3
$\begingroup$

What are the pros/cons of using external GPUs (e.g., connected through thunderbolt) vs. internal GPUs for machine learning?

$\endgroup$

2 Answers 2

1
$\begingroup$

Here is a good discussion on a similar question on reddit:

https://www.reddit.com/r/MachineLearning/comments/twnz1o/d_which_external_nvidia_gpu_should_i_buy_for/

Essentially, when connecting an external gpu, it depends a lot on the pc you're connecting it to. You need a good cpu and general hardware in that pc, in order for the gpu to function properly. that's why many recommend to build a powerful gpu home computer right away instead of buying an external gpu. this however comes at much higher cost. an advantageous use of external gpus may also be using cloud gpus from services like colab or better ones. It comes at a monthly fee, but may be easier, quicker and cheaper than building your own pc ^^.

$\endgroup$
0
$\begingroup$

External GPUs are often more easy to swap out than internal GPUs. Thus, there will be more options for hardware and software upgrades for external GPUs.

Internal GPUs may have better connections. Thus, moving data is will be less likely to be a constraint (i.e., GPU starvation).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.