ASUS unveiled the ASUS Ascent GX10, a groundbreaking AI supercomputer powered by the cutting-edge NVIDIA GB10 Grace Blackwell Superchip. This revolutionary device puts the massive power of a petaflops-scale AI supercomputer directly into the hands of developers, AI researchers and data scientists around the world.
As generative AI models grow in size and complexity, local development efforts are becoming increasingly challenging. Prototyping, tuning, and inference of large models requires massive memory and compute performance. To address these needs, the Ascent GX10 is designed to provide developers with a powerful and efficient desktop solution.
Unparalleled AI Performance
ASUS Ascent GX10 delivers up to 1,000 TOPS of AI performance to power large-scale AI workloads, and with up to 128GB of coherent unified system memory, developers can experiment, fine-tune and infer the latest generation of inference AI models with up to 200 billion parameters.
こちらもお読みください： アルト、ハイレゾと共同で次世代AIデータセンター事業「alt GPU Cloud」を開始
Featuring NVIDIA Grace Blackwell Technology
Inside the new ASUS Ascent GX10 is the advanced NVIDIA GB10 Superchip, powered by the Grace Blackwell architecture and optimized for a compact form factor. This incredibly powerful GB10 Superchip includes a powerful Blackwell GPU with 5th generation Tensor Cores and FP4 support, delivering up to 1,000 TOPS of AI processing power. Additionally, it features a high-performance Grace 20-core Arm CPU to enhance data pre-processing and orchestration, accelerating model tuning and real-time inference. The GB10 Superchip employs NVIDIA NVLink™-C2C , providing a unified CPU+GPU memory model with 5x the bandwidth of PCIe® 5.0.
Handling cutting-edge large-scale parameter-generating AI models
With 128GB of integrated system memory and support for the FP4 data format, ASUS Ascent GX10 can manage AI models with up to 200 billion parameters. This enables AI developers to prototype, fine-tune, and infer state-of-the-art AI models with up to 70 billion parameters right on their desktop. Thanks to integrated NVIDIA ConnectX® network interface cards (NICs) , two GX10 systems can be linked together to handle even larger models, such as Llama 3.1’s 405 billion parameters.
Local development, scalable deployment
ASUS Ascent GX10 provides a powerful and economical experimentation platform for developers to prototype models and AI applications, freeing up critical compute resources in the cluster that are better suited for training and deploying production models. With NVIDIA AI software , Spark Core GX10 users can seamlessly move models from their desktop environments to NVIDIA DGX Cloud or other accelerated cloud or data center infrastructures with little or no code adjustments, easing the prototyping, fine-tuning, and iterative process.
ソース PRタイムズ
