Overview
ParaTools Pro for E4S™ - the Extreme-scale Scientific Software Stack, E4S™ hardened for commercial clouds and supported by ParaTools, Inc., provides a platform for developing and deploying HPC and AI/ML applications. This product is the Ubuntu (arm64) variant of the unified Heidi image; a Rocky Linux variant is available as a separate marketplace listing. It is designed for use with Adaptive Computing's Heidi AI Cloud Supercomputer , which provides multi-cloud HPC orchestration with automated infrastructure deployment, scaling, and cluster monitoring. A single unified image auto-detects its cluster role (head node / compute node) at boot time; the image is also usable standalone as a development environment. When run on EFA-capable instance families, cluster nodes are interconnected by a low-latency, high-bandwidth fabric based on AWS Elastic Fabric Adapter (EFA), fully leveraged by MVAPICH-Plus -- a proprietary MVAPICH derivative from our partners at X-Scale Solutions. NVIDIA NCCL and NVSHMEM accelerate multi-GPU communication, and the NVIDIA Inference Xfer Library (NIXL) provides high-throughput data movement for distributed AI inference. A web-based remote desktop is available for interactive GUI work. ParaTools Pro for E4S™ features a suite of over 100 HPC tools built using the Spack package manager and the MVAPICH-Plus MPI library tuned for EFA. HPC applications include CP2K, GROMACS, LAMMPS, NWChem, OpenFOAM, Quantum Espresso, WarpX, and Weather Research and Forecasting Model-WRF; the AI/ML Python stack includes Hugging Face Transformers, JAX, JupyterLab, Keras, matplotlib, NIXL, NVIDIA BioNeMo, NVIDIA NeMo™, OpenCV, PyTorch, SGLang, TensorFlow, and vLLM, unified into a single Python environment. The Orca and VSCodium IDEs are also pre-installed, along with AI agent tools (Hermes Agent with desktop dashboard, Ollama, OpenClaw). Performance analysis tools include HPCToolkit, TAU, TAU Commander, and the LLVM-based SALT-FM source instrumentation toolkit. New packages can be easily added using Spack and pip and are accessible across cluster nodes. It may be used for developing the next generation of generative AI applications using a suite of Python tools and interfaces.
E4S™ has built a unified computing environment for deployment of open-source projects. E4S™ was originally developed to provide a common software environment for the exascale leadership computing systems currently being deployed at DOE National Laboratories across the U.S. Support for ParaTools Pro for E4S™ is available through ParaTools, Inc. This product has additional charges associated with it for optional product support and updates.
This material is based upon work supported by the U.S. Department of Energy, Office of Science, Office of Advanced Scientific Computing Research (ASCR), under SBIR Award Number DE-SC0022502 ("E4S: Extreme-Scale Scientific Software Stack for Commercial Clouds").
Note: This product contains repackaged and tuned open source software (e.g., E4S™, Spack, and AI/ML tools like NVIDIA NeMo™, NVIDIA BioNeMo, TensorFlow, JAX, vLLM, etc.) which is configured and linked against an MVAPICH MPI implementation specifically developed and tuned for EFA.
The full list of E4S applications installed via Spack is as follows:
- adios
- adios2
- alquimia
- aml
- amrex
- arborx
- argobots
- ascent
- axom
- boost
- butterflypack
- cabana
- caliper
- chai
- chapel
- charliecloud
- conduit
- cp2k
- darshan-runtime
- darshan-util
- datatransferkit
- diy
- dyninst
- e4s-alc
- e4s-cl
- faodel
- fftx
- flecsi
- flit
- flux-core
- fpm
- gasnet
- ginkgo
- globalarrays
- gmp
- gotcha
- gromacs
- guacamole-client
- guacamole-server
- h5bench
- hdf5
- hdf5-vol-async
- hdf5-vol-cache
- hdf5-vol-log
- heffte
- hpctoolkit
- hpx
- hypre
- kokkos
- kokkos-fft
- kokkos-kernels
- kokkos-tools
- laghos
- lammps
- lbann
- legion
- libcatalyst
- libceed
- libnrm
- libpressio
- libquo
- libunwind
- loki
- magma
- mercury
- metall
- mfem
- mgard
- mpark-variant
- mpifileutils
- nccmp
- nco
- nek5000
- nekbone
- netcdf-fortran
- netlib-scalapack
- nrm
- nwchem
- omega-h
- openfoam
- openmpi
- openpmd-api
- papi
- papyrus
- parallel-netcdf
- parsec
- pdt
- petsc
- phist
- plasma
- plumed
- precice
- pruners-ninja
- pumi
- py-cinemasci
- py-h5py
- py-jupyterhub
- py-libensemble
- py-petsc4py
- py-pytuq
- qthreads
- quantum-espresso
- raja
- rempi
- saltfm
- scr
- slate
- slepc
- strumpack
- sundials
- superlu
- superlu-dist
- swig
- sz
- sz3
- tasmanian
- tau
- tomcat
- trilinos
- turbine
- umap
- umpire
- upcxx
- veloc
- vtk-m
- wannier90
- warpx
- wps
- wrf
- zfp
Highlights
- ParaTools Pro for E4S™ and Machine Learning stacks, including NVIDIA NeMo™, built and optimized for AWS EFA and Heidi
- A proprietary MVAPICH-Plus 4 MPI implementation that offers lower latency and higher throughput than default OpenMPI implementations for pre-installed applications and user-installed applications
- Over 100 HPC and AI applications managed via the Spack package manager, with a web-based remote desktop for interactive computing
Details
Features and programs
Financing for AWS Marketplace purchases
Pricing
Free trial
- ...
Dimension | Cost/hour |
|---|---|
c7g.4xlarge Recommended | $0.99 |
c7gn.8xlarge | $0.99 |
r6gd.2xlarge | $0.99 |
c7gn.large | $0.99 |
c7gn.medium | $0.99 |
c6gd.metal | $0.99 |
r7g.2xlarge | $0.99 |
i4g.large | $0.99 |
c6gn.medium | $0.99 |
r7gd.2xlarge | $0.99 |
Dimensions summary
Top-of-mind questions for buyers
Vendor refund policy
Refund Policy: The standard refund policy outlined here will be followed: https://docs.aws.amazon.com/marketplace/latest/userguide/refunds.html
Any additional refund enquiries can be sent to support@paratools.com and will be considered individually on a case by case basis.
How can we make this page better?
Legal
Vendor terms and conditions
Content disclaimer
Delivery details
64-bit (Arm) Amazon Machine Image (AMI)
Amazon Machine Image (AMI)
An AMI is a virtual image that provides the information required to launch an instance. Amazon EC2 (Elastic Compute Cloud) instances are virtual servers on which you can run your applications and workloads, offering varying combinations of CPU, memory, storage, and networking resources. You can launch as many instances from as many different AMIs as you need.
Version release notes
- Adaptive Heidi release: v2.0
- E4S release: 26.11
- Default MPI: MVAPICH-4 Plus
- OS: Ubuntu 26.04
*** UNIFIED IMAGE NOTICE *** Single unified image serves both Heidi Server (head node) and Heidi Node (compute node) roles. Image auto-detects role at boot when deployed with Adaptive Computing's Heidi AI Cloud Supercomputer. Also usable standalone as a development environment.
*** OS VARIANT NOTICE *** Ubuntu 26.04 arm64 variant. For the Rocky Linux 9 arm64 variant, see: https://aws.amazon.com/marketplace/pp/prodview-taspdxtsjkvok
*** UNIFIED PYTHON NOTICE *** The AI/ML Python stack is now a single unified Python 3.12.14 environment. NeMo, BioNeMo, PyTorch, TensorFlow, JAX, and vLLM all share one interpreter; the separate NeMo virtual environment (/usr/local/py-env/nemo) is no longer used.
IMPORTANT:
- Usable standalone as a development environment or shared-memory system. Multi-node SLURM-scheduled jobs and cluster orchestration activate when deployed with Adaptive Computing's Heidi AI Cloud Supercomputer.
- MVAPICH-4 Plus provides optimized multi-node support without strictly requiring EFA-enabled instances; EFA is still recommended for best performance.
