) XPS 16 Creator (64GB)
Qwen3.8 27B, 8 requests (tok/s) 39.8 40.0 33.8 31.9
Llama 3.1 8B, MLPerf (tok/s) 56.3 56.8 56.5 55.7
qwen3.6:35b cold load (s) 8.32 Not run 9.33 Not run
qwen3.6:35b agent, successful runs (of 3) 3 Not run 1* Not run
Speedometer slowdown beside agent (x) 1.78 1.79 2.35 2.79

*Flagged by the suite: at least one of the three agent runs did not pass all tests.

RAG Pipeline

The RAG test embeds a passage set, retrieves against it and answers questions with phi4-mini. The EdgeMesa ingested 114.7 passages per second, 3 percent ahead of the HP at 111.5 and 10 percent ahead of the XPS 16 at 104.4. The GR1X was level with the MSI at 115.6. Answer generation ran at 103.4 tokens per second on the MSI and the HP, 95.0 on the XPS 16 and 104.2 on the GR1X. The median question was answered in 0.23 seconds on both 128GB systems. Retrieval-backed assistants at this scale should be comfortable on any of the four.

Image Generation

Image generation is the one AI category where the EdgeMesa trailed the laptops. FLUX.2 klein 4B at Q4 produced 9.95 images per minute, against 11.01 on the HP and 11.17 on the XPS 16. Z-Image Turbo 6B at BF16 managed 4.03 images per minute, between the two laptops. These tests ran through Vulkan, and the laptops ran on the earlier suite release. The GR1X, on the current release, produced 10.07 images per minute on FLUX.2, level with the MSI. That points to the suite release as the likely source of the gap, though it's not a cause we confirmed separately.

Image Generation at 1024x1024 chart in images per minute. FLUX.2 klein 4B Q4_0: MSI EdgeMesa N AI+ 10, ASUS ProArt GR1X 10.1, HP OmniBook Ultra 16 11, Dell XPS 16 Creator Edition 11.2; Z-Image Turbo 6B BF16: MSI EdgeMesa N AI+ 4, ASUS ProArt GR1X 4.2, HP OmniBook Ultra 16 4.2, Dell XPS 16 Creator Edition 3.9

Qwen-Image 20B at Q4 ran on all four systems at about half an image per minute. The Q8 version ran on the two 128GB machines, at 0.51 images per minute on the EdgeMesa and 0.47 on the HP. The BF16 version belongs to the suite's workstation tier and didn't run on any of them. FLUX.2 klein Q8 and BF16 completed only one pass on the EdgeMesa, short of what the suite expects, so those results are left out.

CPU Performance and Sustained Load

Cinebench 2026 and the 30-Minute Loop

Cinebench isolates the chassis better than anything else in the suite. In a single multi-core run the EdgeMesa scored 7,284 points, 4 percent ahead of the HP at 7,001 and 10 percent ahead of the XPS 16 at 6,631. The GR1X came in level at 7,304. Single-core went the other way by a small margin, with the MSI at 548 and the GR1X at 553, against 561 and 565 on the laptops. That's what you'd expect from a test that loads one core and never stresses the cooler.

The sustained test loops the multi-core render for 30 minutes and compares the steady state with the first pass. All four systems held their first-pass score, so none of them throttled, and the desktops settled above the laptops. The EdgeMesa held 7,276 points at a 3,372MHz mean clock. The GR1X held 7,306 at 3,385MHz, the HP held 6,970 at 3,239MHz and the XPS 16 held 6,550 at 3,083MHz. That's a 4 to 11 percent sustained advantage for the MSI over the laptops, and it's the number that matters most for a desktop in this class.

