Okay, so it's going to be a little shorter today.

I just want to give you

a overview and a little walkthrough

of NVIDIA's GPU architecture.

So as we had with Intel and now with AMD,

we have approximately the same thing for

Nvidia's history of chips they made and micro architectures.

So Nvidia's first GPU came out in 1995,

so quite a long time ago, I guess.

Depends.

Depends.

What do you say, how it's long or how it's not long?

Well, if you consider the advancements that have been made in the last 11 years,

it's quite a long time, I guess.

20 years.

It was a lie.

Who wasn't, right?

Depends.

Anyways, it was one of the first GPUs.

And from 1995 to 2006, it's not like nothing happened there.

It's just that they produced lots of GPUs, but they didn't have any micro architecture

name or anything.

They were really just GPUs.

And back in the day, what came up in the early 2000 was a thing called fixed function pipeline,

where you could program separate parts of your GPU.

Like, if you remember or not, from the last lecture, we had the shader programs, right?

And fixed function pipeline means that you either have a fixed function so you can

dispatch your geometries, your pixels, etc., which is then being worked on by the GPU.

And eventually, the shaders came into play, right?

Where you could program separate stages of your shaders, right?

Pixel shader, geometry shader, vertex shader, which was the beginning of programmable GPUs.

And in 2006, the GeForce 7000 series was the last GPU which had a fixed function

pipeline architecture.

In 2007, the first fully, well, fully to some extent programmable GPU was released.

The micro architecture code name was Tesla.

And since it was the first fully programmable micro architecture, it was also the first

CUDA GPU.

So CUDA is the programming API for NVIDIA.

For NVIDIA GPUs.

Okay.

There was another name.

You say the last time you said that there was...

Because CUDA is like a programming language or is it like the assembly of...

CUDA is the programming language that we in scientific computing and high performance

computing care about.

Okay.

So no, yes, to some extent.

It's a general purpose programming language.

Okay.

So general purpose in terms of GPU architecture.