There are many occasions when one might think that a type of hardware is really simple because its size and everything it incorporates is designed to make it appear that way, but that does not imply that when it is It's a matter of configuring it, that's how it is. . This is demonstrated by a user who has managed to carry out several projects related to the power that a Raspberry Pi can have when additional PC components are connected to it, and this time he has managed to accelerate the functions of AI linked to an LLM in a Raspberry Pi by directly connecting a GPU external.
One of the main reasons why Raspberry Pi They have managed to become one of the most used devices for all types of projects. It is in the simplicity that they offer when first used, but with the possibility of configuring them to the maximum to create anything we can think of. We have seen many times how some users use it to bring their most complicated projects to life, but now we have also seen it work as if it were a normal computer with graphics oriented to gaming , but which is also perfect for using AI-related functions.
It is possible to connect an eGPU to a Raspberry Pi to increase its AI performance
Las Raspberry Pi These are a type of single board computers that have much higher potential than we can imagine. The development company itself has launched different modules that allow us to increase their capabilities by adding functions. AI or other components such as an SSD. But now a user known as Jeff Geerling on YouTube has shown that it is possible to get even higher performance if you connect an external graphics card, which he announced by testing how the various triple A games, but now wants to try artificial intelligence.
This presents quite a challenge, as noted, because it is not at all easy to incorporate a GPU external in a Raspberry Piand even less create compatibility with a LLM to increase the performance of artificial intelligence. In this regard, Jeff says the answer to GPU-accelerated LLM on the Pi 5 is the Vulkan API, but it's still extremely complicated since it requires the user to compile their own Linux kernel, gather a handful of drivers and fixes and more to make it work.
But this is not the only thing that we have to take into account, and that is that if we want to connect a GPU to the Raspberry itself, we will realize that it is quite difficult, since in this case Geerlink used an adapter to convert the Raspberry Pi's PCIe Express FFC connector into an M.2 slot which you then connect the GPU itself to. He does this using an M.2 to OCuLink adapter as well as an OCuLink GPU mount cable, showing in the video that he's using an RX 6700 XT, to which obviously a power supply also needs to be connected.
