Delta GPU

RP account needed

Delta GPU is a GPU-accelerated compute cluster consisting of NVIDIA A40, A100, and H200 nodes and one mixed AMD MI100/MI210 node, with four to nine GPUs, 256 GB to 2 TB of RAM, and 1.5 TB to 2 TB of local storage per node. It is particularly well suited for GPU-accelerated and hybrid CPU-GPU workloads, and is often used for AI and machine learning training and inference, data analytics, and scientific simulations. It includes a great deal of GPU-optimized AI, scientific computing, and parallel programming software.

Submitting Jobs

You can run jobs at different sizes and durations on Delta GPU. The following lists the different queues that you can submit to, describing how many nodes you get, how long you can run, the type of resources you get, and the average wait time.

Delta uses the Slurm scheduler. If you do not request a wall-clock time or memory amount, Delta defaults to 30 minutes and 1,000 MB of memory per core. Submit batch jobs with sbatch; use srun to run a command or start an interactive job; and use salloc to reserve resources for multiple commands. For instructions and Delta-specific examples, see [Delta Batch Jobs] for sbatch and [Delta Interactive Jobs] for srun and salloc.

Compute resources are also available through [Delta Open OnDemand] applications such as JupyterLab, VS Code Code Server, and noVNC Desktop.

Track your work with the usual Slurm commands: squeue -u $USER shows the status of your queued and running jobs, scontrol show job followed by a job ID shows the detail for one job, sinfo -a summarizes all Delta partitions and their node states, and scancel followed by a job ID cancels a queued job or terminates a running one.

Use gpuA100x4 as the default production partition. Use gpuA40x4 for workloads that benefit from A40 GPU memory or ray-tracing support, gpuA100x8 for large multi-GPU A100 workloads, gpuH200x8 for memory-intensive or FP8/FP16 workloads, and gpuMI100x8 for AMD GPU, ROCm, or large-memory workloads. Use the interactive partitions for short development, testing, and debugging sessions. Use the A100 and A40 preemptible partitions only for checkpointing-capable jobs that can tolerate interruption and loss of unsaved progress.

Wall-time limits for each partition are in the table below. Node sharing is the default. To request exclusive use of a GPU node, request all consumable resources for the node or use --exclusive --mem=0. CPU and GPU allocations use separate local charge accounts, so submit GPU jobs with the appropriate Delta GPU account.

For complete job guidance, see [Delta Running Jobs]. GPU, PyTorch, job-array, and other example scripts are available in [Delta Sample Scripts].

