Stanford needed a campus-wide GPU platform that could support interdisciplinary AI research at production scale without forcing labs to depend on fragmented clusters, complex allocations, or external resources for large multi-node training jobs.
Reported impact
Marlowe now supports more than 500 active research accounts, contributes to over 18 million GPU hours across Stanford's compute ecosystem each year, and enabled outcomes such as 90x faster Gaussian-splat fitting and publication-scale world-model research that previously would have been much harder to run on campus.
Stanford designed Marlowe as a DGX SuperPOD-based research platform that complements its Sherlock cluster and gives researchers a clear path from small experiments to large, non-preemptible training runs. The system uses DGX SuperPOD infrastructure with 248 NVIDIA Hopper GPUs, Base Command Manager for cluster provisioning and administration, and Quantum-2 InfiniBand networking so multi-node workloads can scale the way industry-scale model training requires. Researchers apply for access, begin with smaller jobs, and then expand to larger allocations as experiments mature, all on the same governed architecture. That shared platform has already supported work in world models, vision-language alignment, protein folding, and biological foundation models, and it gives Stanford a durable operating model for turning campus research ideas into repeatable, high-throughput AI programs.
Research datasets and training jobs move from Stanford labs into the Marlowe platform, where Base Command Manager provisions DGX resources and Quantum-2 InfiniBand connects multi-node runs across Hopper GPUs. Results then flow back into research groups for model evaluation, publication, and follow-on experimentation across the wider Stanford ecosystem.
Marlowe DGX SuperPOD provides hundreds of NVIDIA Hopper GPUs and supports over 500 active research accounts.
More than 18 million GPU hours are delivered annually across Stanford's compute ecosystem.
The NVIDIA Quantum-2 InfiniBand networking platform capabilities built into the DGX SuperPOD architecture have proven transformational for these users.
Using Marlowe, the team developed a scalable 2D Gaussian splatting (2DGS) pipeline, achieving over 90x faster fitting and around 97% GPU utilization compared to prior implementations.
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
NVIDIA Sample Oct 6, 2026 University of Pennsylvania Penn needed shared AI infrastructure that could support researchers across many disciplines without weeks-long wait times, fragmented storage, or the cost barriers that kept frontier-scale training and inference off… 1.5 years of compute time reduced to three days for a 40,000-document project NVIDIA DGX B200 systems · NVIDIA DGX SuperPOD · NVIDIA InfiniBand NVIDIA Sample Aug 27, 2026 Texas A&M University Texas A&M needed to give researchers across seven institutions access to shared GPU infrastructure for AI and simulation, but isolated workstations, slow grant-funded clusters, and long queues were limiting… $1M saved on a single project DDN high-performance storage · NVIDIA Base Command Manager · NVIDIA DGX SuperPOD NVIDIA Sample Aug 3, 2026 StudyFetch StudyFetch needed to keep AI tutoring affordable while scaling live lecture transcription and personalized learning for millions of students, but off-the-shelf cloud transcription created cost and latency ceilings that… 100M+ learning interactions NVIDIA H100 GPUs · NVIDIA L40S GPUs · NVIDIA NIM 