EMARQUE AI Eclipse AMD Ryzen AI Max+ 395 Local AI Workstation
Pre-order. This price is the 1TB configuration, with Windows 11 or Linux at no extra cost. 2TB and 4TB are quoted on enquiry. Quantities for the initial production run are limited.
EMARQUE AI Eclipse: 128GB of unified memory for local AI, built in Malaysia
The EMARQUE AI Eclipse brings large-memory local AI computing into a compact chassis. It runs the AMD Ryzen™ AI Max+ 395 with 16 Zen 5 CPU cores, Radeon™ 8060S graphics and 128GB of LPDDR5X unified memory, for running large language models, AI agents, development workloads and private AI applications on hardware you own. Available in Windows and Linux configurations.
Key features
Technical specifications
Available ports are based on the EMARQUE AXB35-02-H05-YC configuration.
Why 128GB of unified memory matters
A conventional desktop separates system RAM from dedicated GPU VRAM. The Ryzen AI Max+ 395 instead uses high-speed unified memory addressable across the processor and the integrated Radeon GPU. With 128GB installed, up to 96GB can be allocated to graphics workloads. For local AI, that makes it possible to work with models that would exceed the VRAM of a 16GB or 24GB consumer graphics card.
Local AI model capability
AMD states that this platform, with 128GB of unified memory, holds models of up to 200 billion parameters in device memory. In practice the useful classes are:
Actual performance depends on model architecture, quantisation, context length, runtime, memory allocation, operating system and active parameter count. A model with more parameters is not necessarily slower than a smaller one, particularly with Mixture-of-Experts architectures.
What you can run
Local large language models, coding assistants, AI coding agents, AI agents, private document analysis, Retrieval-Augmented Generation, local knowledge bases, OpenAI-compatible local APIs, image generation, computer vision, speech and transcription, AI development, Python and PyTorch development, local inference servers, business AI applications and edge AI workloads.
Supported AI ecosystem
Depending on the operating system and workload, Eclipse works with llama.cpp, Ollama, LM Studio, Lemonade, AMD ROCm, HIP, PyTorch, ComfyUI and OpenAI-compatible local APIs. Software compatibility and performance vary by operating system, runtime, model format and software version.
Windows or Linux
Windows suits users who want local AI without leaving a familiar desktop: AI assistants, coding and development, creator and productivity applications, gaming, and Windows-compatible AI tools such as LM Studio, Ollama and llama.cpp. The Windows configuration works as both an everyday PC and a local AI workstation.
Linux suits developers and AI practitioners building dedicated systems: AMD ROCm development, PyTorch, inference, containers, automation, local AI servers, API deployments, RAG systems and backend workloads. Linux currently gives access to the broader AMD ROCm ecosystem and is generally the better choice for advanced AI development and deployment.
Use it as a local AI server
Eclipse can operate as an inference server on your own network. Compatible inference engines expose models over REST or OpenAI-compatible APIs, so other applications and devices can use models running on the Eclipse: internal AI assistants, company knowledge bases, coding assistants, AI automation, document processing, chatbot interfaces and private API endpoints. The dual 10GbE interfaces let it sit inside high-speed wired and storage networks.
How Eclipse relates to the AMD Ryzen AI Halo
AMD sells its own developer box on this platform, the AMD Ryzen AI Halo. Eclipse runs the same Ryzen AI Max+ 395 silicon with the same 128GB of LPDDR5X-8000 unified memory and the same 256GB/s of bandwidth, so it carries the same 200 billion parameter platform capability AMD states. Eclipse differs in being configured, built and supported in Malaysia by EMARQUE, with dual 10GbE networking and a choice of Windows or Linux at handover.
Part of the AMD Ryzen AI Max range, the Windows-native x86 route to on-device AI. If you need a CUDA-first Linux platform instead, EMARQUE also supplies NVIDIA DGX Spark.
Important information
AI model compatibility and performance vary with model architecture, format, quantisation, runtime, context length and software configuration. References to maximum model parameter sizes describe memory and platform capability and are not a guarantee of inference performance for every model. Some AI software requires additional installation, configuration or dependencies. ROCm, AI runtime and application support differ between Windows and Linux. EMARQUE AI Eclipse ships without Wi-Fi or Bluetooth; network connectivity is provided by its dual 10GbE Ethernet interfaces.
Configuration and pricing confirmed on enquiry. Built and supported in Malaysia by EMARQUE.
Couldn't load pickup availability
Inside the AMD Ryzen AI Max platform
Up to 50 TOPS
AMD XDNA 2 NPU for on-device AI acceleration.
128GB unified
LPDDR5X memory shared by the CPU and Radeon GPU.
