Skip to main content

Eagle vs. Peregrine Supercomputer Stack-Up

See how NREL's newest high-performance computing system, Eagle, compares to its predecessor, Peregrine.


Peak Performance

a timer that displays 55

In just under a minute (54.4 seconds), Eagle can do as many calculations as there have been seconds in the universe (4.352e17 seconds).

2.24 petaflops (million-billion calculations per second)

8.0 petaflops (3.5X higher performance)


Individual Servers

If compute nodes were cars, Eagle’s cars are so much more efficient at transporting people than Peregrine's that Eagle takes 478 cars off the road.

an illustration of cars on a road

2,592 compute nodes

2,114 compute nodes


Modern laptops have an average of at least 4-8 cores. So, Eagle has as many cores as 19,000 cutting-edge laptops combined.

58,752 cores

76,104 cores

an illustration a computer circuit board

Random Access Memory

More RAM means more applications can run simultaneously.

an illustration of a computer chip

A typical web browser tab uses 50-200 MB of RAM. Assuming an average RAM usage per tab of 125 MB (a typical YouTube video), a computer with Eagle's system memory could have 2,368,000 loaded tabs open at the same time.

145,408 GB

296,000 GB, or 296 terabytes

High Speed Data Storage

Where Information is saved.

To match the same amount of (high-speed data) storage, you'd need 28,000 500-GB laptops—which would stack 1,423 feet, or 433.8 meters, high.

Illustration of a skyscraper

2.25 petabytes, or 2,250,000 GB

14 petabytes, or 14,000,000 GB

Network Speed

Frequency of communication between nodes.

If GB were people and Eagle were a grocery store, Eagle could simultaneously check out almost double the amount of people waiting in line than Peregrine could.

56 gigabytes (GB) per second

100 GB per second

an illustration shopping carts
an illustration of a coffee pot

Energy-per-computation Efficiency

Using the same amount of energy as Peregrine—about the same amount needed to heat water in 1,000 coffee pots—Eagle can perform 3.5 times the number of computations.

3.2e9 calculations per watt

8e9 calculations per watt