Queue specifications

Metrics updated 2026-09-10

Name Purpose CPU cores / node GPUs / node Num nodes Node RAM Max wallclock Wait time
30-day trend
Wall time
30-day trend
Number of jobs run
30 days
gpuA100x4 Standard production GPU jobs; this is Delta’s default GPU partition 1x AMD EPYC 7763 "Milan" (64 cores) 4 NVIDIA A100 (40 GB vRAM) 100 256 GB 48h
gpuA100x4 wait time: average 11.9 hours, range 3.2 to 39.4 hours over 30 days
gpuA100x4 wall time: average 1.6 hours, range 0.1 to 2.5 hours over 30 days, wall-time limit 48h
107,282
gpuA100x4-preempt Checkpointing-capable A100 jobs that can tolerate interruption and loss of unsaved progress 1x AMD EPYC 7763 "Milan" (64 cores) 4 NVIDIA A100 (40 GB vRAM) 100 256 GB 48h
gpuA100x4-preempt wait time: average 11.4 hours, range 0 to 67.4 hours over 30 days
gpuA100x4-preempt wall time: average 0.2 hours, range 0 to 1.6 hours over 30 days, wall-time limit 48h
2,258
gpuA40x4 Production jobs that benefit from A40 GPU memory or ray-tracing support 1x AMD EPYC 7763 "Milan" (64 cores) 4 NVIDIA A40 (48 GB vRAM) 100 256 GB 48h
gpuA40x4 wait time: average 8.5 hours, range 1.2 to 20.2 hours over 30 days
gpuA40x4 wall time: average 1.1 hours, range 0.2 to 2.2 hours over 30 days, wall-time limit 48h
144,078
gpuA40x4-preempt Checkpointing-capable A40 jobs that can tolerate interruption and loss of unsaved progress 1x AMD EPYC 7763 "Milan" (64 cores) 4 NVIDIA A40 (48 GB vRAM) 100 256 GB 48h
gpuA40x4-preempt wait time: average 11.1 hours, range 0.1 to 75.9 hours over 30 days
gpuA40x4-preempt wall time: average 0.2 hours, range 0 to 2.1 hours over 30 days, wall-time limit 48h
6,587
gpuH200x8 Memory-intensive or FP8/FP16 production jobs; maximum 1 node per job 2x Intel Xeon Platinum 8558 (96 cores) 8 NVIDIA H200 (141 GB vRAM) 8 2048 GB 48h
gpuH200x8 wait time: average 9.2 hours, range 0 to 59.7 hours over 30 days
gpuH200x8 wall time: average 1.4 hours, range 0.3 to 4.5 hours over 30 days, wall-time limit 48h
12,831
gpuA100x8 Large-scale production jobs that need up to eight A100 GPUs and 2 TB of node memory 2x AMD EPYC 7763 "Milan" (128 cores) 8 NVIDIA A100 (40 GB vRAM) 6 2048 GB 48h
gpuA100x8 wait time: average 4.0 hours, range 0.2 to 9.3 hours over 30 days
gpuA100x8 wall time: average 0.4 hours, range 0 to 4.2 hours over 30 days, wall-time limit 48h
38,232
gpuMI100x8 AMD GPU and ROCm workloads, or large-memory jobs that can use this single-node partition 2x AMD EPYC 7763 "Milan" (128 cores) 8 AMD MI100, AMD MI210 (32 GB vRAM) 1 2048 GB 48h
gpuMI100x8 wait time: average 3.3 hours, range 0 to 12.2 hours over 30 days
gpuMI100x8 wall time: average 4.2 hours, range 0 to 16 hours over 30 days, wall-time limit 48h
474
gpuA100x4-interactive Short interactive sessions for development, testing, and debugging; maximum 4 nodes per job 1x AMD EPYC 7763 "Milan" (64 cores) 4 NVIDIA A100 (40 GB vRAM) 100 256 GB 1h
gpuA100x4-interactive wait time: average 0.8 hours, range 0 to 7.4 hours over 30 days
gpuA100x4-interactive wall time: average 0.3 hours, range 0.1 to 0.5 hours over 30 days, wall-time limit 1h
8,960
gpuA40x4-interactive Short interactive sessions for development, testing, and debugging; maximum 4 nodes per job 1x AMD EPYC 7763 "Milan" (64 cores) 4 NVIDIA A40 (48 GB vRAM) 100 256 GB 1h
gpuA40x4-interactive wait time: average 0.3 hours, range 0 to 3.5 hours over 30 days
gpuA40x4-interactive wall time: average 0.2 hours, range 0.1 to 0.4 hours over 30 days, wall-time limit 1h
10,044
gpuH200x8-interactive Short interactive H200 sessions for development, testing, and tuning; maximum 1 node per job 2x Intel Xeon Platinum 8558 (96 cores) 8 NVIDIA H200 (141 GB vRAM) 8 2048 GB 1h
gpuH200x8-interactive wait time: average 2.3 hours, range 0.1 to 15 hours over 30 days
gpuH200x8-interactive wall time: average 0.4 hours, range 0.2 to 0.6 hours over 30 days, wall-time limit 1h
3,193
gpuA100x8-interactive Short interactive multi-GPU sessions for development and testing; maximum 2 nodes per job 2x AMD EPYC 7763 "Milan" (128 cores) 8 NVIDIA A100 (40 GB vRAM) 6 2048 GB 1h
gpuA100x8-interactive wait time: average 5.8 hours, range 0 to 32.2 hours over 30 days
gpuA100x8-interactive wall time: average 0.3 hours, range 0 to 0.7 hours over 30 days, wall-time limit 1h
961
gpuMI100x8-interactive Short interactive AMD GPU sessions for development and testing; maximum 1 node per job 2x AMD EPYC 7763 "Milan" (128 cores) 8 AMD MI100, AMD MI210 (32 GB vRAM) 1 2048 GB 1h
gpuMI100x8-interactive wait time: average 0.1 hours, range 0 to 1.1 hours over 30 days
gpuMI100x8-interactive wall time: average 0.0 hours, range 0 to 0 hours over 30 days, wall-time limit 1h
21

Software

The following software is reported as available on Delta GPU by the Software Documentation Service.

