BOINC Projects

MilkyWay@home


MilkyWay@home is a volunteer computing project operated by Rensselaer Polytechnic Institute (RPI). Running on the BOINC platform, it harnesses idle processing power donated by volunteers around the world to study the structure and formation history of the Milky Way galaxy, with particular focus on the galactic stellar halo, tidal debris streams, and the distribution of dark matter.[1][2]

MilkyWay@home
A dwarf galaxy being disrupted by the Milky Way's gravity
Project
StatusActive
CategoryAstrophysics
ComputeCPU, GPU
Development
DeveloperHeidi Jo Newberg, Travis Desell, Carlos Varela
SponsorRensselaer Polytechnic Institute
MaintainerMilkyWay@home team
Initial releaseApril 1, 2007  (19 years ago)
Repositoryhttps://github.com/Milkyway-at-home
Software
Written inC, C++, OpenCL, CUDA
Operating systemWindows, Linux, macOS
Size~50 MB
BOINC statistics
Stats as ofMay 21, 2026  (0 years ago)
Performance200817.71 GigaFLOPS
Active users19,120
Total users257,894
Active hosts61,248
Total hosts1,543,021
Analytics
RAC12,400,000
Credit/day730,000
Metadata
Websitehttps://milkyway.cs.rpi.edu/milkyway/
LicenseGNU GPL v3

The project is a collaboration between RPI's Department of Computer Science and its Department of Physics, Applied Physics, and Astronomy, and is funded by the National Science Foundation.[3] MilkyWay@home was among the earliest BOINC projects to deploy GPU applications at scale, releasing CUDA support for NVIDIA GPUs in June 2009 and subsequently achieving computing rates that rivalled the world's fastest supercomputers.[4]

History

The roots of MilkyWay@home lie in the individual research of Heidi Jo Newberg, then an associate professor of physics, applied physics, and astronomy at RPI. Her work to map the three-dimensional distribution of stars and matter in the Milky Way using data from the Sloan Digital Sky Survey (SDSS) required computational resources far beyond what a single research group could provide. As Newberg later explained, she faced "a very big computational problem to solve and very little personal computational power or time" available to her.[5]

Before bringing the project to BOINC, Newberg worked with Malik Magdon-Ismail, associate professor of computer science, to design a faster and more efficient optimization algorithm. Formal development under the BOINC platform began in July 2006, and the project launched publicly on 1 April 2007.[6][7]

The project's first scientific focus was fitting density models to the Sagittarius tidal stream, a massive stellar stream produced by the ongoing gravitational disruption of the Sagittarius Dwarf Spheroidal Galaxy as it orbits the Milky Way. Nathan Cole's 2008 paper in the Astrophysical Journal presented the first major results from this stream-fitting work, applying maximum likelihood methods to SDSS data.[8]

GPU computing milestone

In June 2009, MilkyWay@home released CUDA applications for a broad range of NVIDIA GPUs, following earlier experimental releases in a separate GPU fork of the project. The effect on computing throughput was dramatic. In mid-June 2009 the project was operating at 31.7 teraFLOPS with approximately 24,000 registered users across 149 countries. By 12 January 2010 the average throughput had risen to 1,382 teraFLOPS, a figure that would have placed MilkyWay@home second on the TOP500 list of supercomputers at that time, with 44,900 users across 170 countries.[9] A press release from RPI confirmed that on 26 January 2010 the project surpassed 1 petaFLOP of sustained computing power, making it at the time the fastest computing project on the BOINC platform and the second fastest public distributed computing program in operation, behind only Folding@home.[10]

At its peak, shortly after GPU code was released, MilkyWay@home delivered approximately 2 petaFLOPS of computing power. By 2012 the project was running at around 0.5 petaFLOPS, which still represented processing power equivalent to the 45th fastest supercomputer in the world at that time.[11]

An OpenCL application for AMD Radeon GPUs was also made available, allowing volunteers with AMD hardware to contribute at similarly high throughput rates.[12]

Scientific objectives

The central scientific mission of MilkyWay@home is to build an accurate, three-dimensional model of the Milky Way's stellar halo and to constrain the distribution of dark matter within it. The strategy relies on analyzing the orbits and disruption patterns of dwarf satellite galaxies, whose tidal debris encodes information about the gravitational potential -- and thus the dark matter content -- of the Milky Way.[13]

The project runs two complementary types of applications. The first fits the spatial density profile of stellar tidal streams using statistical photometric parallax applied to main sequence turnoff (MSTO) stars from the SDSS. Each volunteer's computer evaluates the likelihood of one set of density model parameters given the observed positions and magnitudes of turnoff stars, contributing one step in a global parameter optimization. The second application, the N-body subproject, simulates full dwarf galaxies being disrupted within a model of the Milky Way's gravitational field, iterating initial conditions until the simulated tidal debris matches observed stellar structures.[14]

Formally, the stream fitting procedure seeks to maximize the log-likelihood:

ln⁡ℒ=∑i=1Nln⁡[Φ⋅fs(𝐱i)+(1−Φ)⋅fb(𝐱i)C]

where Φ is the stream fraction, fs and fb are the stream and background density functions, and the sum runs over all N observed turnoff stars in an SDSS stripe.[15] Goodness-of-fit for the N-body application is measured using an Earth-Mover Distance method applied to stellar densities along the stream.[16]

Research areas addressed by the project include:

  • The three-dimensional density profile of the Milky Way stellar halo
  • The Sagittarius tidal stream and its bifurcated components
  • The Virgo Overdensity and other halo substructures
  • The Orphan-Chenab Stream and its progenitor dwarf galaxy
  • Dwarf galaxy progenitor masses and dark matter fractions
  • The gravitational potential of the Milky Way and implied dark matter halo shape

Methods

 
An artist's rendering of the Milky Way galaxy, whose structure MilkyWay@home aims to map and model

Stream density fitting

MilkyWay@home uses SDSS photometric data to measure the spatial density of F-type main sequence turnoff stars in the galactic halo. The sky is divided into wedges of approximately 2.5 degrees width, and for each wedge the volunteer application evaluates a density model consisting of a smooth spheroid background component plus one or more tidal stream components. Self-optimizing evolutionary algorithms, including differential evolution and particle swarm optimization methods, are used to find the best-fit parameters across the high-dimensional parameter space.[17][18]

N-body simulations

The N-body subproject creates simulated dwarf galaxies and evolves them within a model of the Milky Way's gravitational potential. The simulation adjusts the initial conditions of the dwarf galaxy -- its mass, concentration, orbital parameters, and dark matter fraction -- until the resulting tidal debris pattern matches observed stellar stream structures. Because a single N-body simulation can require hours of CPU time, distributing thousands of simulations simultaneously across volunteer computers is essential for exploring the parameter space efficiently.[19]

Optimization algorithms

Because volunteer computing platforms are asynchronous and heterogeneous -- work units are completed on wildly different hardware at unpredictable times -- MilkyWay@home required the development of specially adapted optimization methods. The team published work on robust asynchronous optimization algorithms tailored for this environment, and on validating evolutionary algorithms in the context of volunteer grids.[20][21]

Applications

 
An NVIDIA Tesla GPU of the kind used by MilkyWay@home volunteers. GPU computing transformed the project's throughput from tens of teraflops to well over a petaflop.

MilkyWay@home applications are distributed through the BOINC client and support Windows, Linux, and macOS. The project has historically compiled its code for sixteen different platforms to support the heterogeneous hardware of its volunteer base.[22]

Supported application types include:

  • CPU applications (32-bit and 64-bit), available since 2008
  • NVIDIA GPU applications using CUDA, released June 2009
  • AMD GPU applications using OpenCL
  • Multi-threaded N-body simulation applications

Work units originally required 2 to 4 hours of computation on a modern CPU, with a three-day deadline. By early 2010, the project was routinely issuing much larger units requiring 15 to 20 hours of CPU time, valid for approximately one week. By 2018, GPU-based tasks had become so optimized that many could be completed in under a minute on a high-end graphics card.[23]

The project does not include a traditional graphical screensaver. Instead, animations of the best N-body simulations are shared via YouTube.[24]

MilkyWay@home is a whitelisted Gridcoin project and is reported to be the second-largest generator of Gridcoins among BOINC projects.[25]

Project team and funding

 
The Sagittarius Dwarf Spheroidal Galaxy (center), the primary source of the tidal stream first studied by MilkyWay@home, imaged in the infrared by the 2MASS survey

MilkyWay@home is a collaboration between RPI's Department of Computer Science and Department of Physics, Applied Physics, and Astronomy. The project team has included:

  • Heidi Jo Newberg (astrophysics lead)
  • Travis Desell (volunteer computing and GPU optimization)
  • Malik Magdon-Ismail (optimization algorithms)
  • Carlos Varela (distributed computing infrastructure)
  • Boleslaw K. Szymanski (distributed systems)
  • Matthew Newby
  • Nathan Cole
  • Kevin Roux
  • Hiroka Warren

