High performance clients (GPU, SMP/NVIDIA CUDA™) more than 200 gigaflops.
OfflineHi,
Is there Developers or DIVX TEAM interested in taking advantage of GPUs for parallel processing, and faster computations methods for transcoding divx.
As Nvidia now provides free for download development kits, SDK, tutorials for development of programs in c language to take advantage of GPU's. found at:
http://www.nvidia.com/object/cuda_what_is.html
MP3 LAME Encoder already does it
How would go about it, how would one obatin the a current source code of transcoder for divx, so that i can be enhance with new funtions calls and alogerims in the GPU.
The CUDA contest series will feature varying code topics, and all developers are invited to give their best try at speeding up the code and enter for a chance to win great prizes.
Supported products:
http://www.nvidia.com/object/cuda_learn_products.html
Some quato's from PDF found at http://www.nvidia.com/object/io_1195170003876.html
By simply looking at the clocks and capacities of the GeForce 8800 GTX GPU, we observe that it is capable of 172.8 gigaflops (128 processors, 1.35 GHz each, one floatingpoint operation completed per cycle per processor). Multiply-add instructions (MADs)
perform two floating-point operations every clock cycle, doubling the potential performance.
The Stanford University group
compares their GPU implementation to a highly tuned CPU implementation (SSE assembly language that achieves 3.8 gigaflops, a performance metric we cannot match) and observe the GPU outperforming the CPU by a factor of 25. They provide code
(written in BrookGPU) and analyze what the code and the hardware are doing. The GPU hardware they used achieves nearly 100 gigaflops
Performance Increase with Loop Unrolling
If this performance is achieved, the G80 GPU will perform approximately 10 billion body-body interactions per second (128 processors at 1350 MHz, computing 4 bodybody interactions in 72 clock cycles), or more than 200 gigaflops.
Anyone interest or wound like to shed some more light on this matter?
License Agreement
NO COMMERCIAL USE: This License Agreement grants our community members the right to use the Software downloaded from DivX Labs for personal use only in order to evaluate and provide feedback about it to DivX, Inc. Commercial use of the Software or of the work products resulting from its use is not permitted under the Terms of Use of DivX Labs.- To read the full Terms of Use please visit the Terms of Use page.
- To inquire about commercial licensing please email us
I am surprised there isn't an announcement from Divx to include the use of CUDA.
I am leaning towards buying Nero Move It personally... they are going to enable the use of CUDA starting next month:
http://www.tomsguide.com/us/nero-Cuda-Nvidia-video-encoding,news-3559.html
Plus that program seems to do way more than Divx for the same price(download price).
I am very interested to hear anyone that has a reason to go with Divx vs that Nero Move It program.