Hacker News Reader

Comment by eli_s | Hacker News Reader

Article mentions:

'The only thing I can foresee that could prevent the widespread adoption of manycore mainstream systems in the next decade would be a complete failure to find and build some key parallel killer apps'

I'm no expert in this area, but aren't there a lot of graphics problems that are easily parallelized? Given that gaming is now mainstream I'd imagine the first set of killer apps to be game related.

View on HN

This is exactly what GPUs do today, and it is also why NVIDIA and ATI are trying to make their GPUs available to programmers for general purpose use. See CUDA (http://www.nvidia.com/object/cuda_home_new.html) and Stream (http://www.amd.com/US/PRODUCTS/TECHNOLOGIES/STREAM-TECHNOLOG...). These technologies let programmers use the parallel computation engine of the GPU to solve problems. Using such a tool is non-trivial, as the problem has to benefit from massive parallelization and you also have to have a problem big enough to overcome the data load penalty (getting the data from main memory/disk to the GPUs memory for processing).

Good point, plus they are well motivated to improve performance and typically are on the cutting edge.

Who else does a lot of tech innovation? Since on the server side this relates to cost / page view I'd guess the super-competitive porn industry will do some of it.

And of course keep an eye on the functional programmers. With parallelized map and filter we can do a lot...

Also, just about everything Google does is highly parallelizable.

That's mostly because Google chooses to do only things that are highly parallelizable. Given the number of servers they own, I can't say that's a bad choice, either.

Isn't that exactly what GPUs already do?

and game console CPUs (PS3)

Reply on news.ycombinator.com