Available Software

Application Description Research Discipline
alphafold 3 AlphaFold 3 is DeepMind’s next-generation AI system for predicting the structure and interactions of biological molecules, extending beyond proteins to model complexes involving DNA, RNA, ligands, ions, and other biomolecules. It enables high-accuracy structural predictions that support drug discovery, molecular biology, and biomedical research. Biochemistry and Molecular Biology
aocc The AMD Optimizing C/C++ Compiler (AOCC) is a high performance compiler suite with support for C, C++, and Fortran programming languages. Computer & Information Sciences
aocc-mixed AOCC (AMD Optimizing C/C++ and Fortran Compilers) is a production compiler suite from AMD for building high-performance x86 applications, optimized for AMD "Zen"-based EPYC and Ryzen processors. It is a fork of LLVM/Clang, with Flang as the Fortran front end, that adds AMD-specific optimizations — advanced vectorization, loop and interprocedural transformations, and Zen-targeted code generation — to improve performance of C, C++, and Fortran code on AMD hardware. It supports OpenMP and works with standard MPI libraries for parallel HPC workloads.
apprentice3
aria2 aria2 is a lightweight multi-protocol & multi-source command-line download utility. It supports HTTP/HTTPS, FTP, SFTP, BitTorrent and Metalink. Computer & Information Sciences, Other Computer & Information Sciences
atp Abnormal Termination Processing (ATP) is a tool that monitors Cray system user applications. If an application encounters a fatal signal, ATP will handle the signal and perform analysis on the dying application. Description Source: https://cpe.ext.hpe.com/docs/debugging-tools/atp.1.html
aws-ofi-nccl aws-ofi-nccl is a network plug-in that lets NCCL (NVIDIA's Collective Communications Library) and RCCL (AMD's equivalent) use libfabric — the OpenFabrics Interfaces (OFI) — as their network transport. It maps NCCL/RCCL's transport calls onto libfabric's reliable interface, giving multi-GPU, multi-node collective communication the benefits of libfabric such as reliable messaging and operating-system bypass. It was originally developed by AWS for the Elastic Fabric Adapter (EFA), but works with most libfabric providers, so it is also used on HPC interconnects such as Slingshot and InfiniBand. Computer Science
bamtools BamTools is a software toolkit for manipulating BAM (Binary Alignment Map) files. It provides both a C++ API for BAM file input and output, and a command-line tool for performing various operations on BAM files. Biological Sciences
beagle Beagle is a high-performance software library for phylogenetic inference using maximum likelihood methods. It can be used for reconstructing large phylogenetic trees from molecular sequence data with high efficiency and accuracy. Biological Sciences
bedtools2 Bedtools2 is a powerful suite of tools for manipulating genomic features and annotations. It allows users to perform a wide range of tasks on genomic intervals, such as intersecting, merging, comparing, and analyzing BED, GFF, VCF, and other common genomic file formats. Biological Sciences
boost Boost is a set of C++ libraries that provides support for tasks and structures such as linear algebra, pseudorandom number generation, multithreading, image processing, regular expressions, and unit testing. It aims to extend the functionality of C++ programming language and make it more efficient and powerful. Computer & Information Sciences
bowtie Bowtie is a fast and memory-efficient short read aligner for short DNA sequences. It aligns short DNA sequences (reads) to the human genome at a rate of over 25 million reads per hour on a typical workstation with a large amount of memory. Biological Sciences
bowtie2 Bowtie 2 is an ultrafast and memory-efficient tool for aligning sequencing reads to long reference sequences. It is particularly well-suited for aligning sequencing reads of DNA sequences, such as those generated by high-throughput sequencing technologies. Biological Sciences
btop btop is a resource monitor that provides a visually appealing and interactive interface for monitoring system resources such as CPU, memory, and network usage. Computer Science
cce Compiler Construction Engine (CCE) is a software tool used for building compilers and interpreters. It provides a set of tools and libraries to assist in the creation of programming language implementations. Computer & Information Sciences
cce-mixed cce-mixed is a software package designed for performing calculations in the framework of many-body perturbation theory for the electronic structure of solids. It focuses on the GW approximation, self-energy calculations, and the solution of the Bethe-Salpeter equation for optical excitations. Physical Sciences
cmake CMake is an open-source, cross-platform family of tools designed to build, test, and package software. It is used to control the software compilation process using simple platform and compiler-independent configuration files. Engineering & Technology
colabfold ColabFold is a collaborative project that aims to bring together deep learning and biophysical modeling for protein structure prediction. It provides an easy-to-use interface for researchers to perform protein folding predictions using state-of-the-art techniques. Biological Sciences
cpe CPE (Common Platform Enumeration) is a standardized method to describe and identify classes of applications, operating systems, and hardware devices in a consistent format. It provides unique identifiers for these entities to facilitate accurate and efficient information exchange between different security tools and databases. Computer & Information Sciences
craype Cray Programming Environment (CrayPE) is a suite of tools and libraries designed for high-performance computing on Cray supercomputers. Computer Science
craype-accel The craype-accel module is a Cray Programming Environment (CrayPE) module that provides compiler and linker support for targeting accelerator devices such as GPUs on Cray supercomputers. It configures the programming environment to enable offloading and optimized compilation for accelerators, integrating with Cray's compiler wrappers and runtime libraries. Computer Science
craype-x86
craypkg-gen craypkg-gen is a command-line utility provided by Cray (now part of HPE) for generating and managing software package manifests and build configurations in Cray's software stack. It is used to create package metadata, define dependencies, and automate the packaging process for HPC software environments on Cray systems. Software Engineering, Systems, and Development
cti CTI (Common Threat Intelligence) is an open source library designed to parse and manipulate structured threat intelligence information. It provides a set of APIs to work with different threat intelligence data formats and supports the serialization of data into various formats.
cuda CUDA is a parallel computing platform and application programming interface model created by NVIDIA. It allows software developers to use a CUDA-enabled graphics processing unit for general-purpose processing. Computer & Information Sciences
