External GPUs , is that a good idea for Deep Learning? What’s the experience with it?

Source: Deep Learning on Medium

External GPUs , is that a good idea for Deep Learning? What’s the experience with it?

Source: Engadget

Well eGPU is portable , basically the whole idea of eGpu is to transform ultrabooks that are rather weak in terms of raw power into capable gaming laptops and portable workstations. The benefits of an ultrabook are portability and long battery life. This same laptop can then turn into a much more powerful computer when paired with an external GPU.

Purpose built in gaming laptops have partially filled the need for high performing laptops. However they are often bulky , noisy and has shorter battery life. Currently in a state of 2019 , most of the gaming laptop tend to thinner and powerful but they are not so good as expected, because of the thermal issues, which will not last for longer.

But the reality is using an external GPU doesn’t give you the same performance as it would if you had the same GPU mounted internally. Estimates put the loss at around 10 to 15 percent. That isn’t a big deal, especially considering the monstrous capacity of the latest high-end graphics cards you can buy toady. Furthermore, while a PCIe port can transfer a lot of data very quickly, even the latest Thunderbolt and USB-C ports cannot match that data rate. Your laptop CPU probably wasn’t designed to handle a powerful external GPU, either. Again, it’s not a deal-breaker, but you might notice the effects. This is especially true for older and slower CPUs. Also with the eGPU although you have portability on your hand , on the case of training larger models you have keep you setup on one place for longer period of time, incase if you extra laptop for work then that’s fine. Also keep in mind thunderbolt connection are of different types , older Macbook probably supports only 4 PCIE lanes, if your thunderbolt Type C supports 6 PCIE lanes than it’s better. So these are the things to keep in mind

Source: Razer

Despite the drawbacks, eGPUs do work. You will get better graphics performance out of your laptop and they’ll enable you to train your model. There are plenty of benchmarks showing that external GPUs provide a huge boost in graphics power, especially in MacBooks. Also, NVIDIA no longer distributes official drivers to Apple MacOS, which is no longer offically compatible with NVIDIA GPUs. So you would probably work with High Sierra. Well this doesn’s stops you here, there are many hack and techniques. Installing eGPU on MacOs

Alternatively , if you want to build a desktop or which type of GPUs is bang for your buck then you can go here