Nvidia 2080ti Was Designed for DL

Nvidia 2080ti was originally designed with two versions of v-mem: 11Gb and 22Gb.

However, since large gpu memory was not very useful for gamers, few manufacture of 2080ti GPU chose to provide 22Gb version to the market.

The Pursue for GPU Memory

Recently, due the rise of deep learning applications especially the popularity of stable diffusion, lora finetine, LLM, LLAMA, LCM etc, many people are looking for fast GPUs with large video memory. 

Big corporations go for enterprise level V100, A10, A100, H100. Budget limit is not their top concern. Even if you could afford the premium of these cards, you cannot install these cards into a regular PC due to the difference of heat dissipation. These cards are designed for data center. They lack onboard fans which are vital for home-brew rig equipments.

If you are a student, a hobbyist, or a start-up team, you will find buying GPUs with large large vmem heart breaking. They are just too expensive.

Most modern generative deep learning applications require at least 10Gb vmem. As of the year of 2023, the most commonly choices are 3090 and 4090. Why not 20xx device? Because most 20xx devices come with at most 11Gb vmem.

2080ti the Game Changer 

Although common 2080ti comes with only 11Gb vmem, you can easily leverage its official support of 22Gb video memory by upgrading its vmem manually.

Such manual upgrade does not require any software or bios change. As long as you upgrade the v-mem chips, all software (nvidia-smi, nvidia driver, pytorch, etc) automatically recognize it with 22Gb memory.

Below is a comparison of common deep learning GPUs for individuals. Prices are retrieved from Amazon as of Nov 2023.