Cuda MP3 contest

That’s a great world for Cuda.
Nvidia team is visiting European universities to promote their product, as “Cuda is the best solution …,
we need scientists like you to use it to prove it can boost real world applications …”.
And “you’ll see soon a contest …”

Unfortunaly, Europeans are always put out, “United States and Canda only”.
Why are we excluded, doesn’t cuda API run in spanich, french or german ?

Thanks a lot Nvidia

legal issues

Why did Nvidia corporates make advertisement about it during their “European tour” ?

It’s a pity.

who tells you they knew about these issues?

You’re right, it’s not their fault, it’s mine.

There is another Cuda challenge opened for europeans :

Thanks a lot, one can make an accelerated lame encoder :)

  1. What legal issues?

  2. What is the point in accelerating mp3 encoding (and only under Linux mind you!) when the disk I/O is the bottleneck?

I can encode 1h 18m audio file in 40 seconds using Core 2 Duo E8200 CPU here and my RAID0 hardly manages to keep the CPU busy.

apparently the rules for these kinds of competitions are quite different between us/canada & europe.
You can check the differences in the term&conditions between & There is now also a competition in europe.

Just one example: say you win. Would you like to go through the trouble of filing a tax return in the US?

Because technically that’s what you’d have to do…

Contest prizes are (even in the form of hardware) are considered taxable income and company has to provide tax forms for that etc. Even if you’re living outside of the country, you could be breaking US tax law if you don’t file.

etc etc

I’d propose a European counter-contest to accelerate ogg vorbis. Who’s willing to fund the prize money? ;)

There is already a European contest running……testid=2&page=2

1.27 score = 1.27x speed up ? :blink:

Some experts on the hydrogenaudio forums stated that the LAME encoder spent only about 20% of the CPU cycles in FFT, MDCT and like filters that are easily accelerated in GPU. The remaining code (say, huffman coding, psychoacoustic model) are difficult or even impossible to port to a GPU.

So if you accelerate those 20% by a factor of 100, your entire encoder goes… about 19.8% faster ;)


it is up to 2.25 now :sorcerer:

Lots of submissions today. Will the numbers improve?