VA-API Gets New H.264/MPEG-2 Encoding API Support

[agd5f]

I've always wanted to know why Bumblebee technology (or nvidia Optimus) is only available for Notebooks. Perhaps it could be adapted to Desktop machines and work as Lucid on Windows.

You can already do it with desktop cards with provider stuff the latest X server. The question is why would you want it? If you have a faster card, just use it directly rather than having it render and copy the results to the slower card for display.

[YAFU]

The question is why would you want it? If you have a faster card, just use it directly rather than having it render and copy the results to the slower card for display.

For the same reason that people find useful to use LucidVirtu on Windows. You could have the best of both GPU's at the same time, or when needed.

[Aleve Sicofante]

I've always wanted to know why Bumblebee technology (or nvidia Optimus) is only available for Notebooks. Perhaps it could be adapted to Desktop machines and work as Lucid on Windows.

Don't quote me for this, but I don't think they serve the same purpose. Is it Bumblebee/Optimus engaging on or the other graphics on a PER APPLICATION BASIS. My understanding was it selects the app automatically based on load.

[Aleve Sicofante]

You can already do it with desktop cards with provider stuff the latest X server. The question is why would you want it? If you have a faster card, just use it directly rather than having it render and copy the results to the slower card for display.

QuickSync is both faster and of higher quality when it comes to video encoding. It's a weird thing, I know, but even CUDA and OpenCL on discrete cards do a much worse job at that. So when you're going to encode video, you should use the integrated graphics, not the discrete ones.

How do you use one or the other graphics ON A PER APPLICATION BASIS "with provider stuff the latest X server"?

[Kivada]

For the same reason that people find useful to use LucidVirtu on Windows. You could have the best of both GPU's at the same time, or when needed.

Nah, it's always beena bad idea, it only makes a little sense on laptops since pretty much all of Intel's mobile CPUs that are actually being used in mass production hardware have an IGP, thus they can switch off the dedicated GPU. having to pass off the render from the fast GPU to the IGP incurs increased power consumption and increased latency.

On a desktop however, you shouldn't be buying an IGP system if you need a fast GPU anyways. The fast GPUs have made huge strides in idle power consumption every generation, at least for AMD's GPUs. Nvidia's not so much...

[Ericg]

The fast GPUs have made huge strides in idle power consumption every generation, at least for AMD's GPUs. Nvidia's not so much...

Really? I thought Nvidia and AMD had both been dropping their idle temps bit by bit and were about even on power consumption (with Nvidia winning while under load)?

[curaga]

QuickSync is both faster and of higher quality when it comes to video encoding. It's a weird thing, I know, but even CUDA and OpenCL on discrete cards do a much worse job at that. So when you're going to encode video, you should use the integrated graphics, not the discrete ones.

How do you use one or the other graphics ON A PER APPLICATION BASIS "with provider stuff the latest X server"?

QuickSync does not affect graphics. IIRC you can use the quicksync hw already while not touching any of the graphics stuff, or "running the encoder with the integrated graphics". I remember seeing code related to that on the VA list.

[YAFU]

In windows if you are using the discrete card as primary card (which most people do), you can not use Intel Quick Sync without LucidVirtu enabled.
About LucidVirtu, I read reviews with pros and cons, but in general there are more pros.
About QuickSync, you don't get the best quality of encoding, but a good speed. And some people may be more interested in speed than quality. I do not mean you get a bad quality, I mean maybe it's better to use another encoder if you are looking for the best quality, from what I've read.
Some of the links below are tests with little old versions of intel and Lucid, but you get the point:

http://www.tweaktown.com/articles/4651/lucid_virtu_mvp_hyperformance_tested_with_asrock_z77_and_intel_ivy_bridge/index.html

http://www.anandtech.com/show/4113/lucid-enables-quick-sync-with-discrete-graphics-on-sandy-bridge

http://hothardware.com/Reviews/Lucids-Virtu-Software-Combines-Best-of-Both-Worlds/?page=1

http://www.hitechlegion.com/reviews/processors/18317-intel-core-i7-3770k-ivybridge?showall=&start=18

[pixo]

Nah, it's always beena bad idea, it only makes a little sense on laptops since pretty much all of Intel's mobile CPUs that are actually being used in mass production hardware have an IGP, thus they can switch off the dedicated GPU. having to pass off the render from the fast GPU to the IGP incurs increased power consumption and increased latency.

On a desktop however, you shouldn't be buying an IGP system if you need a fast GPU anyways. The fast GPUs have made huge strides in idle power consumption every generation, at least for AMD's GPUs. Nvidia's not so much...

There are use cases for it.
Like running XBMC seat on it.
But yes, you should not waste your money if you are not going to use it.

[Aleve Sicofante]

About QuickSync, you don't get the best quality of encoding, but a good speed. And some people may be more interested in speed than quality.

QuickSync provides the best quality of HARDWARE BASED ENCODERS. It's better than NVIDIA's and AMD's, but of course it's worse than SOFTWARE BASED encoding.

I've been using QuickSync extensively and I can attest that with most encodings the difference will be invisible to the naked eye.