Infinite Supercomputer Power
On your desk
Run powerful AI models, agents and private workflows locally, with the flexibility of Windows or Linux.
EMARQUE AI Eclipse brings large-model local AI performance to your desk with AMD Ryzen™ AI Max+ 395 and 128GB unified memory. Built for serious inference, agents and AI workloads without the complexity of traditional AI infrastructure.
Available with Windows or Linux, Eclipse fits into the way you already work. EMARQUE AI Utility helps you manage local models, load or release them when needed, and connect them with tools such as Hermes Agent, Onyx, APIs and existing business applications.
Start with EMARQUE AI Utility to run and manage local AI on Eclipse. Use Hermes Agent for personal AI that can work with files, tools and multi-step tasks. You can also run platforms such as Onyx, n8n, coding assistants, RAG systems and other self-hosted AI applications.
Turn local AI into a shared service for your team and applications. Deploy internal assistants, inference services, knowledge systems and agent workflows across your network. Keep suitable workloads local and connect to cloud AI when you need more capability or scale.
Connect local AI to the tools your business already uses. Use APIs, MCP and supported integration to link AI agents with CRM, accounting, ERP, support platforms, databases and internal applications. Build useful workflows around your existing systems instead of replacing them.
Go deeper with AMD ROCm, PyTorch and the growing open-source AI ecosystem. Build and optimize your own workloads with tools such as llama.cpp, Lemonade, vLLM, Ollama and ComfyUI. From local inference and model serving to fine-tuning and custom AI applications, EMARQUE AI Eclipse gives developers direct access to the platform underneath.
EMARQUE.AI Utility
EMARQUE AI Utility brings AI models and tools together in one familiar interface, preconfigured on the AI hardware you own. Add open models in one click, use chat, documents, image and video right away, and share it across your network, all on-device.
Preconfigured. No subscriptions. Runs on your own machine.
Supercharge your edge deployments with EMARQUE AI Eclipse. Designed to run massive LLMs up to 120B with ease, while delivering maximum efficiency where it matters most.
*Numbers are recorded from internal testing and may not reflect actual use cases.
Average Token Speed
Harness elite neural processing performance and full x86 compatibility with EMARQUE AI Eclipse. Built to run next-gen AI models across both Windows and Linux operating systems with uncompromised efficiency and flexibility.
AMD Ryzen™ AI Max+ 395
Central Processing Unit
Radeon™ 8060S
Graphics Processing Unit
128GB 8000MT/s
LPDDR5X Unified Memory
Engineered around the AMD Ryzen™ AI Max+ 395 CPU with Radeon™ 8060S Graphics, EMARQUE AI Eclipse combines a 16-core Zen 5 CPU and 40 RDNA 3.5 compute units in a single, power-efficient architecture. Delivering desktop-class compute and integrated graphics in a compact platform, built for local AI, development, creative workloads and everyday computing without requiring a discrete GPU.
With its unified memory architecture, Eclipse gives the CPU and GPU access to a large pool of high-speed LPDDR5X 8000MT/s memory, helping larger AI models run locally without the limits of dedicated VRAM.
EMARQUE AI Utility simplifies on-device model management, letting you download, organize, load and release models as needed from one place.
Pre-order the EMARQUE AI Eclipse Supercomputer today to lock in exclusive pricing. Quantities for the initial production run are strictly limited, so reserve yours now.
Be the first in line
From MYR18,999
1TB configuration, Windows or Linux at no extra cost. 2TB and 4TB quoted on enquiry.
The EMARQUE AI Eclipse is a compact 2L local AI workstation built around the AMD Ryzen™ AI Max+ 395 processor, formerly known by its codename Strix Halo.
Unlike a conventional desktop with separate system RAM and GPU VRAM, Ryzen AI Max uses a large pool of high-speed unified memory shared between its 16-core Zen 5 CPU and Radeon™ 8060S graphics. This makes the Eclipse particularly well suited to running large language models, AI agents and other memory-intensive AI workloads locally.
The Eclipse is available with either Windows 11 or Linux, depending on how you intend to use the system.
For local AI, having enough memory is often just as important as raw GPU performance.
