NVIDIA GB200 NVL72 Optimized with Slurm for AI Supercomputing

NVIDIA’s GB200 NVL72, a cutting-edge rack-scale AI supercomputer, is now achieving optimized performance through topology-aware job scheduling with Slurm. This advancement is critical as AI models, particularly trillion-parameter large language models (LLMs), demand both unprecedented compute power and efficient resource allocation. The system, built on NVIDIA’s Blackwell architecture, delivers up to 130 terabytes per second (TB/s) of GPU communication bandwidth and supports training and inference for some of the most complex AI workloads.

The GB200 NVL72 integrates 72 NVIDIA Blackwell GPUs and 36 Grace CPUs in a single rack, interconnected via NVIDIA NVLink. According to NVIDIA, this setup not only supports large-scale training but also accelerates real-time inference with over 1.5 million tokens per second for OpenAI GPT models. However, maximizing this performance in shared clusters requires strategic scheduling, as highlighted in NVIDIA’s collaboration with SchedMD to enhance Slurm’s topology-aware capabilities.

Why Scheduling Matters for Exascale Systems

AI workloads often run on shared clusters, where multiple jobs must compete for resources. Without topology-aware scheduling, jobs may span across NVLink domains inefficiently, leading to resource fragmentation and reduced performance. The newly introduced Slurm topology/block plugin aligns jobs with the physical network layout of the GB200 NVL72, preserving locality and minimizing fragmentation. This ensures that GPU resources are allocated in a way that maximizes bandwidth and compute efficiency.

For example, NVIDIA’s simulation of a 5,000-node GB200 NVL72 cluster showed that the new scheduling policies achieved GPU occupancy within 1% of a theoretical maximum while maintaining high job efficiency. The plugin also strategically placed smaller jobs to free up resources for larger AI training tasks, striking a balance between utilization and performance.

Segment Sizing and Best Practices

One of the key features of the GB200 NVL72 system is its support for larger segment sizes. While previous systems like the NVIDIA HGX H100 were limited to a single-node segment size, the GB200 NVL72 can handle segments up to 18 nodes. This flexibility allows operators to tailor segment sizes to specific workloads, such as using 16-node segments for high-bandwidth models like mixture-of-experts (MoE) training, or smaller segments for less demanding tasks.

In practice, NVIDIA recommends segment sizes that align with workload characteristics. For example, large jobs of 128 GPUs or more should use 16-node segments, while smaller jobs can be allocated to single-node segments. These configurations prevent over-constraining the scheduler and maintain high cluster utilization, even as job profiles evolve over time.

Market Context and Adoption

Commercial deployments of the GB200 NVL72 began ramping up in 2025, with systems priced between $2.8 million and $3.4 million per rack. As of March 2026, prices have reportedly climbed to as high as $8.8 million for fully configured systems, reflecting soaring demand for advanced AI infrastructure. NVIDIA’s data center revenue, which reached $39.1 billion in Q1 FY26, underscores the growing reliance on systems like the GB200 NVL72 for AI and HPC workloads.

For traders, NVIDIA’s stock (NASDAQ: NVDA) is currently trading at $221.42 with a market cap of $5.40 trillion. The company’s leadership in AI hardware, combined with innovations in software like Slurm’s topology-aware scheduling, positions it strongly in the rapidly expanding AI and HPC markets.

Looking Ahead

The GB200 NVL72 represents a significant leap forward in AI supercomputing, but its full potential hinges on efficient workload management. NVIDIA’s partnership with SchedMD to refine Slurm demonstrates how software can complement hardware to achieve exascale performance. For organizations deploying these systems, continuous monitoring and simulation-based testing of scheduling policies will be key to maintaining both high utilization and peak performance.

As AI models continue to grow in complexity, the GB200 NVL72 and similar architectures will likely become foundational to large-scale AI training and inference. With further advancements in scheduling algorithms and hardware integration, the era of exascale AI computing is just beginning.

Image source: Shutterstock