Skip to main content

Multi GPU

The Multi GPU feature leverages multiple GPU’s on the cloud in order to either run bigger wave based simulation tasks than was possible before or run small to medium sized wave based tasks faster.

note

The multi GPU feature is only available to subscribers with a premium tier subscription

The feature is automatic in the sense that the user can’t manually activate it. Before a simulation can be started, Treble runs a preprocessing check to determine whether the simulation will benefit from the feature. If the check is positive the simulation will take advantage of it, given that the subscription method is of a premium or enterprise tier. The logic for the check is as follows:

  1. Check whether the simulation fits on 1 GPU (previous limit). If not, check if it fits on 2 or 4 GPU’s. If it does, then the simulation is run on that number of GPU’s. If it doesn’t fit on 4 GPU’s the transition frequency has to be lowered.
  2. If the simulation does however fit on 1 GPU another check is made to see if it would run faster on 2 GPU’s. If it does, the simulation is run on 2 GPU’s. If there is no speed improvement it will remain on a single GPU.

To sum it up, the feature enables wave based simulations with up to four times as many mesh elements than was possible before, enabling much higher transition frequencies for simulations. For simulations with mesh sizes that fit on a single GPU the speed of the wave based solver can improve up to 50%, compared to what was previously possible to achieve in Treble.