Performance of NVIDIA Earth-2 FourCastNet NIM#

Use this documentation for details about the performance of the NVIDIA Earth-2 FourCastNet NIM.

Evaluation Process#

The FourCastNet NIM’s API streams back forecast data as it is generated, such that the user receives each time-step sequentially. Performance is measured based on the time needed to complete a set of three different forecasts. Each forecast uses an initial condition at 2020-01-01T00:00:00, and requests five output variables ['t2m', 'z500', 't850', 'u10m', 'v10m']. The requested length of each forecast is the following:

  • 6 hour forecast: 6 hour / 1 step

  • 5 day forecast: 120 hour / 20 steps

  • 10 day forecast: 240 hour / 40 steps

Note

Both model profiles that the NIM supports are expected to have similar performance values.

Performance Results#

The performance results appear in the following table. Latency is the total forecast time in seconds, and throughput is the average forecast steps per second.

Important

It is expected that measurements listed can deviate between machine and runs. The performance values depend greatly on the specific hardware configuration the NIM and client are running on.

H100#

6 hour

5 day

10 day

Latency

3.49s

9.83s

16.08s

Throughput

0.29 step/s

2.03 step/s

2.49 step/s

A100#

6 hour

5 day

10 day

Latency

3.53s

9.92s

17.14s

Throughput

0.28 step/s

2.01 step/s

2.33 step/s

L40S#

6 hour

5 day

10 day

Latency

3.86s

14.54s

25.82s

Throughput

0.26 step/s

1.37 step/s

1.55 step/s

RTX6000#

6 hour

5 day

10 day

Latency

3.83s

13.18s

23.09s

Throughput

0.26 step/s

1.52 step/s

1.73 step/s