cuda-compat cuda-compat is a compatibility package provided by NVIDIA to enable running CUDA applications compiled with older CUDA toolkit versions on newer CUDA driver versions. It provides runtime libraries and driver components that bridge compatibility gaps between different CUDA toolkit and driver versions, facilitating smooth execution of CUDA applications across diverse system environments. Computer Science
cudatoolkit The CUDA Toolkit is a development environment for creating high-performance GPU-accelerated applications. It includes libraries, debugging and optimization tools, a compiler, and a runtime environment. Artificial Intelligence and Intelligent Systems
cue-login-env Cue-login-env is a Python library that provides an easy way to configure user-specific environments within shared servers or computing clusters. It helps users manage their environment variables, aliases, and paths without interfering with the system-wide configurations. Computer & Information Sciences
default-rh96
default-rh96test
default-rh9test
dos2unix dos2unix is a command-line utility that converts text files with DOS or Windows line endings (CRLF) to Unix line endings (LF) format. Computer & Information Sciences
dsmml dsmml is a deep learning library specifically designed for molecular machine learning applications. It offers a range of tools and techniques tailored for computational chemistry and drug discovery tasks. Other Chemical Sciences
dyninst Dyninst is a versatile and powerful binary instrumentation and analysis tool that provides capabilities for runtime code analysis, modification, and monitoring in C, C++, and Fortran applications. Computer & Information Sciences
fastqc FastQC is a quality control tool for high throughput sequence data. It provides a detailed overview of the quality and potential issues in sequencing data generated by high throughput sequencing pipelines. Biological Sciences
ffmpeg FFmpeg is a free and open-source software project consisting of a large suite of libraries and programs for handling video, audio, and other multimedia files and streams. Computer Science
fftw FFTW is a fast C library for computing the discrete Fourier transform (DFT) in one or more dimensions, of arbitrary input size, and of both real and complex data. It is widely used in areas such as signal processing, data compression, and solving partial differential equations. Mathematics
gcc-native gcc-native is a compiler system produced by the GNU Project supporting various programming languages, most notably C, C++, and Fortran. It is the default compiler for many Unix-like operating systems. Computer & Information Sciences
gcc-native-mixed The gcc-native-mixed compiler is a version of the GNU Compiler Collection (GCC) that is specifically optimized for mixed-language programming, allowing developers to combine different programming languages within the same codebase. Computer Science
gdb4hpc gdb4hpc is a plugin designed to enhance the GNU Debugger (GDB) for High Performance Computing (HPC) applications. It provides additional features to aid in debugging parallel and distributed applications typically used in HPC environments. Engineering & Technology
gmake GNU Make is a tool that controls the generation of executables and other non-source files from a program's source files. It is widely used for managing build processes in software development. Software Engineering
gnuplot Gnuplot is a portable command-line driven graphing utility for Linux, OS/2, MS Windows, OSX, VMS, and many other platforms. It is capable of producing interactive plots and supports various output formats. Physical Sciences
gromacs GROMACS (GROningen MAssive Parallel MD for Molecular Dynamics) is a versatile package for molecular dynamics simulations with a strong emphasis on high-performance computing capabilities. Biological Sciences
gsl The GNU Scientific Library (GSL) is a numerical library for C and C++ programmers. It provides a wide range of mathematical routines such as special functions, linear algebra, interpolation, numerical integration, random numbers, and more. Other Mathematics
gurobi Gurobi Optimization is a high-performance mathematical programming solver for linear programming (LP), mixed-integer programming (MIP), and other related optimization problems. Mathematics
gzip gzip is a file compression and decompression tool used to reduce the size of files for storage or transfer purposes. It is a widely used compression utility in Unix-like operating systems. Engineering & Technology
hdf5 HDF5 (Hierarchical Data Format version 5) is a file format and a suite of tools developed by the HDF Group for managing and storing large and complex data. It is designed to store and organize large amounts of data efficiently, enabling fast data access and sharing. Computer Science
hdf5-parallel HDF5 Parallel is an extension of the HDF5 library that allows for parallel I/O operations, enabling efficient data access and manipulation in high-performance computing environments. Data Management
hisat2 HISAT2 is a fast and sensitive alignment program for mapping next-generation sequencing reads (both DNA and RNA) against the general human population (as well as against a single reference genome). It builds on HISAT and Bowtie, allowing for significantly improved performance and new capabilities. Biological Sciences
hpc-gpt opencode OpenCode is an open-source AI coding agent available for the terminal, desktop, and IDEs that helps developers understand codebases, plan features, and write or refactor code. Powered by Large Language Models, it allows users to interact via text prompts and images to execute codebase changes directly. It streamlines development with built-in tools like fuzzy file searching, an intuitive plan-and-build workflow, quick undo/redo commands, and extensive customization options to fit any coding environment.
htop htop is an interactive system-monitor process-viewer for Unix systems. It provides a detailed overview of system performance and resource usage in a visually appealing and user-friendly interface. Engineering & Technology
htslib HTSlib is a C library for processing high-throughput sequencing data in the form of SAM, BAM, and CRAM formats. It allows for efficient reading, writing, and manipulation of such files, as well as various operations such as format conversion, indexing, and retrieval of specific sequences. Biological Sciences
imagemagick ImageMagick is a free and open-source software suite for creating, editing, composing, or converting bitmap images. Computer & Information Sciences
imagenet Imagenet is a large-scale visual database designed for use in visual object recognition research. It is widely used as a benchmark dataset for training and evaluating deep learning models, especially convolutional neural networks (CNNs). Various software implementations and utilities exist to facilitate downloading, preprocessing, training, and evaluating models on the ImageNet dataset. Artificial Intelligence and Intelligent Systems
impute2 IMPUTE2 is a tool for imputing ungenotyped markers in large-scale genome-wide association studies. Biological Sciences
