Hacker News Reader

Comment by sjcoles | Hacker News Reader

The 10 year span between 2000 and 2010 also had one of the largest relative uplifts in transistor density. Going from 130nm to 28nm was HUGE.

I think people forget how quickly computers (especially in efficiency!) improved from just 2000-2005 and again from 2005-2010.

Everything has leveled out a bit until recently. I think we'll see another 5 year growth spurt once stacked designs become common end of 2022ish. Though TSV vs EMIB is a whole other argument/ball of wax.

View on HN

The 1990s felt similarly insane. In 1991 we we had Hovertank 3D, in 1993 we had Doom, in 1998 we had Half-Life, and in 2000 we had Deus Ex.

Go back nine years today and things don't feel that different from now. People still buy Skyrim.

AR is probably in its "hovertank" era right now, after years of interesting but largely useless demos.

Someday we're going to wonder how anyone managed to navigate or buy things in shops without realtime vision markup.

I just want a little note that reminds me of a person's name and how I know them.

That probably has more to do with the human bottleneck in games, than anything else.

Big open world is good, if you can manage to fill the big open world with stuff. But procedural generation is annoying and repetitive, but at the same time handcrafting a world is time and labor intensive. See: Cyberpunk 2077

> Everything has leveled out a bit until recently.

You can say that again. I'm typing this on a desktop PC powered by a Xeon that is approaching 10 years old and I don't have any complaints. The cost of upgrading it over the years has been modest.

In comparison I have a Nexus 5 that is about the same age and getting modern Android on it is a pain, the processor really struggles and the physical device itself has seen much better days.

Anecdote: I'm typing this on a 2013 Xeon (Haswell E3v3), and it's acceptable for web/productivity, but the x265 encode performance is absolutely abysmal compared to newer CPUs, especially AMD's. GPU encoding is good for ephemeral streams, but good archival encoding is all CPU-bound: https://forum.doom9.org/showthread.php?p=1908148#post1908148

"""Luckily""" for me I do mostly standard-definition encodes and can get a sustained ~11FPS with my own ICC-built x265 binaries and my own custom "SD" encoder tuning. Every time I encode an HD video I wonder why I haven't replaced this system yet. Then I look at component prices and remember why.

The big jump has been in storage. HDD to SSD is 4-10x and from SSD to NVMe is 4-10x again.

Yes, I was going to say -- the increase in IOPS is off the charts. I worked at an ad-tech startup in 2010 and that was the biggest bottleneck for what we were trying to do. At that point SSDs were a thing on the desktop, but not really in the datacenter still. Too pricey.

Kids today with their insanely high IOPS! They don't know the pain!

I remember upgrading my CPU and GPU every year because of the exciting new technology being released almost constantly.

> 28nm was HUGE

... so to speak.

Reply on news.ycombinator.com