The Ryzen AI Max+ 395 supports up to 128GB of LPDDR5X-8000 unified memory, with bandwidth of up to 256GB/s. Because the Radeon 8060S shares this memory with the CPU, much more memory can be made available to AI workloads than on a typical consumer graphics card with 12GB, 16GB or 24GB of dedicated VRAM.
AMD allows up to 96GB to be configured as graphics memory on a 128GB system when required for larger models.
This is one of the main reasons the Eclipse can run models that simply will not fit into the VRAM of many conventional desktop GPUs.
It depends on the model architecture, quantisation, context length and AI runtime being used.
AMD officially positions the Ryzen AI Halo platform for running models of up to approximately 200 billion parameters locally with suitable model formats and memory configurations.
In practice, the Eclipse is particularly comfortable with models in the 7B to 70B range, while 100B+ models and large Mixture-of-Experts models can also be practical when properly quantised.
Ryzen AI Max+ 395 systems have demonstrated models such as Llama 4 Scout 109B, GPT-OSS 120B and other 100B+ class models running locally.
Parameter count alone does not tell you how fast a model will run. A 120B Mixture-of-Experts model may actually run considerably faster than a dense 70B model because only part of the model is active for each generated token.
The Eclipse is not limited to running a chatbot.
You can use it for local LLM inference, coding assistants, AI agents, document analysis, private knowledge bases, retrieval-augmented generation (RAG), image generation, computer vision, speech and transcription, development environments and custom AI applications.
It can also be used as the AI backend for other software. For example, a locally hosted model can be exposed through an API and connected to internal applications, automation platforms or other computers on your network.
Because the model can run directly on the Eclipse, sensitive documents and prompts can remain on your own hardware instead of automatically being sent to a cloud AI provider.
Whether a workflow is completely offline still depends on the software you choose to run. Some applications may use external services even when the AI model itself is local.
Choose Windows if you want the Eclipse to function as both a normal high-performance PC and a local AI workstation.
Windows gives you the familiar Windows environment, creator applications, gaming, productivity software and straightforward local-AI tools such as LM Studio, Ollama and llama.cpp. It is a good choice for users who want to experiment with local AI without turning the system into a dedicated Linux workstation.
Choose Linux if AI development is the primary purpose of the machine.
Linux currently provides the broader AMD ROCm environment and is generally the better platform for PyTorch development, containers, server deployments, automation and many open-source AI frameworks.
AMD has significantly expanded ROCm support on Windows, but the complete ROCm software stack available on Linux is not yet available on Windows.
The software ecosystem around AMD Strix Halo has matured considerably.
AMD officially lists support and workflows for tools including llama.cpp, Ollama, LM Studio, vLLM, PyTorch and ComfyUI, alongside AMD ROCm.
This means the Eclipse can be used for everything from downloading and chatting with a local model through a graphical interface to developing and serving your own AI applications.
For users who simply want to run local models, llama.cpp, LM Studio and Ollama are among the easiest places to start. Developers running more advanced ROCm, PyTorch or server workloads will generally benefit from the Linux version.
Software compatibility continues to evolve quickly, so the ideal runtime can vary depending on the model and workload.
Yes.
The Eclipse does not need to be used only as a desktop computer. Local inference engines such as llama.cpp and LM Studio can expose models through an OpenAI-compatible API, allowing applications and other computers on your network to access the model running on the Eclipse.
This makes it possible to build private AI assistants, internal knowledge systems, development tools and business applications around one locally hosted model.
This platform has also been demonstrated in multi-system AI deployments, where several Ryzen AI Max machines are combined for workloads that exceed the capabilities of a single system.
For larger production environments, the appropriate architecture will depend on the model, expected number of simultaneous users and required response speed.
Not for every workload, and that is an important distinction.
The main advantage of the Ryzen AI Max+ 395 is its combination of large unified memory, compact size, x86 compatibility and relatively low power requirements. It allows surprisingly large models to run on a small desktop system without requiring several expensive high-memory GPUs.
A dedicated NVIDIA GPU or data-centre AI server is still the better choice when you need very high inference throughput, large-scale model training, heavy multi-user concurrency or software that specifically depends on CUDA.
The Eclipse makes the most sense when your priority is having a capable local AI workstation with enough memory to experiment with large models, develop AI applications, run private workloads and use the same machine for normal computing.
It is about bringing useful AI compute onto your desk, not replacing a data centre.