intel-oneapi-compilers Intel oneAPI Compilers provide a unified programming model for high-performance computing, enabling developers to optimize applications across various architectures including CPUs and GPUs. Software Engineering
intel-oneapi-mkl Intel oneAPI Math Kernel Library (MKL) provides highly optimized mathematical functions for scientific computing, including linear algebra, fast Fourier transforms, and vector mathematics. Applied Mathematics
ior ior is a versatile software tool for analyzing and optimizing input/output (I/O) performance in high-performance computing (HPC) systems. It is designed to assess and improve the efficiency of data transfer operations in parallel computing environments. Computer Science
iperf3 iperf3 is a tool for actively measuring the maximum achievable bandwidth on IP networks. It supports various parameters to customize the testing process and is widely used for network performance testing and tuning. Computer & Information Sciences
julia Julia is a high-level, high-performance dynamic programming language for technical computing, with syntax that is familiar to users of other technical computing environments. It provides a sophisticated compiler, distributed parallel execution, numerical accuracy, and extensive mathematical function libraries. Computer & Information Sciences
kraken2 Kraken 2 is a taxonomic sequence classifier that assigns taxonomic labels to short DNA sequences, such as those generated by high-throughput sequencing technologies. Metagenomics
lammps LAMMPS (Large-scale Atomic/Molecular Massively Parallel Simulator) is a classical molecular dynamics code designed for simulating large-scale atomistic systems. It is highly versatile and can be used to model a wide range of materials and complex molecular structures. Chemical Sciences
libfabric libfabric is a high-performance library for fabric interfaces, designed to abstract fabric communication resources to applications while providing flexibility to applications that need to bypass the fabric interfaces. It provides fabric-agnostic interfaces for communication in high-performance computing (HPC) environments. Engineering & Technology
libsci libsci is a library of optimized scientific routines for high-performance computing (HPC) environments. It provides a collection of mathematical functions and computational kernels tailored for scientific and engineering applications that require high computational efficiency. Other Mathematics
libsci_acc libsci_acc is a high-performance, GPU-accelerated math library developed by Cray (now part of HPE) that provides optimized implementations of BLAS, LAPACK, and other scientific computing routines. It is designed to deliver accelerated linear algebra and scientific computations on Cray supercomputers equipped with NVIDIA GPUs, leveraging CUDA for performance improvements. Applied Mathematics
llvm The LLVM Project is a collection of modular and reusable compiler and toolchain technologies. LLVM is used for developing compiler back-ends, static analysis tools, and JIT compilation engines. Computer & Information Sciences
lmod Lmod is an environment module system that helps users manage their software environment by providing a convenient way to dynamically modify the user's environment through modulefiles. It supports multiple collections of modules that can be easily swapped in and out, allowing users to load the software packages and versions they need for their work. Engineering & Technology
matlab MATLAB is a high-performance language for technical computing and visualization. It integrates computation, visualization, and programming in an easy-to-use environment where problems and solutions are expressed in familiar mathematical notation. Mathematics, Applied Mathematics
mdtest MDTest is a file system metadata benchmark that includes tests to measure metadata operations, such as file creates, stat operations, and directory operations. It is often used to assess the metadata performance of parallel file systems. Computer & Information Sciences
mesa Mesa is an open-source agent-based modeling framework that allows users to create and analyze computational models of complex adaptive systems. Computer & Information Sciences
mesa-glu mesa-gl is an open-source implementation of the OpenGL specification, a system for rendering interactive 3D graphics, and a widely used and versatile graphics library. mesa-glu is the OpenGL Utility Library (GLU) component of Mesa, providing useful functions for building OpenGL applications.
miniforge3-python Miniforge is a community-driven minimal installer for conda, allowing users to create and manage environments with Python and other packages. Computer Science
mpich MPICH is a high-performance and widely portable implementation of the Message Passing Interface (MPI) standard. It enables parallel computing on a wide variety of platforms and architectures. Computer & Information Sciences
mpich-abi mpich-abi is a compatibility layer and utility set designed to enable binary compatibility between different MPICH implementations and versions. It facilitates running MPI applications compiled against one MPICH ABI (Application Binary Interface) version on another MPICH runtime without recompilation, easing software deployment and integration in HPC environments. Computer Science
mrnet MRNET (Mendelian Randomization using Network models) is a software package for performing causal inference through Mendelian randomization and network modeling. It allows researchers to investigate causal relationships between traits by leveraging genetic instrumental variables and network structures.
namd NAMD (NAnoscale Molecular Dynamics) is a parallel molecular dynamics code designed for high-performance simulation of large biomolecular systems. It is optimized for the simulation of biomolecular systems containing millions of atoms. Biochemistry and Molecular Biology
netcdf-cxx netcdf-cxx is a C++ interface for the NetCDF data format, which provides implementations to read and write self-describing, machine-independent binary data files. It is designed to simplify the use of NetCDF in C++ applications. Earth & Environmental Sciences
netcdf-cxx4 netcdf-cxx4 is a C++ library for accessing netCDF-4 classic and extended model data formats. It provides a high-level interface for reading and writing netCDF-4 data files in C++. Computer & Information Sciences
nvbandwidth nvbandwidth is a command-line utility provided by NVIDIA to measure the memory bandwidth of NVIDIA GPUs. It is designed to benchmark and analyze the effective bandwidth of GPU memory operations, helping developers optimize memory throughput for CUDA applications. Computer Science
nvhpc NVIDIA HPC SDK is a comprehensive, integrated suite of compilers, libraries, and tools for the development and optimization of GPU-accelerated high-performance computing (HPC) applications. Computer & Information Sciences
nvhpc-byo-compiler The NVIDIA HPC SDK Bring Your Own (BYO) Compiler is a feature that allows users to integrate their own compiler toolchains with the NVIDIA HPC SDK environment. This enables leveraging NVIDIA's HPC libraries, tools, and optimizations while compiling code with a preferred or custom compiler, facilitating flexibility in HPC application development and deployment. Software Engineering, and Development
