|
Page 1 of 3   New graphics cards from ATI and NVIDIA can support hardware acceleration of various video manipulation processes so time needed for some of those tasks can shorten, especially when it comes to video transcoding. Until recently for such task you needed separate application written specially for each GPU. Now, there is one application that can use both GPUs and do exactly the same job: CyberLink MediaShow Espresso. CyberLink promises ten times faster transcoding by using GPU processing power along with ability to transcode few files at once. Application has simple user-friendly interface and since it has Espresso in its name, you guessed it right: there aren’t too many options to bother you.
When it comes to transcoding, some compromises had to be done and some of you won’t like them. Compression is divided into five profiles: Apple, Sony, Microsoft, YouTube and manual. In first three you can pick device for which video will be converted: Zune, Xbox, iPod Nano, PSP, PS3 and similar along with resolution and folder where those will be saved. Other “important” parameters like bitrate or quality settings are absent, even in manual mode. Output file cannot be MKV container also. At the end of conversion you are offered option to watch a video and that is it.
Testing
We decided to use two videos for each format as a way to see if that will influence speed and quality of conversion. One video is 30sec trailer in 1080p MPEG4 format, and the other is also HD trailer around 2min long but in WMV format. Tests were conducted on two test configurations: one was based on Core i7 920 and the other based on “old” Core 2 Duo E8400.We did this because CyberLink MediaShow Expresso is optimized for Core i7 family. All tests were done under 32bit Vista since ATI stream doesn’t work under XP. We used ATI Radeon HD 4890 and NVIDIA GTX 260 graphics cards.
In case of ATI graphics card choice of hardware conversion was automatically set in MediaShow Espresso. We mention this because for NVIDIA choice of hardware acceleration is accessible at same place where other conversion parameters are set. NVIDIA hardware acceleration is available for Apple iPod, Sony PSP and YouTube presets so it is obvious that NVIDIA supports only hardware conversion to MPEG4 and MPEG2 formats.
In results you can notice inconsistencies, and it is obvious that speed of conversion doesn’t depends solely on target format. Format from which is converted also makes significant role when it comes to transcoding speed.
ATI
MPEG4 on Core 2 Duo E8400 and ATI graphics card resulted in nice acceleration except in case of conversion in MPEG4 1080p resolution. In case of second (longer WMV) file there was even deceleration in combination of ATI graphics card and Sony PSP profile (480x272). On the other hand, it will remain mystery why conversion to Zune format offered much better results for same video. We repeated few times this conversion and always came with same results. When it comes to ATI graphics card on Core i7 CPU no improvement in transcoding speeds were noticed, but now small deceleration is noticed in case of Zune format.
NVIDIA
In case of NVIDIA graphics card it can be noticed that speed of conversion raises as resolution of output video raises. In some cases conversion was four times faster compared to traditional CPU (software) transcoding.
CPU
In case of software transcoding Core i7 offers much better performances. All available cores were used so it is obvious that CyberLink did its job when it comes to Core i7 optimizations.
CPU vs. GPU
In both cases, Core 2 Duo and Core i7, during software conversion CPU load was between 80% and 100%, but mostly around 90%. In case of GPU conversion CPU load was between 40% and 50%, so it was obvious that CPU has chance to do something else besides transcoding. As we already mentioned NVIDIA and ATI graphics cards cannot transcode to all formats, so in case you do conversion to unsupported format there will be no help from GPU (only CPU will be used). Since CPU wasn’t 100% loaded with GPU accelerated transcoding, there is option to convert few videos at the same time. That is another way to shorten transcoding times. Our estimation is that 10 times faster transcoding is hard to achieve, but with further improvements (optimizations) in drivers and application it will be possible.
Picture Quality
As far as picture quality goes, some differences between GPU assisted and pure CPU transcoding were noticed. These differences were noticed only by looking at same video frames. These differences cannot be observed when looking at motion video so we could say that there isn’t any real difference between videos converted by GPU and CPU. Therefore, when it comes to quality GPU transcoding gets thumbs up.
|
