Hacker News Reader

Comment by bawolff | Hacker News Reader

Amazing how cpu clock speed seems not out of place at all, while the rest seems so out of date (like 56k modems)

Biggest differentiator being that back then most CPU's were single core - one of the first dual core CPUs was the Pentium D that only came out in 2005 iirc.

Back then having a "Dual Core" of "Quad Core" processor was such a selling point, I remember getting my first Macbook (white plastic one) thinking it was a beast for having a Core2 Duo. Then like a year later, my Android phone had a quad core processor inside.

May 25, 2005 https://en.wikipedia.org/wiki/Pentium_D

May 2005 https://en.wikipedia.org/wiki/Athlon_64_X2

What a race! Looks like the era of 90nm processing, and Intel just beating AMD to the punch, with both of them moving to 65nm in 2006!

Anandtech preview of Pentium D in April 2005.

https://www.anandtech.com/show/1657

Anandtech preview of Athlon 64 X2 in April 2005.

https://www.anandtech.com/show/1665

Review of Athlon 64 X2 May 2005!

https://www.anandtech.com/show/1676

> Both AMD and Intel appear to be playing release date games with their latest dual core processors.

> Intel's affordable dual core desktop solution, the new Pentium D, officially launched in the middle of last month, but has yet to be seen in the channel.

IPC and caches have grown a lot (and obviously core count, dramatically).

A 2000 to 2005 ish pentium might have 50 to 200 million transistors, an M1 Pro is more like 12 billion.

It always comes to intersection of physical world and computers. After all electrons can only move so fast that is so far in single clock cycle... And I think more complex chips are internally "longer"...

We could probably make some older generation chip now and run it much faster, but it is easier and more efficient to add lot more chip...

I don’t think it’s the electrons moving that matter. It’s the electric fields, which in conducting materials, is approaching the speed of light.

The latency you’re thinking of is more related to the speed with which a semiconductor can change from conducting to non-conducting states.

The speed of electrons in a wire with a potential difference across is it is actually quite slow (cm/s or so).

Also, modern chips and electronics aren’t significantly larger in a physical sense, the wires/traces aren’t longer, but the density of transistors has increased.

Someone please correct me if I’m wrong.

The P4-era Intel Netburst pipeline, in order to chase the highest clock speed, at some point (Prescott) degenerated from 20 stages into a 31-stages deep abomination (which would suffer so badly at any branch mispredict).

The pipeline actually has a couple stages called "Drive"... allegedly for allowing signals to propagate between key areas in the CPU.

Interesting, let's take a look at the instructions per clock. Oh, yikes.

https://cpu.userbenchmark.com/Compare/Intel-Pentium-4-380GHz...

36% better memory latency.

Four times the single core performance.

Two decades.

Ten times cheaper

It's not much compared to the two decades of 1980-2000, but it seems there's nothing interesting on the horizon for 2010-2030.

Aren't you comparing a fairly high end processor at the time (I wasn't alive when the pentium 4 came out, I think) with the bargain bin one from today? The new pentiums have the same microarchitecture but I don't know if they have all the execution units turned on + less cache.

Reply on news.ycombinator.com