### Problem to solve with the new feature Currently, during executions on bench or opencarp, when using GPUs for the solver and limpet, data needs to be sent back to the CPUs before every stimulus or clamp_vm event. This becomes a major bottleneck for performance during GPU execution. ### Intended users Unknown ### Further details This feature request addresses the issue discussed during the WP1 meeting on 27 November, Currently, stimuli and the clamp_vm function only have a CPU implementation. Pre-computing them at the start of the simulation or having a GPU implementation (or both) Would allow to remove the need for most CPU-GPU data transfer during computation, improving performance ### Proposal Hi, It was mentioned in the last few meetings that it should be possible to precompute the stimuli, It should be possible to send the precomputation results to the GPUs to apply them during a given time step. Please let me know if you have any further questions about the issue, Thank you! ### Testing Results for computation using precomputed stimuli need to be the same as doing the same computation using the current implementation. ### Links / references None for now