nvhpc-hpcx nvhpc-hpcx is a software stack combining NVIDIA HPC SDK compilers and tools with the HPC-X communication library suite. It is designed to optimize high-performance computing applications on NVIDIA GPU-accelerated clusters by providing high-performance MPI, SHMEM, and NCCL implementations integrated with NVIDIA compilers and CUDA libraries. Computer Science
nvhpc-hpcx-2.20-cuda12 nvhpc-hpcx-2.20-cuda12 is a software module combining NVIDIA HPC SDK (nvhpc) version 22.0 with HPC-X 2.20 and CUDA 12 support. It provides a comprehensive development environment for high-performance computing applications, including compilers, libraries, and tools optimized for NVIDIA GPUs and high-speed interconnects. HPC-X is a high-performance communication library stack that includes MPI, SHMEM, and NCCL, optimized for InfiniBand and NVIDIA networking hardware. CUDA 12 enables GPU acceleration with the latest features and performance improvements. Computer Science
nvhpc-hpcx-cuda12 nvhpc-hpcx-cuda12 is a software stack combining NVIDIA HPC SDK (including CUDA 12) with HPC-X, a high-performance communication library optimized for HPC clusters. It provides a comprehensive environment for developing, building, and running GPU-accelerated HPC applications with efficient MPI and SHMEM communication over InfiniBand and other high-speed interconnects. Computer Science
nvhpc-hpcx-cuda12-openmpi4
nvhpc-hpcx-cuda12-openmpi5
nvhpc-hpcx-cuda13
nvhpc-hpcx-cuda13-openmpi4
nvhpc-hpcx-cuda13-openmpi5
nvhpc-nompi nvhpc-nompi is a variant of the NVIDIA HPC SDK compilers and tools suite that excludes MPI (Message Passing Interface) support. It provides optimized compilers for C, C++, and Fortran targeting NVIDIA GPUs and CPUs, enabling high-performance computing applications without MPI parallelization. Computer Science
nvidia NVIDIA Corporation is an American technology company that designs graphics processing units (GPUs) for the gaming, cryptocurrency, and professional markets, as well as system on a chip units (SoCs) for the mobile computing and automotive market. Computer & Information Sciences
nvidia-mixed This modulefile defines the system paths and environment variables needed to use the NVIDIA HPC C, C++, and Fortran Compilers.
nvtop nvtop is an NVIDIA GPU system monitoring tool with a user-friendly CLI interface. It provides real-time monitoring of GPU utilization, memory usage, temperature, power consumption, and other important metrics. Computer & Information Sciences
nwchem NWChem is an open-source computational chemistry package designed for high-performance simulations of large-scale molecular systems. It provides various tools and algorithms for both classical and quantum chemical calculations. Chemical Sciences
openblas OpenBLAS is an open-source implementation of the Basic Linear Algebra Subprograms (BLAS) library with optimized functions for matrix operations and linear algebra computations. It provides fast and efficient routines for common linear algebra operations. Computer & Information Sciences
opencv OpenCV (Open Source Computer Vision Library) is an open-source computer vision and machine learning software library. It provides a wide range of functionalities for real-time computer vision, including image processing, object detection, object tracking, and machine learning algorithms. Computer & Information Sciences
osmesa osmesa is an extraction of the Mesa 3-D library's software rasterizer and offscreen rendering implementation of OpenGL 2.0 style APIs.
osu-micro-benchmarks OSU Micro-Benchmarks is a suite of parallel micro-benchmarks written in C with MPI and OpenMP implementations, designed for evaluating the performance of parallel computing systems and HPC networks. Engineering & Technology
papi PAPI (Performance Application Programming Interface) is a portable and efficient API that provides a consistent interface and methodology for collecting performance counter data from the underlying hardware. Computer Science
parallel GNU Parallel is a command-line tool for Unix-like operating systems that allows the user to execute shell scripts or commands in parallel. It is designed to simplify parallel job execution on single or multiple processors without the need for complex scripting. Computer & Information Sciences
parallel-netcdf Parallel-NetCDF is a library providing high-performance I/O in parallel environments for big data applications with large scientific datasets. It offers parallel I/O capabilities for NetCDF files, which are commonly used in climate modeling, weather forecasting, and other scientific disciplines. Physical Sciences
paraview ParaView is an open-source, multi-platform data analysis and visualization application. It is designed to handle large data sets and offers a wide range of visualization techniques to analyze and interact with the data. Physical Sciences, Engineering & Technology, Natural Sciences, Computer & Information Sciences, Other Natural Sciences
patch The `patch` utility is a tool used to apply changes to files based on a diff file, which contains the differences between two versions of a file or set of files. Other Computer and Information Sciences
perftools-base perftools is a collection of performance analysis tools for Linux-based systems. perftools-base is the base package of perftools which includes basic performance analysis tools for profiling CPU and memory usage. Computer & Information Sciences
petsc PETSc, the Portable, Extensible Toolkit for Scientific Computation, is a suite of data structures and routines for the scalable (parallel) solution of scientific applications modeled by partial differential equations. Natural Sciences
picard Picard is a set of command line tools for manipulating high-throughput sequencing (HTS) data and formats such as SAM/BAM/CRAM and VCF. It is primarily used for processing and analyzing next-generation sequencing (NGS) data in genomics. Biological Sciences
plumed PLUMED is an open-source library for free energy calculations in molecular dynamics simulations. It allows for the easy and flexible implementation of enhanced sampling techniques and the calculation of various collective variables. Biological Sciences
pmi PMI is a low-level API (Application Programming Interface) designed for managing processes in an HPC environment. It is primarily used to manage resources, handle process launch and communication, and manage job execution within parallel computing frameworks, such as those using MPI (Message Passing Interface) for distributed computing.
pocl Portable Computing Language (pocl) is an open-source implementation of the OpenCL standard which aims to provide a portable programming layer for execution on various devices including CPUs, GPUs, and FPGAs. It enables developers to write parallel programs that can be executed efficiently across different hardware platforms. Computer & Information Sciences
prgenv
prgenv-amd prgenv-amd is a programming environment module that provides compiler and toolchain support optimized for AMD processors on HPC clusters. It configures environment variables and paths to enable building and running applications with AMD-specific compilers and libraries. Computer Science