Updates in this version: Platform:
- Ubuntu 26.04 LTS (from 24.04)
- E4S 26.11 scientific software stack (from 25.11; see the full Spack list below). Newly included: CP2K, DIY, Flux, Kokkos-FFT, Kokkos Tools, Mercury, NWChem, PyTUQ, SALTFM. No longer included: cuSZ, ExaGO, GLVis, GPTune, ParaView, STC, Xyce
- MVAPICH-4 Plus -- MPI library tuned for AWS EFA (MVAPICH 4.1 core, ch4:ofi device over libfabric/EFA)
- SLURM 23.11.4 workload manager + MUNGE 0.5.16 authentication (replaces TORQUE)
- CUDA 13.3 system toolkit, cuDNN 9.28.1, NCCL 2.30.7, NVSHMEM 3.8.0, NVIDIA open driver 620.70.06, gcc 13.4.0
- AWS EFA installer 1.49.0 (libfabric 2.4.0)
- Python 3.12.14, Julia 1.13.3, Go 1.27.4, Node.js 26.13.1
- Web-based remote desktop (replaces TurboVNC + noVNC)
- TAU Commander 1.7.1 (new)
- SALT-FM 0.6.0 -- LLVM-based source analysis toolkit for performance instrumentation (new, via Spack)
- k3s is not included in this release
AI/ML stack (unified Python env):
- PyTorch 2.14.1 + torchvision 0.29.0 + torchaudio 2.11.0
- TensorFlow 2.21.0, Keras 3.15.1, JAX 0.12.0.1
- vLLM 0.33.0, SGLang 0.5.2, NIXL 1.5.2, FlashInfer 0.7.2
- OpenCV 5.0.0, CuPy 14.1.1, mpi4py 4.1.2, RDKit 2025.9.6
- Hugging Face Transformers 5.20.2, HF Hub 1.33.0, Tokenizers 0.23.3
- NVIDIA NeMo (nemo-run 0.12.0, nemo-evaluator 0.2.9, nemo-lens 0.2.0) + Megatron-Core 0.20.1
- NVIDIA BioNeMo (bionemo-ir 0.2.1)
- Flash-Attention 2.8.3, Transformer Engine 2.20.2, Mamba-SSM 2.3.1
- Ray 2.56.1, Gradio 3.36.1, Streamlit 1.67.0, LangChain 1.5.0, OpenAI SDK 2.54.0, Anthropic SDK 2.8.0, Agent Development Kit (ADK) 2.14.0, Weights & Biases 0.31.0, TensorBoard 2.21.0
- XGBoost 3.4.1, PyTorch Monarch 0.7.0
- JupyterLab 4.6.6 + Notebook 7.6.5, Marimo 0.25.1
- NumPy 2.3.5, SciPy 1.18.1, Pandas 2.3.3, Polars 2.1.0, Scikit-Learn 1.9.1, SymPy 1.14.0, GeoPandas 1.2.0
- Matplotlib 3.11.2, Seaborn 0.13.2, Plotly 7.1.0
AI agent tooling:
- Ollama 0.40.16 (local LLM inference)
- Hermes Agent v0.22.2 (CLI + web dashboard; new desktop launcher)
- Orca IDE 1.4.255 (agentic development IDE)
- OpenClaw 2026.11.5 (agentic AI framework)
Containers:
- Docker 29.10.0, Podman 5.7.0, Singularity-CE 4.6.0
- NVIDIA Container Toolkit 1.20.1, Podman-HPC 1.2.2
IDE:
- VSCodium 1.143.17232
The full list of E4S applications installed via Spack is as follows:
-- linux-ubuntu26.04-aarch64 / gcc@13.4.0 --------------
- adios@1.13.1
- adios2@2.12.1
- alquimia@1.1.0
- aml@0.2.1
- amrex@26.10
- arborx@2.1
- argobots@1.2
- ascent@0.9.5
- axom@0.15.0
- boost@1.90.0
- butterflypack@4.1.0
- cabana@0.8.0
- caliper@2.15.0
- chai@2026.07.0
- chapel@2.10.0
- charliecloud@0.44
- conduit@0.9.8
- cp2k@2026.2
- darshan-runtime@3.5.0
- darshan-util@3.5.0
- datatransferkit@3.1.1
- diy@3.6.0
- dyninst@13.0.0
- e4s-alc@1.0.3
- e4s-cl@1.0.9
- faodel@1.2108.1
- fftx@1.2.0
- flecsi@2.4.2
- flit@2.1.0
- flux-core@0.89.0
- fpm@0.13.0
- gasnet@2025.8.0
- ginkgo@1.11.0
- globalarrays@5.8.2
- gmp@6.3.0
- gotcha@1.0.8
- gromacs@2026.3
- guacamole-client@1.6.0
- guacamole-server@1.6.0
- h5bench@1.4
- hdf5-vol-async@1.7
