A 128 GB unified-memory mini PC that runs massive local AI models at a fraction of enterprise hardware pricing
The AMD Ryzen AI Halo Developer Platform is tailored for AI researchers and engineers who need to run expansive language models locally without cloud subscriptions. Its standout capability is 128 GB of unified LPDDR5X memory, enabling the system to host models that crash high-end consumer GPUs. The trade-off is modest raw processing throughput; desktop graphics cards like the GeForce RTX 5090 process tokens far faster whenever models fit inside 32 GB. Compared to the $7,000 NVIDIA DGX Spark, the $4,000 AMD system delivers a much more accessible entry into local high-capacity AI.
Buy it if
- Machine learning developers running large language models up to 70B parameters entirely offline
- Engineers needing a quiet, 120 W compact desktop appliance for multi-agent local workflows
- Researchers who require more than 32 GB of VRAM without purchasing enterprise server hardware
Skip it if
- Gamers seeking pure graphics performance where dedicated discrete GPUs offer better value
- Developers running models small enough to fit comfortably inside a 24 GB or 32 GB discrete GPU
- Shoppers on a strict budget who can purchase third-party 395 mini-PCs for roughly half the price
Where to buy
We may earn a commission from purchases or qualifying sign-ups through our links. Affiliate links explained
What reviewers liked
- 128 GB of unified LPDDR5X-8000 memory allows running large AI models locally
- Compact 150 mm square chassis weighing just 1,016 grams with integrated 10 GbE
- Multi-core CPU performance rivals Intel Core i7-14700K at only 120 W peak board draw
- Substantially lower price than enterprise workstations like the $7,000 NVIDIA DGX Spark
What they didn’t
- Single-core CPU performance lags modern desktop chips in Cinebench 2024
- Official $4,000 developer price tag carries a heavy premium over third-party 395 systems
- PCIe Gen 5 storage operates capped at Gen 4 speeds due to APU bus limitations
Key specs
- Launch price
- $4,000
- Reported price
- $4,000 (developer platform); $1,999 to $2,399 (third-party systems)
- Processor
- AMD Ryzen AI Max+ 395 (16 cores, 32 threads, up to 5.1 GHz)
- Integrated Graphics
- AMD Radeon 8060S (40 RDNA 3.5 CUs, 2560 stream processors)
- Unified Memory
- 128 GB LPDDR5X-8000 (soldered, 256-bit bus)
- Storage
- 2 TB Micron 4600 NVMe SSD (PCIe Gen 4 mode)
- TDP / Power Consumption
- 120 W TDP (approx. 10 W idle, 120 W peak)
- Networking
- 10 GbE RJ-45, Wi-Fi 7, Bluetooth 5.4
- Dimensions
- 150 x 150 x 45.4 mm (5.9 x 5.9 x 1.79 in)
- Weight
- 1,016 g (2.24 lb)
- Launch Price
- $4,000
In depth
Productivity and multi-core performance
TechPowerUp tested the 16-core, 32-thread Zen 5 CPU in Cinebench 2024, recording a multi-threaded score of 1,839 points. That output surpasses eight-core desktop chips like the Ryzen 7 9800X3D (1,378 points) and lands right alongside the desktop Intel Core i7-14700K, while reaching roughly 80% of a full 170 W desktop Ryzen 9 9950X (2,285 points). However, TechPowerUp noted that single-threaded performance is more restrained: it posted 116 points, trailing standard desktop Zen 5 processors that range between 134 and 148 points due to the APU's lower sustained single-core clock frequencies under its mobile-derived power profile.
Gaming performance
The integrated Radeon 8060S GPU features 40 RDNA 3.5 compute units and 2,560 shaders. In TechPowerUp's testing, the graphics processor performed roughly in line with an entry-level desktop discrete GeForce RTX 4060, making it capable of solid 1080p and moderate 1440p gaming. TechPowerUp also experimented with the platform's Variable Graphics Memory allocation settings, finding that altering the memory carve-out between system RAM and video memory produced no measurable change in gaming frame rates. While competent for an integrated chip, reviewers emphasize that gaming is not the primary purpose of this machine.
Power consumption and cooling needs
Designed around a compact 150 mm square form factor, the system draws modest power compared to standard desktop towers. TechPowerUp recorded an idle system draw of approximately 10 W, which climbed to roughly 120 W under peak GPU inference and full computing workloads. The integrated thermal solution maintained stable operation within its 1,016 g chassis without extreme thermal throttling, providing a workstation-class computing environment that operates quietly on a standard desk without the cooling overhead of multi-slot discrete graphics cards.
Platform: socket, motherboard and memory costs
Unlike standard AM5 desktop processors that slot into user-upgradable motherboards, the Ryzen AI Halo Developer Platform is a fully integrated system-on-chip device. The 128 GB pool of LPDDR5X-8000 memory is soldered directly to the board along a wide 256-bit bus, delivering approximately 256 GB/s of bandwidth. This design eliminates aftermarket RAM purchases and ensures high memory bandwidth, but it removes socket upgradability entirely. TechPowerUp also observed that while the system ships with a fast Micron 4600 SSD, it runs at PCIe Gen 4 speeds because the APU platform does not support PCIe Gen 5 connectivity.
Upgrade value versus previous generation
Compared to standard consumer APUs and earlier mobile silicon, the Strix Halo architecture introduces a jump in graphics compute units and unified memory capacity. TechPowerUp noted that AMD has already outlined its next-generation Gorgon Halo (Ryzen AI Max+ PRO 495), which supports up to 192 GB of LPDDR5X-8533 memory with a 100 MHz clock speed bump on the same underlying silicon. For users who already own a 128 GB Strix Halo device, the upcoming refresh offers only incremental processing gains, but for owners of standard desktop systems, the primary value is moving from a memory-constrained 16 GB or 32 GB GPU to a 128 GB unified pool.
Price
AMD set the launch price for this official Developer Platform package at $4,000, positioning it directly against workstation hardware. TechPowerUp and ServeTheHome highlighted that while $4,000 is steep for a mini PC, it undercuts NVIDIA's comparable 128 GB DGX Spark ($7,000) by $3,000. For buyers who do not need AMD's official developer bundle, third-party mini PCs packing the exact same Ryzen AI Max+ 395 silicon and 128 GB of RAM—such as the GMKtec EVO-X2 and Framework Desktop—typically retail between $1,999 and $2,399, offering a much more cost-effective avenue to the same computing performance.
Is it worth the money?
At $4,000, AMD's turnkey developer unit is $3,000 cheaper than an NVIDIA DGX Spark, but third-party mini-PCs using the identical 395 processor offer equivalent hardware for around $2,000.
Also consider
- NVIDIA DGX Spark (128 GB)
A $7,000 AI appliance offering enterprise-grade networking and native CUDA optimization for commercial teams.
- GMKtec EVO-X2
Uses the identical Ryzen AI Max+ 395 APU and 128 GB RAM configuration for roughly $1,999.
- Framework Desktop (Ryzen AI Max+ 395)
A modular mini-workstation option offering modularity and 128 GB unified memory for about $2,198.
Reviews we read
This roundup summarizes professional reviews and published specifications; Next Interrupt did not test the product. Roundup updated October 20, 2026. How we work
- TechPowerUp — AMD Ryzen AI Halo Review - Local-AI Powerhouse with 128 GB of Memory
- ServeTheHome — AMD Ryzen AI Halo Developer System Review AMD Goes for Local AI
