2 Sources
[1]
MSI's RTX Spark mini PC beats Strix Halo at local AI, but Windows on Arm holds it back
As we close out 2026, running top-tier local models at home is becoming more and more feasible. I've spent the past week with MSI's EdgeMesa N AI+ RTX Spark-powered mini PC and have put it through its paces. Add How-To Geek on Google Add as Preferred Source Follow on Google Discover From running
[2]
MSI EdgeMesa N AI+ Series Review - The Full RTX Spark PC Experience Arrives
In a stylish, quiet, and compact aluminum chassis with 128GB of memory, RTX Spark delivers impressive AI, gaming, and productivity performance. NVIDIA's RTX Spark Superchip, or the N1X, combines a 20-core NVIDIA Grace CPU with a 6144-core RTX Blackwell GPU, making it the company's first all-in-one
Share
Copy Link
Micro-Star International's EdgeMesa N AI+ has landed in the hands of two reviewers, and it is one of the earliest chances to see what a desktop built on Nvidia's RTX Spark chip can actually do
1
2
.
Source: How-To Geek
The EdgeMesa is what the company calls an "AI Compact Desktop": a 1.2-liter, 2.91-pound aluminum mini desktop, essentially silent, powered by a 240-watt USB-C adapter
1
. TweakTown calls the N1X Nvidia's first all-in-one consumer chip, putting a CPU plus GPU on one piece of silicon2
. It competes with Advanced Micro Devices and Intel's flagship integrated platforms without using an x86 CPU2
. The N1X pairs a 20-core Arm CPU running at up to about 4GHz with an integrated Blackwell RTX GPU carrying 6,144 CUDA cores1
2
.
Source: TweakTown
It is on sale now at B&H Photo, Micro Center and direct from the company: $4,399 for the 64GB/1TB version, $7,499 for 128GB/4TB
1
. The company cut both prices from $4,699 and $7,999 the week of the review1
. TweakTown calls the 128GB configuration a luxury purchase in a market where memory prices are climbing2
.In How-To Geek's own tests the Spark platform beats Advanced Micro Devices' Strix Halo on CUDA-backed AI workloads, though Strix Halo wins on Vulkan
1
. Image generation is the widest CUDA gap in the reviewer's test: the Spark produced an SDXL image in about 22 seconds against 38 on Strix Halo, roughly half the time1
.The 128GB of unified memory is soldered in
1
2
. By the reviewer's measurement the GPU can use up to 105GB of it; TweakTown puts the assignable figure at 96GB1
2
. That is enough to run an 88GB model fully on the GPU1
. The reviewer ran two Qwen models at once with 40-60GB of memory still free1
. Claude-driven video editing also went through it, rendering a 15-minute 1080p video in about seven minutes1
.Real-world text generation ran at roughly 30-50 tokens per second in benchmarks, and about 23 in actual agent calls
1
. Memory bandwidth is around 300GB/s, against about 1.8TB/s for an RTX 50901
. The chip games as well. TweakTown calls the 1440p gaming decent, with some 1080p path tracing2
. How-To Geek measured Forza Horizon 6 at over 100 frames per second at 1440p with frame generation on1
.Related Stories
There is no USB4, no Thunderbolt and no high-speed clustering link like the DGX Spark has, so clustering multiple units is not possible
1
. The reviewer's biggest complaint is Windows on Arm: some game anticheats won't run, and there is no official Linux option1
. Support for vLLM, a widely used tool for serving AI models, is also immature on Arm1
. How-To Geek scored it 8/10 and says it suits people running larger local models at home, with the caveat to skip it if you need clustering, x86-only software or Linux1
."Windows on Arm is the single reason why I couldn't make the EdgeMesa my main desktop," the reviewer writes
1
.Summarized by
Navi
[1]
15 Oct 2026•Technology

21 Sept 2026•Technology

16 Nov 2026•Technology
