Error when running transfer learning model using GPU in MacPro Intel

I am running Geneformer, a transfer learning model for scientific research and got into this error: torch not compiled with CUDA enabled. I would appreciate any help on that! Thanks. I am running this model locally on my MacPro Intel processor with GPU build-in. Graphics: Intel HD Graphics 630. Open for more discussion…

CUDA requires an NVIDIA GPU to run. To my knowledge it has been many years since Apple last offered a Mac with an NVIDIA GPU. The last CUDA version that supported macOS shipped four years ago.

Based on my understanding, it seems that this model necessitates an NVIDIA GPU for optimal performance, rather than any other graphics card. Additionally, I’d like to note that the MacPro I’m currently using is approximately four years old and features an Intel processor. Initially, attempts to run the model on a Silicon Apple chip proved incompatible.

Considering this, if I were to connect my MacPro Intel to an eGPU, specifically the OWC Mercury Helios FX, which houses an NVIDIA card (GeForce GTX 980 Ti), do you anticipate compatibility? Moreover, could you please advise on how I can confirm whether the card supports Thunderbolt connectivity?

Your guidance is greatly appreciated as I explore the viability of executing the model on my Mac with the assistance of this external device.

I am not familiar with the products and technologies mentioned and as a result I do not know the answer to any of these questions.

Likewise with njuffa, I have no experience with your situation, but offer this for what it’s worth, if you haven’t already come across it.

I also found this Helios FX compatibility guide, which doesn’t seem encouraging for the card you mention. Indeed it shows no Mac driver support at all for any of the Nvidia cards they list.