Stanford University deploys Marlowe DGX SuperPOD with 248 Nvidia GPUs for research access

Stanford University has brought its first GPU-based supercomputer online, and it’s named after a fictional detective. Marlowe, a nod to Raymond Chandler’s hard-boiled Philip Marlowe, is now hunting down answers to research questions across political science, astrophysics, and everything in between.

The system is built on an Nvidia DGX H100 SuperPOD architecture. It houses 31 DGX H100 nodes containing 248 Nvidia H100 80GB GPUs, paired with 2.5 petabytes of DDN Lustre high-performance storage.

What Marlowe actually does

Marlowe offers non-preemptible access, meaning researchers get dedicated, uninterrupted computing time rather than preemptible cloud instances where jobs can be kicked off the server when someone with higher priority needs the resources.

The supercomputer became operational for the Stanford research community on January 15, 2025. It now supports over 500 active research accounts spanning all seven of Stanford’s school divisions.

Each of Marlowe’s 31 nodes includes two Intel Xeon Platinum 8480C CPUs with 112 cores. The system is optimized specifically for large-scale AI model training. Computations that previously took weeks can now be completed in days or hours.

Why a university needs a supercomputer in 2025

Stanford already operates smaller CPU and GPU clusters for research workloads. Marlowe complements that existing ecosystem by offering a robust multi-node reference architecture. A single DGX H100 node is powerful. Thirty-one of them networked together as a SuperPOD is a different category of machine entirely, designed so nodes can work in concert on a single massive training job rather than just handling a queue of smaller independent tasks.

The deployment also reflects a strategic move to reduce Stanford’s dependence on cloud computing for GPU-intensive work, providing persistent access to high-performance computational resources and lessening reliance on preemptible cloud alternatives.

What this means for the broader AI research landscape

Marlowe is Stanford’s first GPU-based supercomputer. Nvidia’s DGX SuperPOD is a standardized, scalable blueprint that other institutions and enterprises can replicate, and Stanford’s deployment serves as a case study for other universities considering similar investments.

With over 500 research accounts across all seven school divisions, the supercomputer is being used by political scientists, astrophysicists, biologists, and dozens of other fields that have discovered GPU-accelerated computing can transform their work.