34 comments

[ 2.7 ms ] story [ 139 ms ] thread
Render two chicks at the same time?
I just saw this post now and was going to add a comment with something similar. I wonder how many people's response to any question that begins with: "What would you do with..." results in an answer of something that includes the phrase "two chicks at the same time".
(comment deleted)
Calculate pi to 48 places.
write an essay about using that sun t2000s shipped with 64 cores years ago.... well at least its x86 now so you can run more... proprietary binaries?
Sun wanted to sell you theirs for $$$$. Now that AMD and Intel are at it, we'll have 8-12 cores for $200 in a few years. Things did seem to slow down a bit on the multi-core front. I was beginning to wonder what happened.
I need 42 to find the answer to the Ultimate Question,

the 43rd to satisfy the nerd,

the 44th to find north,

the 45th to bust a myth,

and the 46th idles because nothing rhymes with it

... and well, you finish it.

probably the same thing I do with my 8 core mac pro- under utilize it...
I have a Mac Pro too, and I know the feeling. Have you looked into distributed computing (BOINC, etc)? That's how I rationalize it. I see the electricity bill as part of my donation to science.
I would have them sit idle most of the time, since most of my tasks are usually IO bound.
Add them to a grid, so others can use them when you don't need them? Seti@home, but where you get paid for your cycles?
Well, it depends how you view the original question.

Sure, you could eventually find some purpose to the 48 cores, but that implies an active search for such a purpose and basically a side-project. I dunno, I might just collocate the server and sub-rent virtual machines or do some shared hosting -- but that's not what I care about.

So -- for my current tools and work-flow, 48 cores isn't needed. Pretty sure some scientist or video editor might replace his cluster with this thing but as a normal user and developer I would severely under-utilize the hardware.

For some reason, this idea came into my mind: that the existence of this kind of contests proves either the marketing departments have really trouble figuring out how to promote many-core processors, or there is no real demand for these monsters except a few niches.
Couldn't most of games use it for higher complexity graphics? It looks like people are still not developing for multicore... maybe it's still "too hard" for an average game developer?
Games face bottlenecks from heavy statefulness. Animations, physics, and AI have design-required interdependencies where a change in one will trigger a change in the other - e.g., the player character throws a box, it collides with an enemy, and the box breaks, while the enemy shows a visible hitreact or knockback, and possibly changes their future tactics. Sometimes you can queue and defer all of these changes for a frame and lose some responsiveness, but other times it's really critical to have them all sync up.

The render stage of graphics benefits from massive parallelization, but most subsystems can't see the same improvement. A common fallacy of new game programmers is to think that "one thread per actor" is a clean architecture - only to discover that each actor's thread blocks the others and causes massive slowdown.

As an OS developer trying to solve the I/O bottlenecks on 8 cores, probably cry...

Seriously though? I don't know -- maybe get a bunch of RAM and virtualize 32 machines for some distributed systems research.

I'd see if I could view flash banner ads efficiently.
I believe the next version of Firefox will be able to run flash in its own thread, so having a bunch of extra cores would indeed isolate the evil from the rest of your browser.
I've been running nightly's lately and it does indeed "isolate evil". Instead of a browser crash I now see a message at the top of the respective page telling me that the plugin crashed.
Just 16 cores short of having one per chessboard square, like the pioneering parallel chess machine Hitech...

https://chessprogramming.wikispaces.com/HiTech?f=print

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.126...

I'm glad this has been done :) I thought of the same thing when I first heard of Tilera's TILE64.
As Minsky has said, parallelism isn't worth a damn thing if you don't have better heuristics to prune the search space.

If you have a million processors to play chess with, and your search tree still has a branching factor of thirty, you're only looking two moves ahead.

The human mind is amazing because our pattern-matching-based search heuristics are great.

Our pattern-matching-based search heuristics are great because the human mind is amazing.

At least that's how I see it ;)

(comment deleted)
I'd run 48 instances of Rosetta@home.
I'm cheap so I would simply use it for an embarrassingly parallel problem. I'd personally have the most fun with ray tracing or brute force cracking.
The MapReduce tasks I run on Hadoop are heavily CPU bound. This would be a dream come true, at least until the light bill arrived. I have to wonder, with so many MR tasks running on a single box, would IO become an issue again? Suppose you have 12 HDDs and 96 tasks running. At some point the tasks are going to fight each other for the needle.
Crack passwords and keep my office warm.

Alternatively I'd set up a transcoding farm and build a service to help people to edit their videos online and download them in multiple formats. My goal would be something like iMovie but cross-platform with subscription meaning exclusive filters, wipes and effects with more formats.