Overview
As AI models continue to grow in complexity, developers need local compute resources that can keep up with large-scale model development without relying solely on cloud infrastructure. The AceleMax® AXG-AB10 AI Supercomputer, built on the NVIDIA Grace Blackwell Superchip, brings data center-class performance to the desktop.
With up to 1,000 AI TOPS of compute and 128GB of unified memory, the AXG-AB10 enables fast iteration, efficient fine tuning, and local inferencing. Whether used as a standalone development station or as part of a broader AI stack, it delivers the performance and flexibility needed to accelerate AI workloads at every stage.
Benefits of Desktop AI with the AXG-AB10
Organizations and research teams are rethinking their approach to AI development. While the cloud offers flexibility, local systems like the AXG-AB10 provide cost predictability, performance control, and secure access to powerful compute resources.
- Faster model iteration
Develop, fine tune, and run inference without waiting on shared cluster resources. - Support for large models
Handles AI models with over 200 billion parameters and scales further when linked to a second system. - Local control and data privacy
Keep sensitive data on-site and avoid the compliance risks of third-party cloud storage. - Seamless software compatibility
Built to support the NVIDIA AI Enterprise stack for easy deployment across environments. - Scalable desktop compute
Add capacity by linking two systems via ConnectX-7 for expanded model support.
Key Features
- Up to 1,000 AI TOPS compute performance
- 128GB of unified CPU-GPU memory
- NVIDIA Grace Blackwell Superchip with Arm CPU and Blackwell GPU
- NVLink-C2C interconnect with 5x the bandwidth of PCIe 5.0
- Supports AI models with over 200 billion parameters
- ConnectX-7 NIC for system expansion
- NVIDIA AI Enterprise software compatibility
System Specifications
| Component | Specification |
|---|---|
| Processor | 20-core NVIDIA Grace Arm CPU |
| GPU | Blackwell GPU with FP4 and Tensor Cores |
| Memory | 128GB Unified CPU-GPU Memory |
| Performance | Up to 1,000 AI TOPS |
| Interconnect | NVLink-C2C (5x PCIe 5.0 Bandwidth) |
| Networking | NVIDIA ConnectX-7 |
| Model Support | 200B+ parameter models |
| Software Stack | NVIDIA AI Enterprise compatible |
Supports Every Stage of the AI Workflow
From prototyping and testing to fine tuning and inference, the AXG-AB10 gives teams the power to iterate on large models without external dependencies. It’s a practical solution for AI labs, R&D teams, and developers working on LLMs, computer vision, or simulation workloads.
Advanced Unified Architecture
The AXG-AB10 combines a fifth-generation Tensor Core GPU with FP4 support and a 20-core Arm CPU, optimized for AI model development and real-time inferencing. NVLink-C2C delivers unified memory across CPU and GPU with significantly higher bandwidth than PCIe 5.0, reducing bottlenecks and enabling a more efficient development process.
AI Solutions by AMAX
AMAX offers the AXG-AB10 as part of our portfolio of AI systems designed for accelerated computing. Our team supports every step of the deployment process—from selecting the right platform to ensuring it’s properly configured for your workload, power, and thermal requirements. Whether it's for a single developer or part of a broader research effort, AMAX delivers systems ready to work.