Up to 96GB to GPU
Variable Graphics Memory for running local LLMs.
Ryzen AI Max+ 395
16-core Zen 5 CPU with Radeon 8060S graphics.
FAQ
Frequently asked questions
What this AMD Ryzen AI Max system is, what it can do locally, and how it compares to NVIDIA's personal AI computers.
-
What is the EMARQUE AI Eclipse?
Eclipse is EMARQUE's own local AI workstation, configured, built and supported in Malaysia. It pairs the AMD Ryzen AI Max+ 395 with 128GB of LPDDR5X-8000 unified memory, Radeon 8060S graphics with 40 RDNA 3.5 compute units, and an XDNA 2 NPU rated up to 50 TOPS. Up to 96GB of the memory can be assigned to the GPU, so large language models run on the device rather than in the cloud. It is available in Windows and Linux configurations.
-
How does Eclipse compare to the AMD Ryzen AI Halo?
AMD sells its own developer box on this platform, the AMD Ryzen AI Halo. Eclipse uses the same Ryzen AI Max+ 395 silicon, the same 128GB of LPDDR5X-8000 unified memory and the same 256GB/s of memory bandwidth, so it carries the same 200 billion parameter platform capability AMD states.
The differences are in the build and the support. Eclipse is configured, assembled and validated by EMARQUE in Malaysia, ships with dual 10GbE networking, and you choose Windows or Linux at handover. Warranty and support are handled locally rather than through an overseas channel.
-
What size AI models can it actually run?
AMD states this platform holds models of up to 200 billion parameters in device memory with 128GB installed. In practice, 7B models are fast for local assistants and coding, 14B to 32B balance capability against speed, 70B-class models suit higher-quality reasoning, and selected quantised or Mixture-of-Experts models above 100B are workable depending on architecture, quantisation, runtime and context length.
Parameter count alone does not determine speed. A Mixture-of-Experts model with more total parameters can run faster than a smaller dense model, because only a fraction of its parameters are active per token.
-
Should I choose the Windows or the Linux configuration?
Choose Windows if you want local AI without leaving a familiar desktop. It runs LM Studio, Ollama and llama.cpp alongside your normal productivity, creator and gaming software, so the machine works as an everyday PC as well as an AI workstation.
Choose Linux if you are building a dedicated AI system. It gives access to the broader AMD ROCm ecosystem, and suits PyTorch, containers, inference servers, API deployments, RAG systems and backend automation. For advanced AI development and deployment, Linux is generally the better environment today.
-
Why does it have no Wi-Fi, and what is the dual 10GbE for?
Eclipse is configured as a wired LAN-only workstation by design. Its two 10 Gigabit Ethernet ports are there for serving models across your network, connecting to NAS and storage at speed, and linking multiple systems, which is where a machine like this usually sits. Wired networking is more predictable under sustained load than consumer wireless.
If your deployment needs wireless, tell the EMARQUE AI team when you enquire and we will advise on the options.
-
Eclipse or NVIDIA DGX Spark: which should I choose?
Both are compact 128GB personal AI computers, and EMARQUE supplies both. Eclipse runs full x86 Windows or Linux, so your existing Windows software and tooling work alongside local AI, and it offers dual 10GbE networking. NVIDIA DGX Spark runs Linux-based DGX OS on the CUDA ecosystem, which is deeper and faster for AI inference work.
Independent testing has DGX Spark ahead of this AMD platform on most vLLM inference workloads, while the AMD platform leads on CPU-bound tasks. Choose Eclipse for x86 Windows compatibility, dual-OS flexibility and networking; choose DGX Spark for CUDA-first inference throughput. Talk to the EMARQUE AI team and we will scope which fits your workload.
-
How do I order, and how is it configured?
Eclipse is supplied to order. The EMARQUE AI team specs the storage and operating system around your workload and target models, validates the build, and confirms configuration and pricing on enquiry. WhatsApp 012 627 2280 or email business@emarque.co and we will scope the right build.
-
Can it really run large AI models locally?
Yes. AMD Ryzen AI Max+ 395 pairs up to 128GB of unified LPDDR5X memory with a 50 TOPS XDNA 2 NPU and Radeon 8060S graphics. Up to 96GB of that memory can be assigned to the GPU, so large language models run on the machine itself under Windows using tools such as LM Studio and Ollama, without sending data to the cloud. Exact model sizes and speeds depend on the model and quantization.
-
How do I order and get support from EMARQUE?
Pricing and configuration are confirmed on enquiry. EMARQUE supplies and supports AMD Ryzen AI Max systems in Malaysia, and can spec the memory, storage and operating system for your workload. WhatsApp 012 627 2280 or email business@emarque.co.