Video Processing Perspectives

Now that we have some products out the door, we have been focusing on making the core technology even faster than ever using off-the-shelf NVIDIA GPUs.  One of the biggest boosts in speed will come from taking advantage of multiple GPUs in the system.  Here is a sneak peek of what we are doing to leverage a multi-GPU system. Multi-Movie Parallel Transcode The most straightforward way to take advantage of multiple GPUs in the system is to send separate transcode jobs to each GPU.  This means one GPU could be transcoding one movie while another one can transcode a second one each at their already very fast speeds.  We have been prototyping this using Badaboom already by launching 2 Badabooms each using a different GPU in the system transcoding different movies.   Doing this essentially doubles the already fast speed achieved with Badaboom.

This system has 2 of the older 8800GTS video cards in it and a 3 GHZ dual core CPU. What this shows is transcoding 2 entire movies in under 20 minutes (one will complete in 12 and the other in 18)! Content, GPU proximity to the CPU (PCIe slot), and other system factors make it so both speeds are not exactly the same, but It is showing effectively 350 fps to 400 fps of total sustained throughput.   Splitting up the movie Many folks will want to simply convert 1 movie faster instead of 2 at the same time.  We can take advantage of the two cards by transcoding 1/2 the movie on one GPU and the other 1/2 on the other.  We will then stitch the two back together which is a very fast operation.  This means an entire movie transcode in under 7 minutes! Today, you can get 2 8800GTs for a total cost of about $240 which would be even faster.  Of course if you want to go BIG, then you could get 2 GTX280's and hit somewhere around 600fps of sustained throughput.  Transcoding 2 entire 2 hour movies in about 13 minutes. When can I get it? Look for initial multi-GPU support in Badaboom before the end of the year!

• ‹ previous
• 94 of 106
• next ›