Cinebench 2026 Multi Core chart in points. Multi core: MSI EdgeMesa N AI+ 7,284.4, ASUS ProArt GR1X 7,304.1, HP OmniBook Ultra 16 7,001.4, Dell XPS 16 Creator Edition 6,631.4; 30-minute loop, steady state: MSI EdgeMesa N AI+ 7,276.3, ASUS ProArt GR1X 7,305.9, HP OmniBook Ultra 16 6,970.1, Dell XPS 16 Creator Edition 6,549.8

Temperatures in the loop separate the chassis types. Both laptops ran at a 97C mean, pinned against their 98C ceiling. The EdgeMesa averaged 88.4C with peaks to 107.3C, which is the one place where the MSI ran hotter than its ceiling-bound rivals on paper. The GR1X ran the same loop at a slightly higher clock, averaged 83.3C and peaked at 88.7C. The two desktops delivered essentially the same sustained score, but the ASUS cooler did it with more margin under the chip's limit.

Render, Compression and Build Work

The all-core application tests repeat the Cinebench pattern at different sizes. In 7-Zip all-threads, the EdgeMesa scored 152.6 GIPS, against 153.8 on the GR1X, 150.6 on the HP and 147.0 on the XPS 16. Single-thread 7-Zip slightly favored the laptops at 9.10 and 9.13 GIPS, with both desktops at 8.92.

7-Zip 24.09 All Threads chart in GIPS. Total rating: MSI EdgeMesa N AI+ 152.6, ASUS ProArt GR1X 153.8, HP OmniBook Ultra 16 150.6, Dell XPS 16 Creator Edition 147

Blender 4.5 is where the desktop lead shows up most clearly. The Monster scene rendered at 215.1 samples per minute on the EdgeMesa, 2 percent ahead of the HP and 8 percent ahead of the XPS 16. The Classroom scene widened those leads to 3 and 10 percent. The GR1X matched the MSI in Monster and Classroom and sat 2 percent behind it in Junkshop, at 127.2 against 129.7.

Blender 4.5 CPU chart in samples per minute. Monster: MSI EdgeMesa N AI+ 215.1, ASUS ProArt GR1X 215.6, HP OmniBook Ultra 16 211.4, Dell XPS 16 Creator Edition 198.9; Junkshop: MSI EdgeMesa N AI+ 129.7, ASUS ProArt GR1X 127.2, HP OmniBook Ultra 16 127.7, Dell XPS 16 Creator Edition 119.1; Classroom: MSI EdgeMesa N AI+ 95.7, ASUS ProArt GR1X 95.7, HP OmniBook Ultra 16 92.8, Dell XPS 16 Creator Edition 86.7

Video encoding gives the mini PC its largest application lead. HandBrake's x265 encode ran at 43.7 fps on the EdgeMesa, against 40.6 on the HP and 38.8 on the XPS 16, leads of 8 and 13 percent. SVT-AV1 ran at 16.3 fps against 15.4 and 14.0. The GR1X encoded at 44.1 and 16.4 fps. In the LLVM build the MSI finished in 217 seconds, level with the HP at 218, 8 seconds ahead of the GR1X at 225 and 15 seconds ahead of the XPS 16 at 232.

HandBrake CPU Encode chart in frames per second. x265: MSI EdgeMesa N AI+ 43.7, ASUS ProArt GR1X 44.1, HP OmniBook Ultra 16 40.6, Dell XPS 16 Creator Edition 38.8; SVT-AV1: MSI EdgeMesa N AI+ 16.3, ASUS ProArt GR1X 16.4, HP OmniBook Ultra 16 15.4, Dell XPS 16 Creator Edition 14

Short, Lightly Threaded and HPC Work

Short and lightly threaded tests show little or no chassis effect. Stockfish with all threads ran at 14.42 Mnps on the EdgeMesa, between the HP at 15.19 and the XPS 16 at 13.87, and the GR1X hit 15.13. Primesieve to 10^12 on all threads finished in 9.40 seconds on the MSI, against 9.39 on the GR1X, 9.53 on the HP and 10.39 on the XPS 16. The single-thread Python tests and Speedometer 3.1 landed within a few percent of each other, with the XPS 16 slightly ahead in the browser at 36.5 against 34.7 on the MSI. Everyday responsiveness is the same across the desktops and laptops.

