AI & ComputeLOT 261122 · 4 min read
MSI's $7,499 RTX Spark Desktop Brings GB10 AI Power to Windows
MSI's $7,499 EdgeMesa N AI+ pairs Nvidia's GB10 chip, 128GB unified memory and Windows 11 Pro. It matches DGX Spark AI performance but drops clustering NIC and ships only three 20 Gbps USB-C ports.
Key takeaways
- MSI EdgeMesa N AI+ costs $7,499.99 with 128GB unified memory and 4TB SSD
- AgentPerf-Local performance matches Surface Laptop Ultra on Qwen 3.8 27B with DFlash2 decoding
- AMD Ryzen AI Halo takes four times as long on the same Windows LLM benchmark
- Three 20 Gbps USB-C ports and no ConnectX clustering NIC versus DGX Spark
- Noise measured 38.2 dBA under LLM inference at about 140W power draw

MSI has priced its first RTX Spark Windows desktop at $7,499.99. The EdgeMesa N AI+ pairs Nvidia's GB10 superchip with 128GB of unified memory and a 4TB NVMe SSD in a mini PC aimed at creators and local AI users who want to avoid Linux entirely.
The system carries the same silicon Tom's Hardware tested in Microsoft's Surface Laptop Ultra: a 6144-core, 48-SM Blackwell integrated GPU and a 20-core Arm CPU built from ten X925 performance cores and ten A725 efficiency cores. Memory is LPDDR5X with 273 GB/s of shared bandwidth. Windows 11 Pro ships preinstalled.
How does it perform on AI workloads?
In the AgentPerf-Local inference benchmark, the EdgeMesa delivered performance practically identical to the Surface Laptop Ultra running the same Qwen 3.8 27B model with DFlash2 speculative decoding. The desktop chassis adds no performance headroom for that configuration, though other workloads could differ.
Against AMD's Ryzen AI Halo, the only other Windows-ready platform with a large unified memory pool, the gap is wide. AMD's LLM performance under Windows with the Vulkan backend took four times as long as the EdgeMesa to complete the benchmark, and earlier testing under Linux with ROCm also trailed GB10. Reviewer Jeff Kampman concluded that a GB10 system is the superior choice for local AI in late 2026, especially where tokens equal dollars.
CPU results are closer. Geekbench 7 shows the Ryzen AI Halo slightly ahead in single-threaded score and behind in multithreaded work under Windows. Apple's M4 Max Mac Studio still holds a large lead in the benchmark.
What are the trade-offs versus DGX Spark?
The EdgeMesa omits the ConnectX RDMA NIC that lets DGX Spark systems cluster together to run larger AI models. That omission is permanent for the RTX path. Buyers who already own a capable Windows PC or Mac can link it to a DGX Spark through the Nvidia Sync app for many local AI jobs.
The pricing gap is narrow. MSI's own EdgeXpert with 128GB of memory and a 4TB SSD sells for $8,099 on sale, or $8,499 at full price, roughly $1,000 more for the clustering capability.
Peripheral connectivity is thin for the price. The rear panel offers three USB 3.2 Gen 2x2 Type-C ports at 20 Gbps with DisplayPort Alt Mode, one 10 Gigabit Ethernet jack, and one HDMI 2.1 port. A fourth USB-C port is dedicated to the 240W USB-PD power adapter. A similarly priced Mac Studio provides four 80 Gbps Thunderbolt 5 ports, two more USB-C ports, an SD card reader and a headphone jack. The EdgeMesa has no audio jack at all.
Can it game?
Results are mixed. The RTX Spark scored roughly 56% higher than the Radeon 8060S in 3DMark Steel Nomad. In Control Resonant at 1440p High with Performance upscaling, the two platforms landed close together, but the EdgeMesa posted considerably worse 1% low framerates, producing visible stutter.
The 273 GB/s of shared LPDDR5X bandwidth constrains the GPU; an RTX 5070 enjoys 672 GB/s of dedicated GDDR7 bandwidth. Prism x86 emulation overhead also limits GPU utilization in most games, since few major titles have Arm ports. Enabling ray tracing at Medium cost the Spark 14% of performance versus 38% on the AI Halo, a result the review credits partly to DLSS Ray Reconstruction. With MFG Preset C, average latency fell to about 84 ms with a 145 FPS average.
Is it quiet?
Power draw measured about 140W from the wall during LLM inference and about 200W under gaming loads. Noise reached 38.2 dBA at 46 cm during inference and 40.7 dBA while gaming. The Ryzen AI Halo emitted 46.9 dBA while pulling about 209W in the same game. The review singled out MSI's dual-fan, vapor-chamber cooling design for high praise.
The review closes with a cautious verdict: buyers uncertain about local AI under Windows should wait for prices to fall before spending thousands on a still-fuzzy vision, while Windows-native developers with cash in hand get the strongest Microsoft-powered foundation for agentic AI work available today.
bestbuy.7tiv.net (Original)
More from Grace Kim
Show full bio
Market editor covering people, launches and day-to-day practice in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.
1,213 articles