Processor progression

[Rigel]

Back in the 90s / aughts, you had a lot of talk about "Moore's Law," which was mistakenly stated as "Computer speeds double every other year." Especially during the aughts, you'd see articles about the impending death of Moore's Law, and how computers would eventually peak in performance.

What Gordon Moore actually observed was that "the number of transistors we can place within a given area doubles roughly every eighteen months." This has, in fact, held true even now. But it means something very different than it did 20 years ago.

I spent quite a bit of time with an indie video game this summer (Rimworld, if you're interested), and the performance with a larger colony got me thinking about replacing my laptop. So, I checked the resource usage to see what I'd need: light on RAM, one core at full load, the other cores mostly idle.

"Easy peasy," I thought, "My laptop has a 7-year-old processor in it, I'm certain I can find one with several times the performance."

Heh.

It turns out the fastest processor on the market (not in laptops, but anywhere) doesn't even have twice the single-core performance of mine. Depending on the site and benchmark the fastest Intel workstation processor has single-threaded performance somewhere between 60-90% higher than what I have now.

While it's unfortunate that single-core performance has started to lag so much behind the curve from a few decades back, there are two areas where performance has increased dramatically (one of which would help me, one of which would not).

The obvious one is multi-core. Where my laptop has two cores (and four instruction decoders), it's common today to find laptops with 6 or 8 cores in them. Unfortunately, this does absolutely nothing for a single-threaded workflow.

The second improvement, though, is in power efficiency. Further profiling of my system under load revealed that the CPU was being throttled due to thermal conditions, in some cases running at less than 1/3 of its peak speed! (It'll boost up to 3.2GHz, under load it'll throttle to 1GHz). New laptop processors can run 4 cores using 1/3 the electricity of my laptop's 2 cores, so they should perform dramatically better (I've seen quad-core chips with 15W TDP, whereas mine is rated at 45W).

This discovery led me to profile my system in quite a few other situations, and I discovered that through everything I do other than web browsing and email my processor is being throttled due to heat. I tried a decent cooling pad from Targus, but it only got me around 10% real-world improvement.


TL;DR Rigel's laptop is a space heater with a web browser.

[wayfriend]

I had been thinking about that, too.

My one contributing thought is that the industry stopped focusing on increasing raw CPU speed. Instead they have been focusing on "parallel processing", which covers everything from multi-core processors to cloud computing to massively distributed calculations such as are done by NASA. This is, in turn, in service to the Internet of Things, which is all about CPUs everywhere, but not necessarily big ones.

Then again, have you seen the latest news on quantum computing. It's FAST!.

[High Lord Tolkien]

TL;DR Rigel's laptop is a space heater with a web browser.

Good post!
Hysterical ending too.

[Rigel]

Then again, have you seen the latest news on quantum computing. It's FAST!.

They're fast, but they don't operate the same way as traditional computers.

They're extremely fast at solving complex calculations where you can solve for an unknown variable. But they don't perform imperative operations as traditional computers do.

**Everything above is with the caveat that I don't actually know what I'm talking about when it comes to quantum computing.

[Rigel]

I grabbed a Ryzen 5 laptop on sale yesterday, so the heat problems shouldn't be an issue anymore.

The 3500U is slightly faster in theoretical single-threaded performance than my old processor (3500U vs 4210H is 5-20% depending on the benchmark and the review site), but it's a quad core rather than dual, and it only draws 15W so it doesn't get throttled due to heat anymore. In fact, I ran a 10 minute stress test on it, and it stayed at boosted clock speeds the whole 10 minutes. Never even reached 75C where my old laptop would be at 99C.

The integrated graphics are actually quite a bit slower than the dedicated GPU in my old laptop, but I'd be hard pressed to find an instance of the GPU being the bottleneck for my usage anyway.

[wayfriend]

All's well that ends yada yada ??

[Rigel]

Yup.

Also, I didn't realize how quiet it's possible for a laptop to be! My old one is like a jet engine compared to this thing!

[Linna Heartbooger]

Then again, have you seen the latest news on quantum computing. It's FAST!.

They're fast, but they don't operate the same way as traditional computers.

They're extremely fast at solving complex calculations where you can solve for an unknown variable. But they don't perform imperative operations as traditional computers do.

^Dunot worry; wf knew that and was joking.
and that they are prohibitively expensive, etc. etc.

The progress in bringing them from the realm of theory into physical reality lately... yeah, kind of exciting.

Also, enjoyed this thread Rigel.
(I'm with HLT on majorly appreciating your "TL;DNR" version.)

(I don't keep up with hardware stuff either.)

[wayfriend]

Not joking. I was just trying to say that it seems like the semi-conductor industry has stopped trying to push the envelope with regard to instructions per second, and has been innovating in different directions.

[Rigel]

Well the "raw clock speed" you mentioned above is difficult... Intel had a horrible time with the Netburst architecture almost 20 years ago, when the clock alone was talking up more than half the wattage of the processor. As everything about it was designed for extremely high clock speeds, and they weren't (at the time) able to achieve those clock speeds, they were well and truly forked until they switched over to the Core architecture.

But IPS isn't all about clock speed. There were grad-student theses back in the 90s about clockless designs that basically operated as quickly as the wires carried the signals, and they were almost twice as fast in terms of IPS as the leading procssors of the time. They were also incredibly difficult to work with; it turns out that using a clock to synchronize the scheduling simplifies life dramatically.

Overall I'm quite pleased with my new lappy. Sure, the IPC is around par for the 7 year old one I was replacing, but the improvements to power draw are so dramatic that it can maintain the top speed continuously, meaning it's roughly 3 times faster than my previous. Plus, lower power draw means a smaller strain on the grid, which helps keep our planet clean.