GpuInfo

GpuInfo possess 2 different behaviours depending or the supplied arguments.

The related code is in gpuhelper.hpp

GpuInfo Argument and return type 1

This case is what happens when no argument is given : vship.GpuInfo()

This mode allows to list the available GPUs. Beware to manage Errors as this function can return 2 errors: DeviceCountError and NoDeviceDetected

Here is the return format:

GPU 0: {GPU 0 Name}
GPU 1: {GPU 1 Name}
...

GpuInfo Argument and return type 2

This case is what happens when one argument is given : vship.GpuInfo(gpu_id = x)

Note that by default, HIP and CUDA will use GPU 0

gpu_id : integer

In that case, vship understands that you want more details about the GPU specified. The same as before, beware to manage Errors. This function can return 3 error types : DeviceCountError, NoDeviceDetected and BadDeviceArgument

Here is the return format and all the infos that vship will give :

Name: {GPU Name string}
MultiProcessorCount: {multiprocessor count integer}
ClockRate: {clockRate float} Ghz
MaxSharedMemoryPerBlock: {Max Shared Memory Per Block integer} bytes
WarpSize: {Warp Size integer}
VRAMCapacity: {GPU VRAM Capacity float} GB
MemoryBusWidth: {memory bus width integer} bits
MemoryClockRate: {memory clock rate float} Ghz
Integrated: {0|1}
PassKernelCheck : {0|1}

PassKernelCheck is special because it is not given by HIP or CUDA, it corresponds to a Sanity Check and gives clues showing that vship will fail if it gives 0 and the given gpu is used to compute. The corresponding error that would arise if unmanaged at that step is BadDeviceCode