The project is supported by the National Science Foundation under Grant Numbers 0612213, 0607618, 0448407, 1009670, 1615688, and 1908653.[26] Individual grants include:

  1. NSF Grant #1908653
  2. Charting the Structure of the Milky Way Stellar Halo and Disk, NSF AST Grant #1615688, 09/15/2016 -- 08/31/2019, Principal Investigator: Heidi Jo Newberg
  3. Stars and Dark Matter in the Halo of the Milky Way, NSF AST Grant #1009670, started 09/15/2010
  4. Data-Driven Discovery of the Milky Way Origin and Evolution from the Sloan Digital Sky Survey, NSF IIS Grant #0612213
  5. Middleware and Programming Technology for Grid Computing, NSF CAREER Grant #0448407

BOINC statistics

As of May 2026, MilkyWay@home remains one of the largest and most active astronomy-focused projects on the BOINC platform, with over 257,000 total registered users across more than 1.5 million total hosts. At its peak throughput following the GPU release, the project delivered approximately 2 petaFLOPS of sustained computing power. The project has maintained a strong and active GPU cruncher community due to the efficiency of its OpenCL applications and characteristically high credit generation rates.[27]

Scientific publications

Astrophysics journal papers

  1. Eric J. Mendelsohn. Using MilkyWay@home to Measure the Mass of the Orphan-Chenab Stream Progenitor Dwarf Galaxy. PhD thesis. Rensselaer Polytechnic Institute, 2022.
  2. Eric J. Mendelsohn, Heidi Jo Newberg, Siddhartha Shelton, Lawrence M. Widrow, Jeffery M. Thompson, Carl J. Grillmair. Estimate of the Mass and Radial Profile of the Orphan-Chenab Stream's Dwarf-galaxy Progenitor Using MilkyWay@home. The Astrophysical Journal, 2022. doi:10.3847/1538-4357/ac498a
  3. Siddhartha Shelton et al. An Algorithm for Reconstructing the Orphan Stream Progenitor with MilkyWay@home Volunteer Computing. The Astrophysical Journal, 2021.
  4. Heidi Jo Newberg et al. Streams and the Milky Way Dark Matter Halo. International Astronomical Union, 2020.
  5. Jake Weiss, Heidi Jo Newberg, Travis Desell. A Tangle of Stellar Streams in the North Galactic Cap. The Astrophysical Journal, 2018.
  6. Jake Weiss, Heidi Jo Newberg, Matthew Newby, Travis Desell. Fitting the Density Substructure of the Stellar Halo with MilkyWay@home. The Astrophysical Journal Supplement Series, 2018.
  7. Heidi Jo Newberg, Matthew Newby, Travis Desell, Malik Magdon-Ismail, Boleslaw Szymanski, Carlos Varela. MilkyWay@home: Harnessing Volunteer Computers to Constrain Dark Matter in the Milky Way. Proceedings of the International Astronomical Union, 2014.
  8. Matthew T. Newby. The Sagittarius Tidal Stream and the Shape of the Galactic Stellar Halo. PhD thesis. Rensselaer Polytechnic Institute, 2013.
  9. Matthew Newby, Nathan Cole, Heidi Newberg, Travis Desell, Malik Magdon-Ismail, Boleslaw Szymanski, Carlos Varela, Benjamin Willett, Brian Yanny. Spatial Characterization of the Sagittarius Dwarf Galaxy Tidal Tails. Astronomical Journal, 2013.
  10. Benjamin Arthur Willett. Simultaneous Orbit Fitting of Stellar Streams: Constraining the Galactic Dark Matter Halo. PhD thesis. Rensselaer Polytechnic Institute, 2010.
  11. Nathan Cole et al. Maximum Likelihood Fitting of Tidal Streams with Application to the Sagittarius Dwarf Tidal Tails. Astrophysical Journal, 2008.

