Folding at home

XDA-Developers made the point this week that most gamers are running an idle 300W GPU for 20 hours a day. It renders one thing at 8 PM, sits at 20W the rest of the time, and eats the depreciation of a $600 card doing nothing. Volunteer distributed computing has been the answer for a decade: donate the idle cycles to a research project that needs them. The projects are real (protein folding for drug discovery, gravitational wave analysis, cancer screening), the clients are mature, and the modern GPUs on them chew through work that used to take supercomputers weeks.

We tested 7 desktop apps that route your idle GPU (and, on some, CPU) to a research project. The list covers the two giant middleware clients (BOINC, Folding@home), the IBM-founded humanitarian grid (World Community Grid), gravitational wave and pulsar work at Einstein@Home, the protein-design project that Rosetta powers, the umbrella scheduler for BOINC newcomers (Science United), and the GPU-only physics grid (GPUGRID). Every option runs on Windows, macOS, or Linux; most support all three.

What to look for in a distributed compute client

Quick comparison

App Best for OS Uses GPU Cost
Folding@home Protein folding, largest home compute network Windows, macOS, Linux Yes (CUDA, OpenCL) Free
BOINC Middleware for dozens of projects Windows, macOS, Linux Depends on project Free
World Community Grid Humanitarian and health research Windows, macOS, Linux (via BOINC) Some projects Free
Einstein@Home Gravitational waves and pulsar detection Windows, macOS, Linux (via BOINC) Yes (CUDA, OpenCL) Free
Rosetta@home Protein design Windows, macOS, Linux (via BOINC) CPU primarily Free
Science United BOINC scheduler for newcomers Windows, macOS, Linux (via BOINC) Depends Free
GPUGRID GPU-only molecular dynamics Windows, Linux (via BOINC) Yes (CUDA) Free

1. Folding@home — best overall

Folding@home simulates how proteins fold, which underpins drug discovery for cancer, Alzheimer’s, and infectious disease. It’s the largest home compute network in the world; at the pandemic peak it briefly exceeded the top 500 supercomputers combined by a factor of several. The GPU client is mature, hits high utilisation on modern Nvidia and AMD cards, and pauses automatically when it detects an active user or a full-screen game.

Where it falls short: Only one project (protein simulation). No CPU-heavy alternative work when the GPU queue drains, though that is rare.

Pricing:

Platforms: Windows, macOS, Linux

Download: foldingathome.org

Bottom line: Best if you want the largest impact per idle hour with the least configuration.

2. BOINC — best middleware

BOINC (Berkeley Open Infrastructure for Network Computing) isn’t a project itself; it’s the client that talks to dozens of them. Attach to one or several, set a resource share, and BOINC schedules work between them. The classic use is running Einstein@Home, Rosetta@home, and MilkyWay@home from a single dashboard.

Where it falls short: UI feels old. First-run setup (attach to a project, get an account key, verify credit) has friction that Folding@home doesn’t.

Pricing:

Platforms: Windows, macOS, Linux, Android

Download: boinc.berkeley.edu/download.php

Bottom line: Best when you want to donate to more than one project or switch projects without changing clients.

3. World Community Grid — best humanitarian focus

World Community Grid was IBM-founded, now Krembil Research Institute-run, and coordinates humanitarian research: cancer screening (OpenPandemics), rainfall prediction for climate-vulnerable regions (Africa Rainfall Project), and tuberculosis drug discovery. Runs on top of BOINC, so once BOINC is installed, attaching to WCG is one form submission.

Where it falls short: Not every WCG project has GPU work units; several are CPU-only, which means your modern gaming card sits partly idle. Turnover between projects varies.

Pricing:

Platforms: Windows, macOS, Linux, Android (via BOINC)

Download: worldcommunitygrid.org

Bottom line: Best when the cause matters as much as the throughput.

