New Strix Halo Mini-PC Arrives with 96GB Memory, Targeting Sub-$1500 LLM Builds

The stream of mini-PCs built around AMD’s Ryzen AI 300 “Strix Halo” platform continues, this time with a new model named the X+ RIVAL. While the market is quickly becoming crowded with similar hardware, this particular entry is notable for its specific memory configuration and pricing, creating a new value-oriented option for local LLM enthusiasts. This system, along with a handful of others, is beginning to define a new mid-range tier for compact, high-memory inference machines.

Core Hardware and System Configuration

The X+ RIVAL is built around the flagship Ryzen AI MAX+ 395, which features 16 Zen 5 CPU cores and the integrated Radeon 8060S graphics with 40 RDNA 3.5 compute units. The chassis itself appears to be a familiar OEM design, also used by brands like Bosgame and Peladn, indicating a focus on component configuration rather than custom enclosure engineering. It includes a dual-fan cooler with a copper block and three heatpipes to manage the processor thermal output.

The most significant specification is its 96GB of soldered LPDDR5X-8000 memory. Unlike other manufacturers that have focused on 64GB or 128GB configurations, the X+ RIVAL, similar to the BOSGAME M5, targets a specific capacity that sits between these two points. For storage, the system includes two M.2 2280 PCIe 4.0 slots and is offered with a 2TB NVMe drive by default. The availability of Ubuntu as a pre-installed option is a practical consideration for users who plan to dedicate the machine to LLM workloads.

Analyzing the 96GB Memory Configuration for LLMs

For a local LLM builder, the 96GB unified memory pool is the main point of interest. After accounting for operating system overhead, users can expect to allocate approximately 80GB of this memory as VRAM for the integrated GPU. This capacity is substantial and opens the door to running a wide range of large models that are inaccessible on typical consumer hardware.

With around 80GB of VRAM, this mini-PC can comfortably load 4-bit quantized versions of powerful models such as the 70-billion parameter Llama, GLM-4.5-Air, and the Kimi Dev 72B. It should also be capable of handling certain 106B parameter models and the recently released 120B gpt-oss model. The gpt-oss is particularly interesting as it is a Mixture-of-Experts (MoE) model released with native quantization, which could lead to very efficient prompt processing and token generation speeds even on Strix Halo’s 256 GB/s memory bandwidth. However, models requiring more than 80GB of VRAM, such as the Qwen3-235B-A22B or DeepSeek’s larger variants, will remain out of reach for this configuration.

Market Context and Pricing Analysis

The introduction of 96GB models from X+ and Bosgame at a sub-$1500 price point disrupts the initial pricing structure we have seen for Strix Halo systems. Until now, the market has been segmented into high-end 128GB models like the PELADN YO1, often priced near $2000, and more moderate 64GB versions like the GMKTEC EVO-X2, which sell for around $1600. This new 96GB tier provides a middle ground, offering a significant memory upgrade over 64GB without the steep price increase of the 128GB options.

This makes the 96GB configuration a compelling choice for price-conscious builders who need to run 70B-class models efficiently. It is important to note that the absolute cheapest entry into the Strix Halo ecosystem is expected to be an unconfirmed laptop with just 32GB of unified memory for under $1000, but for serious LLM work, 96GB represents a more practical and capable starting point. Even with these new, more accessible price points, it is arguable that the entire Strix Halo product stack is still priced with a premium, likely driven by the ongoing AI hardware hype.

Trending Stories