Closed mikhail-khalizev closed 2 years ago
@mikhail-khalizev Thanks for this feature request 👍 I guess these queries would be possible via the NVML library for Cuda-capable devices. Maybe we can create a wrapper for the ILGPU.Algorithms
package. However, this also requires adding several properties to the CudaAccelerator
class in order to instantiate an NVML context.
Raised PR that adds some NvmlAPI
bindings, and an NvmlDevice
wrapper.
This wrapper supports easy access to:
I wonder if this should be in the scope of this library...
Is there a way to get the percentage of GPU utilized by an application in real-time?
I propose to add the ability to add methods that return the current parameters of the accelerator: