I am thinking of building a new workstation with 2 GPU's for compute purposes (FP64 >= 1/4 FP32). The main ways I am thinking of using this heterogeneous compute workstation is :
- Use the two GPUs separately for compute loads by programming Cpu+AMD card using OpenCL and CPU+NVIDIA card using CUDA
- Use the two cards together with CPU for compute loads by using AMDs new portable HIP API infrastructure.
Does anyone know if AMD and NVIDIA GPUs can be used in the above two usage ways ?
- Use the two GPUs separately for compute loads by programming Cpu+AMD card using OpenCL and CPU+NVIDIA card using CUDA
- Use the two cards together with CPU for compute loads by using AMDs new portable HIP API infrastructure.
Does anyone know if AMD and NVIDIA GPUs can be used in the above two usage ways ?