- hdf5-vol-cache@v1.1
- hdf5-vol-log@1.4.0
- hdf5@1.14.6
- heffte@2.4.1
- hpctoolkit@2026.0.1
- hpx@1.11.0
- hypre@3.2.0
- kokkos-fft@2.0.0
- kokkos-kernels@5.2.2
- kokkos-tools@5.2.0
- kokkos@5.2.2
- laghos@3.1
- lammps@20250722.6
- lbann@0.104
- legion@26.09.1
- libcatalyst@2.1.0
- libceed@1.0.0
- libnrm@0.1.0
- libpressio@0.99.4
- libquo@1.4.1
- libunwind@1.8.3
- loki@0.1.7
- magma@2.10.0
- mercury@2.4.1
- metall@0.30
- mfem@4.10.0
- mgard@1.6.0
- mpark-variant@1.4.0
- mpifileutils@0.12
- nccmp@1.11.0.0
- nco@5.4.1
- nek5000@19.0
- nekbone@17.0
- netcdf-fortran@4.6.4
- netlib-scalapack@2.2.3
- nrm@0.1.0
- nwchem@7.3.1
- omega-h@11.2.0-scorec
- openfoam@2606
- openmpi@5.0.10
- openpmd-api@0.17.1
- papi@7.2.0
- papyrus@1.0.2
- parallel-netcdf@1.14.1
- parsec@4.0.2411
- pdt@3.25.2
- petsc@3.26.0
- phist@1.12.1
- plasma@24.8.7
- plumed@2.10.1
- precice@3.4.0
- pruners-ninja@1.0.2
- pumi@2.2.9
- py-cinemasci@1.7.9
- py-h5py@3.16.0
- py-jupyterhub@1.4.1
- py-libensemble@1.6.1
- py-petsc4py@3.26.0
- py-pytuq@1.0.0
- qthreads@1.18
- quantum-espresso@7.6
- raja@2026.07.0
- rempi@1.1.0
- saltfm@0.6.0
- scr@3.1.0
- slate@2025.05.28
- slepc@3.26.0
- strumpack@8.1.0
- sundials@7.9.0
- superlu-dist@9.2.1
- superlu@7.0.1
- swig@4.0.2-fortran
- sz@2.1.12.5
- sz3@3.2.0
- tasmanian@8.1
- tau@2.36
- tomcat@9.0.96
- trilinos@14.4.0
- trilinos@17.2.1
- turbine@1.3.0
- umap@2.1.1
- umpire@2026.07.1
- upcxx@2025.10.0
- veloc@1.7
- vtk-m@2.3.0
- wannier90@4.0.2
- warpx@26.09
- wps@4.6.0
- wrf@4.7.1
- zfp@1.0.1
Additional details
Usage instructions
The 1-Click Security Group opens port 22 only so that you can access your instance via SSH using login 'ubuntu', you may change this later.
For software development and basic usage:
- Launch the ParaTools Pro for E4S (TM) AMI via 1-Click
- On the 'EC2 Launch an Instance' page pick the key pair you will use to login
- On the 'EC2 Launch an Instance' page optionally edit the network settings by pressing the edit button. Adjust the firewall rules if needed to ensure ssh access and enable Auto-assign public IP if you plan to access the instance remotely from a non-AWS IP address.
- Click 'Launch Instance'
- Find your running instance in the EC2 Instances section of the EC2 dashboard, and connect to the instance via SSH using the key pair you previously selected by picking the instance and pressing the connect button.
For more advanced usage, including launching a ParaTools Pro for E4S (TM) cluster with the Heidi AI Cloud Supercomputer, and submitting multi-node jobs please see:
Support
Vendor support
For general support questions, please email support@paratools.com
Paid support contracts, custom AMIs, and computing environments are available. Please see https://paratoolspro.com/ for additional details.
AWS infrastructure support
AWS Support is a one-on-one, fast-response support channel that is staffed 24x7x365 with experienced and technical support engineers. The service helps customers of all sizes and technical abilities to successfully utilize the products and features provided by Amazon Web Services.