Gearing for the upcoming release of its XA NB3I-E12 AI server, ASUS has announced that the system has delivered top-tier results in MLPerf Training v6.0, thanks to the power of the NVIDIA HGX B300 platform and Intel Xeon Scalable processors.
Specifically, Llama 3.1-8B across both server and offline modes has been topped, which shows its superiority in scheduling and concurrency handling in real-world applications under server mode, and maximization of GPU utilization under offline mode. Meanwhile, GPT-OSS-120B interactive inference showcases ero-latency processing for complex reasoning in massive models.
This is done through the combination of 8x NVIDIA Blackwell Ultra GPUs and 2x Intel Xeon 6 Scalable processors that can support terabytes of DDR5 ECC RAM. As such, the system has proven that it can handle dense LLM workloads that prioritize throughput, latency, and memory bandwidth all at once, which is essential for enterprise AI deployment.
ASUS will be shipping the XA NB3I-E12 AI server on a global scale in the near future.
