Hey there, fellow AI enthusiast! If you’re like me, you’re excited to dive into the world of artificial intelligence and start experimenting with models locally. But before you do, you might be wondering – what memory size do I need? Specifically, do I consider the dedicated VRAM on my GPU or the total memory, which includes shared memory?
I recently came across this question while researching packages for my own AI project. My NVIDIA card has 8 GB of dedicated RAM, but it also indicates 16 GB of shared memory, making the total size 24 GB. So, when choosing a package, do I look at the total size or just the dedicated size on the card?
The answer lies in understanding how VRAM and shared memory work together. Dedicated VRAM is the physical memory on your GPU, while shared memory is a portion of your system’s RAM that’s allocated to the GPU. When it comes to running AI models, the dedicated VRAM is what matters most, as it’s responsible for handling the complex computations involved.
So, when selecting a package, it’s generally recommended to consider the dedicated VRAM size. This ensures that your model can run smoothly and efficiently, without relying too heavily on shared memory. Of course, having more shared memory can be beneficial, but it’s not as critical as having sufficient dedicated VRAM.
I hope this clears up any confusion, and happy AI experimenting!