Computer science and volunteer computing papers

  1. Travis Desell, Anthony Waters, Malik Magdon-Ismail, Boleslaw Szymanski, Carlos Varela, Matthew Newby, Heidi Newberg, Andreas Przystawik, Dave Anderson. Accelerating the MilkyWay@Home Volunteer Computing Project with GPUs. PPAM 2009. Published in Lecture Notes in Computer Science, vol. 6067, Springer, 2010. doi:10.1007/978-3-642-14390-8_29
  2. Travis Desell et al. Robust Asynchronous Optimization for Volunteer Computing Grids. IEEE e-Science 2009.
  3. Travis Desell et al. Validating Evolutionary Algorithms on Volunteer Computing Grids. DAIS 2010.
  4. Travis Desell et al. An Analysis of Massively Distributed Evolutionary Algorithms. IEEE CEC 2010.
  5. Nathan Cole et al. A Study of the Sagittarius Tidal Stream Using Maximum Likelihood. ADASS XVIII, 2009.
  6. Carlos Varela. Enabling Synchronous Computation on Volunteer Computing Grids. The 4th Pan-Galactic BOINC Workshop, 2008.

See also

External links

References

  1. ↑ MilkyWay@home. Rensselaer Polytechnic Institute. Retrieved 2026-05-21.
  2. ↑ Newberg, Heidi Jo.(2014).MilkyWay@home: Harnessing Volunteer Computers to Constrain Dark Matter in the Milky Way. Proceedings of the International Astronomical Union. pp. 98--104.
  3. ↑ Project Information. MilkyWay@home. Retrieved 2026-05-21.
  4. ↑ (2010})."Accelerating the MilkyWay@Home Volunteer Computing Project with GPUs".DOI: 10.1007/978-3-642-14390-8_29.
  5. ↑ Template:Cite news
  6. ↑ Template:Cite news
  7. ↑ MilkyWay@home. Rensselaer Polytechnic Institute. Retrieved 2026-05-21.
  8. ↑ Cole, Nathan.(2008).Maximum Likelihood Fitting of Tidal Streams with Application to the Sagittarius Dwarf Tidal Tails. The Astrophysical Journal. pp. 750--766.
  9. ↑ MilkyWay@home. HandWiki. Retrieved 2026-05-21.
  10. ↑ Template:Cite news
  11. ↑ Newberg, Heidi Jo.(2014).MilkyWay@home: Harnessing Volunteer Computers to Constrain Dark Matter in the Milky Way. Proceedings of the International Astronomical Union. pp. 98--104.
  12. ↑ MilkyWay@home. Wikipedia. Retrieved 2026-05-21.
  13. ↑ Newberg, Heidi Jo.(2014).MilkyWay@home: Harnessing Volunteer Computers to Constrain Dark Matter in the Milky Way. Proceedings of the International Astronomical Union.
  14. ↑ Newberg, Heidi Jo.(2014).MilkyWay@home: Harnessing Volunteer Computers to Constrain Dark Matter in the Milky Way. Proceedings of the International Astronomical Union.
  15. ↑ Cole, Nathan.(2008).Maximum Likelihood Fitting of Tidal Streams with Application to the Sagittarius Dwarf Tidal Tails. The Astrophysical Journal. pp. 750--766.
  16. ↑ Newberg, Heidi Jo.(2014).MilkyWay@home: Harnessing Volunteer Computers to Constrain Dark Matter in the Milky Way. Proceedings of the International Astronomical Union.
  17. ↑ Weiss, Jake.(2018).Fitting the Density Substructure of the Stellar Halo with MilkyWay@home. The Astrophysical Journal Supplement Series.
  18. ↑ (2009})."Robust Asynchronous Optimization for Volunteer Computing Grids".link.
  19. ↑ Shelton, Siddhartha.(2021).An Algorithm for Reconstructing the Orphan Stream Progenitor with MilkyWay@home Volunteer Computing. The Astrophysical Journal.
  20. ↑ (2010})."Validating Evolutionary Algorithms on Volunteer Computing Grids".link.
  21. ↑ (2010})."An Analysis of Massively Distributed Evolutionary Algorithms".
  22. ↑ Newberg, Heidi Jo.(2014).MilkyWay@home: Harnessing Volunteer Computers to Constrain Dark Matter in the Milky Way. Proceedings of the International Astronomical Union.
  23. ↑ MilkyWay@home. Wikipedia. Retrieved 2026-05-21.
  24. ↑ MilkyWay@home. Wikipedia. Retrieved 2026-05-21.
  25. ↑ MilkyWay@home. Wikipedia. Retrieved 2026-05-21.
  26. ↑ Project Information. MilkyWay@home. Retrieved 2026-05-21.
  27. ↑ Server Status. MilkyWay@home. Retrieved 2026-05-21.
Retrieved from "https://boincsynergy.ca/wiki/index.php?title=MilkyWay@home&oldid=1580"