The memory-bound HPC kernels are identical across the four machines, and the numbers show it. STREAM Triad measured 122,423MB/s on the EdgeMesa and 124,529 on the GR1X, against 120,783 and 122,872 on the laptops. Memory latency on 4KiB pages ranged from 140.5ns to 143.6ns. GROMACS benchMEM landed at about 17.4 ns/day on all four. OpenBLAS DGEMM and LU results sat within 6 percent, with the HP slightly ahead. Bandwidth-bound work gains nothing from a bigger cooler.

Graphics, Encoding and Thermals

SuperTuxKart at its ultra preset averaged 269.8 fps at 1080p on the EdgeMesa. That's level with the HP at 269.7 and ahead of the XPS 16 at 248.6, with the GR1X highest at 274.0. At 4K on our external display, the EdgeMesa averaged 111.8 fps and the GR1X 114.1. The laptops didn't run the 4K test. FFmpeg's hardware HEVC encode through Media Foundation ran at 54 fps on the MSI, matching the HP at 53, with the XPS 16 at 61 and the GR1X at 65. The hevc_nvenc encoder wasn't available in the FFmpeg build the suite uses on Arm.

The thermal picture across the whole suite is where the desktop design earns its keep. In the mean CPU temperature data, the EdgeMesa ran 14 to 22 degrees C cooler than the laptops on the single-run Cinebench, Blender and LLVM build tests, finishing the same work sooner. The GR1X ran cooler again at the same throughput. Its mean CPU temperature sat 3 to 8 degrees C below the EdgeMesa across the seven CPU tests, and 11 to 18 degrees below it during LLM serving, RAG and image generation. Peaks on the GR1X stayed under 90C in nine of ten tests.

Mean CPU Temperature Under Load chart in degrees C. Cinebench 30-minute loop: MSI EdgeMesa N AI+ 88.4, ASUS ProArt GR1X 83.3, HP OmniBook Ultra 16 97.4, Dell XPS 16 Creator Edition 97.3; Blender 4.5: MSI EdgeMesa N AI+ 76, ASUS ProArt GR1X 70, HP OmniBook Ultra 16 96.7, Dell XPS 16 Creator Edition 94.2; LLVM build: MSI EdgeMesa N AI+ 70.5, ASUS ProArt GR1X 67.2, HP OmniBook Ultra 16 91.6, Dell XPS 16 Creator Edition 92.9; LLM serving: MSI EdgeMesa N AI+ 85, ASUS ProArt GR1X 73.4, HP OmniBook Ultra 16 81, Dell XPS 16 Creator Edition 94

Peak readings are less tidy for the MSI. Its maximums went above the laptops' 98C ceiling in a few tests, reaching 105.3C in Stockfish and 107.3C in the 30-minute loop. In GPU-heavy LLM serving the EdgeMesa averaged 85.0C, which sits between the HP at 81.0C and the XPS 16 at 94.0C. Across the storage test, the SSD stayed at 62C or below.

Storage Performance

The 4TB Samsung drive in the EdgeMesa links at PCIe Gen5 x4 and delivers Gen5 sequential speeds. On the 25GiB footprint it read at 13,315MB/s and wrote at 12,910MB/s, close to the 2TB Sandisk Gen5 drive in the HP at 13,598 and 11,899MB/s. Both Gen4 drives trailed badly. The GR1X managed 6,511 and 5,235MB/s, and the XPS 16 6,502 and 5,670MB/s. For a machine that will hold several multi-gigabyte model files, both capacity and read speed matter, and the model-load times above reflect that.

