Originally posted by MartjeB
View Post
Announcement
Collapse
No announcement yet.
The UVD/UVD2 thread.
Collapse
X
-
Why do you think emulating this using shaders is efficient when there is already a dedicated part in the chip for that purpose? Then you can use raw cpu power too, because you only need about 3 ghz cpus and those will get cheaper over this year - when you oc a cpu it is already cheap.
Comment
-
Originally posted by Kano View PostWhy do you think emulating this using shaders is efficient when there is already a dedicated part in the chip for that purpose? Then you can use raw cpu power too, because you only need about 3 ghz cpus and those will get cheaper over this year - when you oc a cpu it is already cheap.
I doubt he will understand that. Try in his tongue.
openCL=GPU in highest power state
GPU in highest power state power consumption > CPU in highest power state power consumption
GPU in highest power state heat > CPU in highest power state heat.
openCL = more power consumption + more heat
vdpau netbook/notebook user = Lord of the Rings Extended Edition
openCL netbook/notebook user = an episode of MacGyver + roasted testicles + looking for power plug-in
Comment
-
Originally posted by deanjo View PostI doubt he will understand that. Try in his tongue.
openCL=GPU in highest power state
GPU in highest power state power consumption > CPU in highest power state power consumption
GPU in highest power state heat > CPU in highest power state heat.
openCL = more power consumption + more heat
vdpau netbook/notebook user = Lord of the Rings Extended Edition
openCL netbook/notebook user = an episode of MacGyver + roasted testicles + looking for power plug-in
Comment
-
Originally posted by Vighy View Postfunny, but most of all: true!
Thank you. Just wanted to make it as clear as possible as he seems to have some delusions that any tech that is not opensource is the death to it which is far from the truth. Ultimately a large part of that falls upon the opensource developers themselves and their willingness to work with their closed counterparts.
His "counter point" basically has little merit. Yes openCL could be made to handle x264 playback. No doubt about it. Can it do it efficiently? Not likely. GPGPU solutions only achieve their stellar efficiency when taxed enough to compensate for the additional draw on resources. A pickup truck (vdpau) will always be more efficient hauling around a refrigerator then a semi tractor (openCL at least pertaining to a usage on a GPU) hauling around the same fridge. Even a car towing a trailer (CPU decoding) will have better efficiency.
So where does openCL and kin all fit in to video? It's more suited for loads that can truly benefit the massive data parallelism that can offer. Such video applications that can benefit from that are jobs like encoding, transcoding, filtering and effects in real-time for example. It can even benefit audio. A real time complete FIR crossover could be done with GPGPU. Currently it cannot be done on a CPU without loosing time accuracy and experiencing some serious latency. Now couple that with multiple channels and the situation gets worse.
Applications such as this is where GPGPU fits into the grand scheme of things. It gets even better when coupled with solutions such as vdpau. Imagine being able to apply all your video effects on the fly in real-time and seeing the end results as well without frame drop, or being able to slam that 2 channel audio into a 7.1 stream with real time processing. Pure sweetness (not to mention diminishing dedicated hardware solutions like creative offers).
openCL is not a replacement for hardware playback of x264 and the likes. Encoding and processing is where it's strength lies.
( I do find Kano being accused of being the death to opensource quite funny though considering his background. He just calls it as he see's it, honest, truthful and usually with out blind fanaticism . Now his clubbing of baby seals on the other hand..... )
BTW Nvidia has openCL drivers to download if your part of their developer program.Last edited by deanjo; 20 April 2009, 10:37 PM. Reason: No matter how hard I tried, I had to go back and correct the typo's and spelling. Qaridarium, it is a video CARD not a CART.
Comment
-
Originally posted by Qaridariumnot true becourse UVD2 only the controll unit to send the file to the shaders and bag!
and the only think what can only the UVD2 unit do is the DRM and Copy protection!
so in fakt! UVD so not save Powerconsuming!
UVD2 is using more powerconsuming becorce UVD2 is a crypto unit what send the VGA signal in the secure way to the HD-ready and HDCP protectet monitor!
UVD2 do not save Powerconsuming!
Wrong wrong wrong. You are confusing decryption with decoding.
Comment
-
Originally posted by deanjo View PostWrong wrong wrong. You are confusing decryption with decoding.
Even Wikipedia thinks it that way.
That's a Unified Video Decoder we're talking about, and it (UVD2) aims at "full bitstream decoding of MPEG4-class video streams".
If using the UVD block, I assume that no shaders are involved in the video stream decoding. It would be... counterproductive, at least. Remember that the different iterations of UVD, are heavily based on the Xilleon decoder unit used in systems like TVs, where there aren't other "things" that could work as video stream decoders. And, yeah, a Xilleon is a complete MIPS SoC (probably a similar technology used by AMD while developing the Wii's northbridge/video chip/embedded MIPS).
OT: Qaridarium, you're making me crazy with that language of yours. Where did you pick it up? (joking)Last edited by Loris; 21 April 2009, 05:56 PM.
Comment
-
Originally posted by QaridariumDHCP Copy protection is decryption.. UVD2 is the Copyprotektion unit in the ATI VGA!
the UVD2 decode wit the shader internaly
As far as power consumption goes, run ANY gpgpu solution and you will see for yourself what the power consumption is like.
and if your thinking of brute force decrypting a BluRay with openCL, lol, well, good luck with that 256 AES encryption with a room full of GPU'sLast edited by deanjo; 21 April 2009, 05:59 PM.
Comment
Comment