Mind-blowing

[Sussexlad]

I bought a cheap second-hand Nvidia graphics card in an attempt to solve an update niggle with my Centos Linux - still to be sorted. However I looked up the card details, just out of interest and came across this.

https://www.techpowerup.com/gpu-specs/nvs-300.c1452
"The GT218 graphics processor is a relatively small chip with a die area of only 57 mm² i.e. 7.5 x 7.5, and 260 million transistors"

I guess this is pretty small-fry now, it's from 2011, but I find the transistor figure mind-blowing, i mean, how do you design something to manufacture them? I used to play with transistors in a small way, when they were individual items !

[kiloran]

I bought a cheap second-hand Nvidia graphics card in an attempt to solve an update niggle with my Centos Linux - still to be sorted. However I looked up the card details, just out of interest and came across this.

https://www.techpowerup.com/gpu-specs/nvs-300.c1452
"The GT218 graphics processor is a relatively small chip with a die area of only 57 mm² i.e. 7.5 x 7.5, and 260 million transistors"

I guess this is pretty small-fry now, it's from 2011, but I find the transistor figure mind-blowing, i mean, how do you design something to manufacture them? I used to play with transistors in a small way, when they were individual items !

Indeed.
I worked in the semiconductor industry for 33 years, and never ceased to be boggled by the small geometries and mega-transistor-count.
Such were the small dimensions, our wafer fab was built on a damped spring-loaded platform to minimise vibrations from road traffic, and the air-conditioning fans were designed to minimise air vibrations.

--kiloran

[Julian]

I once had a tour of the Intel fabrication plant in Ireland when I was working in the IT industry. There was a standard talk that went with the tour and they'd obviously honed the analogies to perfection because they really hit home.

The one that really stuck in my mind, and this was about 20 years ago so I hope I'm remembering it right, is that if the silicone wafer that they were working on was as tall as (i.e. as thick as - we're not talking about its diameter) the Empire State Building then the smallest structures that they were actually laying down on the surface of the silicon at that time would be the thickness of an A4 sheet of paper. At the time I seem to remember that the thinnest layer being laid down was 5 atoms thick, I suspect it's fewer than that now.

Transistor counts can get well into the billions on todays stuff and that's not even high-end data centre equipment. In fact the Apple A12X has hit the 10-billion-transistor mark (https://wccftech.com/apple-a12x-10-bill ... rformance/)

For a tech enthusiast it's an exciting time to be alive.

- Julian

[Julian]

...
I worked in the semiconductor industry for 33 years, and never ceased to be boggled by the small geometries and mega-transistor-count.
Such were the small dimensions, our wafer fab was built on a damped spring-loaded platform to minimise vibrations from road traffic, and the air-conditioning fans were designed to minimise air vibrations.

--kiloran

That's another thing I remember from the Intel tour/presentation. I think they had 4 aircon units to control the air in the fab and each one was "the size of a 747 engine". There was also mention of how much purer the air needs to be inside the fab compared to an operating theatre. I can't remember that soundbite but it was orders of magnitude. 100 times less particulate contamination? 1000 times less? As I say, I forget, but it was another impressive statistic. Plus the sheer amount it cost to build a big fab like that Intel one which was building, amongst other things, whatever Intel's flagship CPUs were at the time. I seem to remember the figure in circa 2000 was something like $1 billion to build the fab.

- Julian

[PhaseThree]

The cynical insiders view of Moore's Law was always that the cost of a new wafer fab doubled every 18-24 months, so you needed to double the number of transistors you sold just to break even.