Ask YC : Are multicores useful? What will you do with 32/64/128 cores? | Hacker News Reader

Ask YC : Are multicores useful? What will you do with 32/64/128 cores?

2 pointsrandomhack6 comments

Before you remark that people need more than 640k, read this patiently. My parents run a 3-4 old year system on XP and it runs just fine. They are able to browse the web and run skype. Most of their meets are already met by web based apps. I see no reason why they would ever need a 64 core computer. On the other hand, they do have a real need for a portable machine. They would also benifit by better connectivity, easy to use user interfaces and reliable (no viruses please) software.

Is the same thing happening to the "normal user" the world over? Apart from servers and games, are there any real needs for much better CPU performance? Or is the focus shifting to "other things" ? Perhaps to cost, mobility, reliability and ease of use? If so what changes do you see happening in the computer industry?

Notice the big push of Atom processors by Intel. Notice the success of EEE Pc. Are big changes coming?

Please tell me what do you think people will do with 64 cores? On the technical side, keep the Amdahl's law in mind too. For applications with even 3-4 % serial component, you will be very hard pressed to double performance with doubling of cores. People talk about better programming languages. But have parallel algorithms matured to a point where we can indeed take advantage of 64 cores?

Does the emperor have no clothes? Will the multicore revolution be relevant for the populace at large?

If history is any guide, that compute power will be used by new applications, not better versions of existing ones (although 1080p Skype will probably also happen). So the inability to scale existing software is irrelevant; new apps will be coded from scratch using new parallel tools (assuming that those appear).

It's possible that no such apps will emerge, but I'm not betting on it. Many of them will look pointless at first. (Check out this 150MB screen saver: http://scr.sc/products/dropclock/ A waste of disk space, you say? 150MB costs a few cents.)

There's a lot going on here.

First thing is Moore's Law. If it continues to hold, we'll be able to have more processors working on the same task, but more importantly, we'll have more processors working on different tasks, concurrently. This is huge. Think "computronium." A revolution? Yes.

Secondly, the problem of optimizing our algorithms for maximizing multithreaded performance is not solved. There's still much research being done in this area (e.g. Terascale, good CS departments). We can assume that our ability to use parallel threads will increase, although it will level off eventually, of course. The question is, how good are we at using many processors? I would wager that we aren't very efficient at that yet.

You are correct in pointing out the applicability of Amdahl's law, but exponential decay in the processor/performance ratio does not immediately mean we still can't reap major benefits from more cores. In this vein, I encourage you to think about the "long tail," and what that could mean for (parallel) computing power.

Probably there needs to be another network infrastructure change before the masses adopt large numbers of cores. By that I mean, that current broadband will have to increase again to higher speeds. I mean, it makes sense that your parents can still run a 3-4 year old system if the most cpu intense program they run is Skype. These programs probably won't need the crazy cores until they start processing far greater amounts of data.

However, there is the issue that your typical user interface is becoming more and more graphic intensive. And there's always that end consumer that runs a million apps and never closes them. A few of my non-computery friends are like that and have a bajillion things in their system tray. They too will probably see performance improvements from multiple cores, long before the apps they use take advantage of it, simply because of their habits.

I forget where I saw the lecture on the internet, but the argument basically stated that people are only starting to realize what multi-core processors are able to achieve. It's a relatively new technology to the average user, and it is going to take some time before applications start moving down the development pipe that takes advantage of multi-core.

To answer your questions, no one knows yet what the average user could do with that many cores, but I can almost guarantee you that their will be an answer within the next five years.

640k blah blah bill gates.

there are uses for that multiple cores. will you need 32 cores in your average home PC? i highly doubt it. just like the average user doesn't need a RAID array or 4 gigs of RAM. yet. perhaps things will progress far enough in the future to require that many cores, but right now they don't.

Yes. Build Skynet.

Reply on news.ycombinator.com