GPU encoding possibly not ready for nominal use yet in any operating system

For those who have not seen the article, I thought it a good read: The wretched state of GPU transcoding - Slideshow | ExtremeTech

In essence the article notes for MS-Windows users that they should not rely/invest into GPU encoding … where to quote the article

If you want a video encoder that’ll run on virtually any system, has well-thought, easy-to-use presets, and neatly balances quality and file size, go download Handbrake. It’s fast, free, and efficient — it just doesn’t use the GPU.

having typed that … there is a big thread on slashdot debating the accuracy of this: The Wretched State of GPU Transcoding - Slashdot

… of course as a user who does a lot of encoding, one of my disappointments with GNU/Linux is neither the open source nor proprietary video drivers openly provide support for GPU encoding (to off loading the encoding from the CPU to the GPU). [Where I refer to ‘encoding’ and not ‘decoding’, as GNU/Linux does offer some offloading of ‘decoding’ to the GPU]. As it turns out, MS-Windows with its drivers that DO offer such encoding offloading, but debateably do not do such a great job. Hence from a GNU/Linux vs MS-Windows video driver perspective, there is no need to be jealous (yet) of the GPU encoding capability of MS-Windows proprietary graphic drivers. … Note I am not talking about decoding nor general graphic driver performance, where from what I have read in the most part, for decoding and general performance, MS-Windows proprietary graphic drivers are still significantly superior to those of GNU/Linux.

… but to get back to my point behind posting (about encoding) … it turns out with the lack of GNU/Linux GPU encoding, with myself being a GNU/Linux user, I may not be missing that much by my still using Handbrake on GNU/Linux.

On 2012-05-09 22:36, oldcpu wrote:
> … but to get back to my point behind posting (about encoding) … it
> turns out with the lack of GNU/Linux GPU encoding, with myself being a
> GNU/Linux user, I may not be missing that much by my still using
> Handbrake on GNU/Linux.

Without having read the article yet, there is a technique where the
graphical GPU is used to do some general computation not related at all to
graphics: it comes from some user application that needs heavy maths. That
is, when the computer is not displaying complex graphics, the GPU is used
to help the CPU, contrary to the normal usage.


Cheers / Saludos,

Carlos E. R.
(from 11.4 x86_64 “Celadon” at Telcontar)

On 5/9/2012 4:43 PM, Carlos E. R. wrote:
> On 2012-05-09 22:36, oldcpu wrote:
>> … but to get back to my point behind posting (about encoding) … it
>> turns out with the lack of GNU/Linux GPU encoding, with myself being a
>> GNU/Linux user, I may not be missing that much by my still using
>> Handbrake on GNU/Linux.
>
> Without having read the article yet, there is a technique where the
> graphical GPU is used to do some general computation not related at all to
> graphics: it comes from some user application that needs heavy maths. That
> is, when the computer is not displaying complex graphics, the GPU is used
> to help the CPU, contrary to the normal usage.
>

That is all possible. Just think of the old C64 days where fractals
where computed with blanking the screen and use the floppydrive for
computing power. :slight_smile:
Nothing new really. Its just not mainstream.


ACCESS DENIED…
Linux Counter: 548299 https://linuxcounter.net/

/_/
/ o o
/~
=ø= /
(______)__m_m) el cato

Am 10.05.2012 04:29, schrieb JoergJaeger:
> That is all possible. Just think of the old C64 days where fractals
> where computed with blanking the screen and use the floppydrive for
> computing power. :slight_smile:
> Nothing new really. Its just not mainstream.
>
GPU programming for high performance computing is mainstream on Linux as
well as on Windows. It is just not very well standardized. The main
players in the graphics world (AMD and NVidia) have their own standards
for that. CUDA for Nvidia is well established, the SDK is available for
Linux and works. OpenCL is an attempt for an open standard in that field.
Not every GPU supports that kind of things, until now it is rather
proprietary thing.
So no wonder that not that many open source programmers build in support
for it in their software.


PC: oS 12.1 x86_64 | i7-2600@3.40GHz | 16GB | KDE 4.8.3 | GeForce GT 420
ThinkPad E320: oS 12.1 x86_64 | i3@2.30GHz | 8GB | KDE 4.8.3 | HD 3000
eCAFE 800: oS 12.1 i586 | AMD Geode LX 800@500MHz | 512MB | KDE 3.5.10

Hi
The BOINC client has been using CUDA for a while now. I played with the
SDK awhile back with my EN8600GTS card, always thought about playing
some more with it, if I had the time…


Cheers Malcolm °¿° (Linux Counter #276890)
openSUSE 12.1 (x86_64) Kernel 3.1.10-1.9-desktop
up 5 days 1:57, 5 users, load average: 0.01, 0.02, 0.05
CPU Intel i5 CPU M520@2.40GHz | Intel Arrandale GPU

Without having read the article yet, there is a technique where the
graphical GPU is used to do some general computation not related at all to
graphics: it comes from some user application that needs heavy maths. That
is, when the computer is not displaying complex graphics, the GPU is used
to help the CPU, contrary to the normal usage.

That technique was used by a company I used to work (a few years ago), with some big screen LED servers (running Linux) they had developed. The actual video data was streamed to LED panels as UDP data.