.avif)
The NVIDIA HGX Rubin platform is the successor to Blackwell, built on an entirely new GPU architecture paired with the NVIDIA Vera CPU. It is designed for the scale and complexity of next-generation AI workloads, with expected improvements in compute throughput, memory bandwidth, and system-level efficiency.
.avif)
Arc Compute is preparing to offer HGX Rubin platforms from leading OEM partners. System configurations and availability will be announced as Rubin hardware becomes commercially available.
The HGX Rubin platform introduces a new GPU architecture paired with the NVIDIA Vera CPU, building on the foundation established by previous HGX generations. It is designed to deliver significant gains in compute density, interconnect bandwidth, and energy efficiency for large-scale AI infrastructure.
Next-generation NVIDIA GPU architecture with upto 288 GB HBM4 per GPU, up to 22 TB/s memory bandwidth, and hardware-accelerated Transformer Engine with NVFP4 support. Up to 50 PFLOPS per GPU.
Sixth-generation NVLink delivering 28.8 TB/stotal GPU-to-GPU bandwidth across all 8 GPUs. Enables the entire server tooperate as a tightly coupled compute domain.
Flexible CPU pairing: NVIDIA Vera CPUs (88 ArmOlympus cores, 1.5 TB LPDDR5x) or next-generation x86 processors. Choose basedon workload and ecosystem requirements.
Hybrid/liquid cooling required for the thermal envelope of 8 Rubin GPUs at approximately 24 kW per system. Designed for dense rack deployments.
Train frontier models and next-generation AI architectures that require significantly more compute than current platforms can efficiently deliver. Rubin is built for the training jobs that will define the next wave of AI capability.
Serve increasingly large and complex models in production at massive scale. The Rubin platform is designed for the throughput and efficiency demands of next-generation inference workloads.
Run next-generation simulations, climate models, drug discovery pipelines, and other compute-intensive workloads that will benefit from Rubin's expected gainsin density and efficiency.