4. Einstein@Home — best GPU utilisation on physics

Einstein@Home analyses data from LIGO (gravitational waves), the Fermi telescope (pulsars), and radio telescope surveys, hunting for signals in noise. GPU work units run on CUDA and OpenCL and use the card as hard as anything on this list, which is the point if the goal is “put that GPU to work.”

Where it falls short: Runs through BOINC; setup carries the same friction. Some GPU tasks generate meaningful heat, so check thermals on tight cases.

Pricing:

Platforms: Windows, macOS, Linux (via BOINC)

Download: Attach via einsteinathome.org

Bottom line: Best if you want a GPU-heavy load on a physics project with clear results.

5. Rosetta@home — best CPU work on protein design

Rosetta@home designs new proteins: enzymes, antibodies, vaccine candidates. Work is CPU-heavy, so it complements a machine that already runs Folding@home or GPUGRID on the GPU without competing for the same silicon.

Where it falls short: Mostly CPU, so a modern gaming GPU sits idle unless you pair it with a GPU project too. Some project cycles run slower than others.

Pricing:

Platforms: Windows, macOS, Linux (via BOINC)

Download: Attach via boinc.bakerlab.org/rosetta

Bottom line: Best CPU-side companion to a GPU-heavy project.

6. Science United — best for newcomers

Science United is a BOINC front-end that skips the project-picking. Sign up, pick broad science areas (biomedical, physics, climate), and Science United attaches your machine to the projects that fit and need volunteers most.

Where it falls short: Less control if you have a specific project you want to support. Still requires BOINC underneath.

Pricing:

Platforms: Windows, macOS, Linux (via BOINC)

Download: scienceunited.org

Bottom line: Best if you want to donate cycles but don’t want to research projects yourself.

7. GPUGRID — best pure GPU workload

GPUGRID runs molecular dynamics simulations for biomedical research (viral proteins, drug binding), entirely on the GPU. If Folding@home is oversubscribed at peak, GPUGRID is where you go for a fresh queue of GPU-only work.

Where it falls short: CUDA-first; AMD and Intel GPU support has lagged. Windows and Linux only. Project’s activity level has varied over the years.

Pricing:

Platforms: Windows, Linux (via BOINC)

Download: Attach via gpugrid.net

Bottom line: Best for CUDA-heavy GPUs looking for a second project after Folding@home.

How to pick the right one

If you want maximum impact per idle hour with one install: Folding@home.

If you want more than one project: BOINC with Einstein@Home (GPU) and Rosetta@home (CPU) attached.

If the cause matters most: World Community Grid through BOINC.

If you want BOINC to pick for you: Science United.

If you have an Nvidia card and want pure GPU work: GPUGRID, alongside Folding@home.

Skip distributed computing entirely if the machine is battery-powered or on a metered network. Both categories cost you more than they contribute.

FAQ

Will running Folding@home or BOINC damage my GPU?

Modern GPUs handle sustained compute loads for years without measurable degradation, provided the card stays within its rated temperatures. Watch thermals for the first few days and set a fan curve if idle temps rise. Bigger risk is dust build-up over months; clean the case.

How much does this add to my electricity bill?

A 250W GPU running 12 hours a day at $0.15/kWh adds around $12/month. Halve that if you undervolt and cap the client’s runtime.

Can I game while Folding@home runs?

Yes. The clients pause GPU work when they detect a full-screen 3D application or active input. You’ll see the fold pause within seconds of launching a game.

Which project produces the most research output for a given hour of GPU time?

That varies by project cycle and hardware. Folding@home consistently ranks near the top for GPU throughput on modern Nvidia and AMD cards. Einstein@Home and GPUGRID are competitive for CUDA-heavy loads.

Do I need an Nvidia card?

No. Folding@home, Einstein@Home, and most others support OpenCL on AMD and Intel GPUs. Nvidia’s CUDA path is often the fastest on a given tier because the clients are tuned there first.

Is my contribution actually anonymous?

You register a username (real name optional) and your contribution appears on public leaderboards. The work itself is anonymous science; the credit is public. If you want no visibility at all, use a pseudonym at signup.