About Multi-GPU Computation

[saffahyjp]

Hi developers,

Thanks for this awesome project again! Recently I am porting my cryptographic application to a machine with four GPUs. Specifically, I have two possible scenarios in mind:

1. Operations (e.g. addition, multiplication, rotation) are sent one at a time, and each operation is processed by all four GPUs before moving to the next operation.
2. There are a lot of (independent and parallelizable) operations to process, so each operation is processed by a single GPU and all four GPUs work simultaneously on different operations.

Since the current examples begin with cudaSetDevice(0) which means the computation is limited to a single GPU, I would like to ask whether it is currently supported (or planned to be supported) to utilize multiple GPUs in parallel, according to either of the two scenarios above? If so, could you please give me some hints on how to implement it?

Thanks a lot in advance!

[Alisah-Ozcan]

Thank you for your kind words and interest in HEonGPU.

HEonGPU currently does not support multi-GPU execution, but we are working on it and plan to release it soon.

This feature won't split a single FHE operation across GPUs. Instead, it enables parallel execution of multiple independent FHE operations on different GPUs, aiming for system-level throughput rather than per-operation speedup.

[saffahyjp]

Hi! Thank you for your information and I'm really glad to hear that. Looking forward to the release!