prgenv-aocc prgenv-aocc is a programming environment module that configures the AOCC (AMD Optimizing C/C++ Compiler) compiler toolchain for HPC applications. It sets up environment variables and paths to enable users to compile and optimize their code using the AOCC compiler suite on supported HPC systems. Computer Science
prgenv-cray-amd The PrgEnv-cray-amd modulefile loads the mixed CRAY/AMD Programming Environment, which uses the AMD Clang C/C++ compiler and the Cray Compiling Environment (CCE) Fortran compiler.
prgenv-gnu The prgenv-gnu is a programming environment module that provides the GNU compiler collection (GCC) for compiling and running codes. It sets up the environment variables needed to compile and run programs using the GNU compilers. Computer & Information Sciences, Software Engineering, Systems & Development
prgenv-gnu-amd The PrgEnv-gnu-amd modulefile loads the mixed GNU/AMD Programming Environment, which uses the AMD Clang C/C++ compiler and the GNU compiler suite Fortran compiler
prgenv-nvidia prgenv-nvidia is a software environment that provides compilers and tools optimized for NVIDIA GPUs, allowing users to develop and execute GPU-accelerated applications efficiently. Computer & Information Sciences
python Python is a high-level, interpreted programming language known for its simplicity and readability. It supports multiple programming paradigms and has a vast ecosystem of libraries and frameworks. Computer & Information Sciences, Software Engineering, Systems & Development
pytorch-conda A package that provides a conda environment for PyTorch, a popular deep learning framework. Artificial Intelligence and Intelligent Systems
r R is a free software environment for statistical computing and graphics. It compiles and runs on a wide variety of UNIX platforms, Windows, and MacOS. Computer Science
samtools Samtools is a suite of programs for interacting with high-throughput sequencing data generated in sequence alignment/map (SAM) format, such as those produced by the Short Read Mapping (SHRiMP) or Burrows-Wheeler Aligner (BWA) aligners. It allows various operations on SAM/BAM files, including indexing, sorting, merging, and manipulating sequence alignments. Biological Sciences
sanitizers4hpc sanitizers4hpc is a toolset designed for detecting and removing memory errors and data races in high-performance computing (HPC) applications. It offers various sanitizers to enhance the reliability and performance of parallel programs. Engineering & Technology
settarg Settarg is a command-line utility designed for setting target specific optimizations in C/C++ compilers. It helps in specifying optimization flags tailored for a particular CPU architecture to improve performance.
slurm-env slurm-env is a tool that helps manage environment variables for SLURM job scripts. SLURM, which stands for Simple Linux Utility for Resource Management, is an open-source job scheduler for Linux clusters. Engineering & Technology
star STAR (Spliced Transcripts Alignment to a Reference) is a RNA-seq read aligner that maps RNA-seq data to a reference genome to identify splice junctions between exons. It is particularly useful for mapping reads that span multiple exons, thereby aiding in the analysis of alternative splicing events. Biological Sciences
stat Stat is a statistical software package designed for data analysis, statistical modeling, hypothesis testing, and visualization. Mathematics
tar tar is a software utility used for packaging files together into an archive file often referred to as a "tarball". It is commonly used in Unix-like operating systems for creating, storing, and managing files in a single file. Computer & Information Sciences
tcl Tool Command Language (Tcl) is a dynamic programming language often used for creating scripts, applications, and embedded control systems. It provides a simple scripting language and a powerful code execution environment for various platforms. Computer & Information Sciences
tensorflow-conda TensorFlow is an open-source machine learning framework developed by Google, designed for building and training deep learning models. Artificial Intelligence and Intelligent Systems
tinker Tinker is a software package for molecular modeling and simulations, providing tools for energy minimization, molecular dynamics, and Monte Carlo simulations. Chemical Sciences
tk Tkinter is a standard GUI toolkit for Python. It is included with most Python installations and is widely used for building desktop applications in Python. Computer & Information Sciences, Computer Science, Software Engineering, Systems & Development
trimmomatic Trimmomatic is a flexible and efficient tool for trimming raw next-generation sequencing data. It provides several trimming modes and options to cater to different sequencing platforms and quality control requirements. Biological Sciences
util-linux-uuid util-linux-uuid is a command-line utility that allows users to generate universally unique identifiers (UUIDs) in various formats. UUIDs are 128-bit numbers used as identifiers for entities in computer systems with a high probability of being unique. Computer & Information Sciences
valgrind4hpc Valgrind4HPC is a performance analysis tool suite specifically designed for high-performance computing (HPC) environments. It provides a comprehensive set of tools for profiling, memory debugging, and thread analysis to help optimize and improve the performance of parallel and distributed applications on HPC systems. Computer & Information Sciences, Engineering & Technology
vcftools VCFtools is a software package for manipulating and analyzing genetic variation data stored in Variant Call Format (VCF) files. It provides a wide range of tools for quality control, filtering, summarization, and other analyses of genetic variants. Biological Sciences
visit Visit is an open-source interactive parallel visualization and graphical analysis tool. It is designed to handle very large data set visualization and enable users to visualize and analyze data from a variety of sources. Computer & Information Sciences
xcb-util-cursor xcb-util-cursor is a library that provides an interface for managing cursors in X11 applications using the XCB (X C Binding) protocol. Other Computer and Information Sciences
xpmem XPMEM is a kernel module that provides a kernel-level interface for sharing memory between processes. Computer & Information Sciences
xterm xterm is a terminal emulator for the X Window System. It provides a command-line interface for users to interact with the system and run shell commands. Other Computer and Information Sciences
zlib-ng zlib-ng is a fast, efficient, and portable data compression library that is a drop-in replacement for zlib. It is designed to be compatible with zlib while providing improved performance and additional features. Software Engineering
zmqnet ZMQNet is a .NET binding for ZeroMQ that enables high-performance asynchronous messaging between applications, processes, and distributed systems. It provides access to ZeroMQ communication patterns through .NET languages such as C#.

