ASUS Ascent GX10 Personal AI Workstation, 2000 B Parameter Fine-Tune Ready, Whisper-Quiet 1.1 L Mini Tower
Introduction
Meet the ASUS Ascent GX10—the world’s first 1-PFLOPS personal AI supercomputer that fits next to your monitor. Powered by NVIDIA’s GB10 Grace-Blackwell Superchip, it fuses a 20-core Arm Grace CPU and a Blackwell GPU with 128 GB of unified LPDDR5X memory, giving developers, data scientists and creators enough horsepower to fine-tune 200-billion-parameter models without sending data to the cloud. A whisper-quiet 1.1-liter chassis, tool-less dual M.2 bays and 200 Gb/s networking make GX10 equally at home on a desk or in a rack. Whether you are prototyping generative-AI applications, building vision-language robots or running RAG pipelines for your enterprise, GX10 delivers workstation-class performance with server-grade scalability.
Features
[1 PFLOPS AI Power] 1.1-liter desktop delivers 1 PFLOPS (FP8 sparse) Grace-Blackwell compute—train & infer 200-billion-parameter LLMs on your desk.[128 GB Unified Memory] CPU + GPU share 128 GB LPDDR5X via 900 GB/s NVLink-C2C—no PCIe bottleneck, instant tensor swaps.[Dual-Box Cluster Ready] Snap two GX10 together with the built-in 200 Gb/s ConnectX-7 for 2 PFLOPS & 256 GB pooled memory.[Whisper-Quiet Mini Tower] Custom vapor-chamber + dual 140 mm fans keep temps under 75 °C at only 32 dB—perfect for open-office or home labs.[Zero-Code Software Stack] Pre-loaded NVIDIA AI Workbench, CUDA, cuDNN, TensorRT, NeMo—develop locally, deploy to DGX Cloud in one click.[Enterprise-Secure On-Prem] Keep sensitive data in-house; optional 4 TB PCIe 5.0 SSD and TPM 2.0 module for compliance-ready AI workflows.
Shipping List
ASUS Ascent GX10 standard package
Applications
Local LLM fine-tuning & 200 B-parameter inference without cloud costs
Generative-AI copilots, chatbots and RAG pipelines for SMBs
Vision-language-model (VLM) prototyping for robotics & edge devices
Secure on-prem code generation, document AI and compliance-sensitive analytics
Real-time recommendation engines, NLP services and multimodal search
University labs: AI courses, student projects and publishable research
scalable render/simulation nodes—link two GX10 for 2-PFLOPS cluster on demand
Specification
CPU | 20-core NVIDIA Grace (Arm v9.2)
GPU | Blackwell GPU, 5th-gen Tensor Cores
AI Performance | 1 PFLOPS (FP8 sparse) | 1000 TOPS (FP4)
Memory | 128 GB LPDDR5X unified (CPU+GPU)
Storage | 1 TB NVMe PCIe 4.0 (M.2 2242); optional 2 TB / 4 TB PCIe 5.0
I/O | 2× USB-C 3.2 Gen2, 3× USB-A 3.2 Gen2, 1× HDMI 2.1, 1× 200 Gb/s ConnectX-7 Ethernet, Wi-Fi 6E, Bluetooth 5.3
Power | 330 W external adapter, 100-240 V
Dimensions | 150 × 150 × 51 mm (1.1 L)
Weight | 1.5 kg
