What's up with multicore CPUs, anyway?

[Naniyue]

I know my dual core mini laptop is weak, but it's bad when old games from the early 2000's only recognize one core and thus don't run well, yet enough games from recent years run OK with some tweaking. What is so special about these cores? Why not just continue to have one base CPU? Is there some sort of heating/cooling, or computational advantage? What's the deal? Thanks!

[Hellser]

CPU cores can split the work load. Back in the day, we had to increase our processor speed further to get things done quicker. We're looking WAY BEYOND the 3.40 GHz range for a single core processor to do what our 2, 4, 6, 8 and so on cores can do today. So yes. Computational advantage and cooling advantage is part of the reason why we abandoned the Single Core way of life.

If you need an analogy - imagine running to get a bucket of water to douse out a fire. Now imagine two, four, six or even eight of your friends helping you with that task. Things get done quicker.

[wildweasel]

If you need an analogy - imagine running to get a bucket of water to douse out a fire. Now imagine two, four, six or even eight of your friends helping you with that task. Things get done quicker.

At the same time, to keep costs low, lower-priced multicore CPUs may emphasize more cores over single-core performance, making them faster with things that acknowledge the extra cores, but slower with legacy software that doesn't.

[Hellser]

At the same time, to keep costs low, lower-priced multicore CPUs may emphasize more cores over single-core performance, making them faster with things that acknowledge the extra cores, but slower with legacy software that doesn't.

Exactly. Hence the entire line of AMD FX processors. I will admit they can kick the crap out of my 3570k in a multicore task. But single core, Intel has AMD beat.
(It's logical, it's been proven a hundred times by many websites, let's not turn this into a hardware flamewar).

[Naniyue]

Thanks for the explanations! I'll be more careful in choosing my next laptop.

[Gez]

If you need an analogy - imagine running to get a bucket of water to douse out a fire. Now imagine two, four, six or even eight of your friends helping you with that task. Things get done quicker.

In that analogy, the water buckets were originally just thimbles, but rapid improvements in water container technology have allowed to increase the volume each "bucket" carries, moving onto glasses, pots, and barrels. Eventually, however, there was the issue that larger buckets would be too heavy for a human to carry, so instead of keeping up with continually increasing bucket size, increasing the number of buckets started to make more sense.

[GooberMan]

Why not just continue to have one base CPU?

Engineers have been battling with the realities of electricity and silicon for some time.

But another thing I've observed: Look at what supercomputers are doing now. In 20 years, that will be roughly normal.

[Naniyue]

Yes, it's amazing how far computing technology has come. I remember when things like an ipad were mere science fiction. Yet I'm sure there's still a ways to go. I'll be long dead before there's finally a Star Trek holodeck, though.

[Caligari87]

I'll be long dead before there's finally a Star Trek holodeck, though.

Don't sell yourself short on that. Maybe not a "holodeck", but I imagine stuff like this is gonna be amazeballs in 10-15 years.

[DaMan]

In the before time long long ago Gordon Moore made a law saying CPUs have to double transistors every 2 years or people on the internet would say the sky is falling. In the olden days people would say to Intel your CPUs suck RISC will take over someday then Intel would throw more transistors at their CPUs overcome whatever these fancy RISC CPUs did better. That stopped working a decade ago so they throw more CPUs and now GPUs onto the die. A single core 14nm CPU would probably be smaller than a 6502.

---

For reference here a similar size (135mm² 169M transistors) single core Prescott2M.

---

[Gez]

Note that RISC still haven't taken over.

[GooberMan]

And in fact, the way that Intel processors break x86/x64 instructions in to microcode would have to make the internals more RISC-like than the instruction set is.

[Graf Zahl]

Note that RISC still haven't taken over.

And for that they'd have to provide some very significant speed advantage per core, which can't be easy with how current x86's work internally.

But yeat, the speed gains over the last 4 years are close to nothing so I wonder where things will head in the future. Fact is, parallel programming needs a lot more support from both programming languages and operating systems to be of more use. I once tried a bit of multithreading inGZDoom's renderer but it was all cancelled out by the synchronization overhead.

[printz]

Isn't multi-core of more benefit if you have a lot of working processes in the background? Just so you don't experience sudden slowdown in your main app because something else is happening elsewhere.

[Graf Zahl]

Of course, but it's also useful if you have some time consuming tasks that can be handled in parallel.

What I'd like to do in GZDoom is to have two worker threads that receive items from the BSP traversal and transform it into render data. But the big problem is that OpenGL is fundamentally incapable of multithreading in any usable form, so I can't get enough workload together to make such an approach pay off. Since all rendering needs to be done from the main thread (and in the same small slices as is now) the bit of work I could offload gets cancelled out by the synchronization calls on the data queues this would require.

I guess if I could migrate to Vulkan things would look quite a bit different, but my 4 year old graphics card can't even support that, making the API totally useless for a game like Doom with so many users with old hardware.