<div class="text-justify"

Hello everyone. Usually here I talk about different stuff from technology to crypto and sometimes geopolitics or finance. I'm sure you all know that I am quite interested in these topics by now as is also labeled on my profile. Today. I want to share something personal. And by personal I mean something I was able to do. So, it can be said as a personal experience. If I throw in a little bit of context, then I am studying as a computer science engineering student in a university in my country. Joy, after I got admitted to the degree and after chatgpt took the world by storm, I thought long and hard about it. I was trying to decide where I should head after graduation. More specifically, which field should I pursue for my future career. And I landed on AI. Now that is something you have heard a lot of times but I believe I am serious enough about this.

---

![1000033126.jpg](https://files.peakd.com/file/peakd-hive/abrar-fahim/23y8zPX6kDQEi1LG9jgBKxAiifhQ7wLXLHoy23NEYW1kMKBhkdnnmVSedktCdg8ocx5kc.jpg)
[Source](https://www.pexels.com/photo/person-holding-smartphone-while-using-laptop-1181244/)

---

So part of pursuing a career in artificial intelligence is actually knowing to work with one Joy. I did some research and realized that most AI stuff requires a certain part of python language and that is pytorch. The biggest problem about running pytorch natively on your system is that you need an Nvidia GPU. Without the Nvidia GPU, you do not get native support for pytorch. But considering these Nvidia gpus are very expensive at the moment. Even if they are expensive, they aren't even available for purchase in the first place. Makes it very hard for people to get into AI and machine learning and they have this Monopoly because AMD does not support pytorch the same way Nvidia does. And the main reason behind all of this is because Nvidia's Cuda architecture helps accelerate pytorch code in the best way possible. So due to this hardware limitation Nvidia had the upper hand. But thank the Lord until came to save us apart. Students there need the released gpus along with an extensive collaboration with the pytorch developers has allowed Intel to run pytorch on their gpus with a few extensions that weave the pytorch code into the xmx engine of Intel gpus which can be called the Intel counterpart to Nvidia's cuda cores. This has been a game changer. For example, it has allowed me to run pytorch code without breaking the bank. I have successfully tested my first pytorch code today hopefully on my machine which wasn't expected to be possible even 2 years before. And it has been possible in the past 2 years but the integration wasn't seamless before. Today, I just had to write two extra lines of code during the installation of pytorch and that enabled me to run pytorch on Intel GPUs.

From a costing standpoint I have spent about 260 US dollars to get the GPU which has 12 GB of vram. You must remember that vram is the most important spec in the GPU for running llms whereas if I had to wait for Nvidia then I had to spend more than $280 and still get only 8 GB of vram which is too low to work with large language models. If I actually wanted an Nvidia GPU with an ample amount of vram I would have to spend at least $400. So I believe the cost of an extra extension and a relatively slow processing rate is still worth it because I am saving almost half the money.

</div>