Today we will see a video card with 24GB of memory sold for less than $ 350. This is the Nvidia Tesla K80, read the specs on the Nvidia Website.

Tesla K80 Deep Learning

Why is the Nvidia Tesla K80 the solution?

Due to the shortage of microchips, and the use of video cards for cryptocurrency mining, the price of the hardware can reach up to 3 times the original price. In this case, the Nvidia Tesla K80 becomes the solution for Deep Learning.

If you have used Google Colab for example for the training of Mask RCNN you will surely have noticed that among the video cards that Google makes available there is also the Nvidia Testa K80. It sounds perfect for deep learning but doesn’t seem to be among the first options for deep learning professionals for their projects.

Why is it so cheap but not a common solution?

The Nvidia Tesla K80 is a server solution from 2014. It has 24 GB of ram split between 2 cards and 2946 Cuda core and all this is great for Deep Learning projects. Being a server solution, it is hardware subject to frequent replacement for updates and often Server Farms want to get rid of it quickly.

But it isn’t suitable for everyone. For example, it seems that they are not optimal for crypto mining and installation on a simple PC is problematic.

How do I install the Nvidia Tesla K80 on my pc?

If you want to use the Nvidia Tesla K80 as a home solution for your Deep Learning projects you have to pay a little more attention than a normal video card.

Cooling of the Nvidia Tesla K80

It has no cooling fans and there are no standard solutions on the market. You have to build an adapter and choose the fan that has the necessary airflow for the purpose.

Power

This video card consumes a lot: 300 WATTS. The recommended PSU for this card should be of at least 700 WATTS, and you should be very careful not to use a power supply that is below what’s required by all your components as that could damage the power supply itself, but also the components like (GPU, motherboard, CPU and so on.)