fio Sequential 128K, 25GiB chart in MB/s. Read: MSI EdgeMesa N AI+ 13,315, ASUS ProArt GR1X 6,511.4, HP OmniBook Ultra 16 13,597.8, Dell XPS 16 Creator Edition 6,502.5; Write: MSI EdgeMesa N AI+ 12,910.2, ASUS ProArt GR1X 5,235.4, HP OmniBook Ultra 16 11,898.7, Dell XPS 16 Creator Edition 5,670.2

Random performance at queue depth 32 across four jobs was mixed. The EdgeMesa wrote at 386,814 IOPS, well ahead of the HP at 234,888 and the XPS 16 at 256,856. It read at 331,187 IOPS, behind the HP at 449,043 and the GR1X at 552,652, and ahead of the XPS 16 at 300,063. The suite also records a peak across queue depths, where the Samsung drive reached 674,013 IOPS read and 608,063 IOPS write. The GR1X's Micron drive led in random reads but trailed badly in writes at 143,165 IOPS.

fio Random 4K QD32 x 4 Jobs, 25GiB chart in IOPS. Read: MSI EdgeMesa N AI+ 331,187, ASUS ProArt GR1X 552,652, HP OmniBook Ultra 16 449,043, Dell XPS 16 Creator Edition 300,063; Write: MSI EdgeMesa N AI+ 386,814, ASUS ProArt GR1X 143,165, HP OmniBook Ultra 16 234,888, Dell XPS 16 Creator Edition 256,856

Where the EdgeMesa Lands

The EdgeMesa answers a fair question: does leaving the laptop chassis buy you anything? Once the load is sustained, yes. It delivers 4 to 11 percent more multi-core throughput than the fastest N1X laptops, 8 to 13 percent more in x265 encoding and 18 to 25 percent more output when eight requests hit a 27B model, and it does so at lower average temperatures. Single-request AI work and short tasks look essentially the same on all four machines.

MSI EdgeMesa N AI+ rear panel beside an NVIDIA DGX Spark on the StorageReview lab bench

The 128GB configuration is the reason to choose this unit over a smaller one. It ran Qwen-Image at Q8 and a 35B coding-agent model that the suite didn't run on the 64GB RTX Spark systems. The 4TB Gen5 drive keeps those larger model files local and loads them quickly. Bigger memory lets you hold bigger models, but token rates drop as model size grows, so the extra capacity buys headroom more than speed. The real limits are the peak CPU readings above 100C in a few tests and the single M.2 2242 slot.

Against the GR1X, performance is a tie. The two machines finish within 2 percent of each other in the Cinebench loop, Blender, x265 encoding and LLM serving. The differences are cooling and expansion. The GR1X runs 3 to 8 degrees cooler on average during CPU work and has two M.2 2280 slots where the EdgeMesa has one 2242. The EdgeMesa's advantages come from its 128GB memory and 4TB Gen5 drive, and those two parts decided which model tiers ran on each unit.

On price, the 128GB EdgeMesa at $7,499 MSRP sits about $500 above the 128GB XPS 16 Creator Edition at $6,999.99 and about $1,270 above the 128GB OmniBook Ultra 16 at $6,229.99. The 64GB ProArt GR1X at $4,499 is the cheaper route to the same N1X on a desk. Buyers paying for the EdgeMesa get 10GbE, the larger memory, a 4TB Gen5 drive and the sustained-load gains. They also give up the portability a laptop offers.

Within MSI's own line, the choice depends on the operating system you need. Developers and creators who want Windows and the RTX Spark platform on a desk, with a 10GbE link and four displays, should look at the EdgeMesa. Teams that want DGX OS and clustering over a 200GbE link point to the EdgeXpert. What to watch next is whether NVIDIA's memory allocation guidance changes as more image and video models arrive, whether the suite release gap in image generation closes on the laptops, and whether the 120B claim gets tested on Windows at all. Until then, the answer for this box is clear on sustained work and mostly a wash everywhere else.

MSI EdgeMesa N AI+ Product Page

Source: Brian Beeler · www.storagereview.com

Same flag