Storage

Delta GPU provides Home, Projects, Work HDD, Work NVMe, and Node-local storage. The work file systems are shared with DeltaAI and should be used for job I/O; home directories are separate, and home-directory snapshots are not backups. For details, see [Delta File Systems].

File System

Directory Path Quota Purge Backup Notes
Home $HOME 100 GB, 750,000 files Never purged Snapshots only, retained 30 days Software, scripts, and job files; not intended for job I/O. Snapshots live on the same hardware as the primary copy and are not a backup - back up anything you cannot lose.
Projects /projects/<project> 500 GB, 300,000 files Never purged Not backed up Shared project data, software, and results. Up to 25 TB by allocation request.
Work HDD /work/hdd/<project> 1 TB, 500,000 files Never purged Not backed up Active computational data and job I/O. Up to 100 TB by allocation request. Shared with DeltaAI.
Work NVMe /work/nvme/<project> See notes Never purged Not backed up Available upon request. Use quota to check assigned capacity and inode limits. Best for frequent small-file I/O. Shared with DeltaAI.
Node-local /tmp See notes Purged at job end Not backed up 1.5 TB on A40, A100, and AMD GPU nodes; 2 TB on H200 nodes; no enforced quota. Capacity may be shared by concurrent jobs. Use for fast small-file I/O and copy results before the job ends.

File Transfer

Use Globus for large transfers or many files and scp or rsync through Delta login nodes for small to modest transfers. The "ACCESS Delta" Globus collection uses your ACCESS identity; scp and rsync require NCSA Duo. For instructions, see [Delta Transferring Data].

Supported Methods Data Transfer Node / Globus Collection Notes
GLOBUS | RECOMMENDED ACCESS Delta Delta Globus
SCP login.delta.ncsa.illinois.edu SCP
RSYNC login.delta.ncsa.illinois.edu RSYNC

Datasets

Name Description
ColabFold

ColabFold predicts protein folding structures and complexes using MMseqs2, which gives faster performance than AlphaFold. The ColabFold databases are available on Delta at /work/hdd/data/colabfold. The colabfold module may be loaded to set the COLABFOLD_PATH environment variable to this path.

ImageNet

ImageNet is an image database organized according to the WordNet hierarchy, where each node of the hierarchy is depicted by hundreds to thousands of images. ImageNet database(s), version ILSVRC 2012-2017, are available on Delta at /work/hdd/data/imagenet. The imagenet module may be loaded to set the IMAGENET_PATH environment variable to this path. Access requires membership in the delta_imagenet group.

AlphaFold

Protein structure prediction datasets used for bioinformatics research with AlphaFold.

Ollama models

Pretrained large language models used with Ollama for local inference and AI experimentation.