FLACCL: CUDA-enabled FLAC encoder by Gregory S. Chudov (prev. FlaCuda), Formerly "lossless codecs and CUDA" Options

[Gregory S. Chudo...]

I'll try to find a HD4xxx GPU to test it on...

[viktor]

I'll try to find a HD4xxx GPU to test it on...

thanks gregory.

[hardboiled]

Hi Gregory,

Any chance for a Fermi based FLACUDA in the near future? I screwed myself by upgrading from my GTX260 to the GTX460 and didn't realize FLACUDA didn't support Fermi based cards. Thanks!

[Gregory S. Chudo...]

I'm currently testing next version on GTX460. Looks good so far. Will probably release it next week.

[Metroid]

I'm currently testing next version on GTX460. Looks good so far. Will probably release it next week.

Look forward to seeing it. I just got a GTX 460 today, now I can try it on.

My 8800GT has speeds up to 220 on cuetools flacuda

This post has been edited by Metroid: Nov 6 2010, 04:36

[viktor]

I'll try to find a HD4xxx GPU to test it on...

couldn't i help you with debugging?

[Gregory S. Chudo...]

I think i can find one at work. I'll PM you if i can't.

[JapanAudio]

I wouldn't kill nVidia just yet. AFAIK, as of now, it is the only card that supports GPU video transcoding [...]

I've transcoded a bunch of videos with ATI Avivo.

[IgorC]

I wouldn't kill nVidia just yet. AFAIK, as of now, it is the only card that supports GPU video transcoding [...]

I've transcoded a bunch of videos with ATI Avivo.

Maybe it's not quite on topic but it might be informative.
GPU video transcoding is different. Mainly it is used for H.264 lossy coding. Untill now every GPU H.264 encoder was very inferior to good CPU H.264 encoder. Speed comes at cost of quality.

This post has been edited by IgorC: Nov 7 2010, 02:21

[JapanAudio]

I wouldn't kill nVidia just yet. AFAIK, as of now, it is the only card that supports GPU video transcoding [...]

I've transcoded a bunch of videos with ATI Avivo.

Maybe it's not quite on topic but it might be informative.
GPU video transcoding is different. Mainly it is used for H.264 lossy coding. Untill now every GPU H.264 encoder was very inferior to good CPU H.264 encoder. Speed comes at cost of quality.

Yeah... I woudn't recommend it for general use. It's just fast. That thing is basically just a bunch of ALUs and hardwired processing units. No tweaking at all for x264.

[bb10]

Any plans for a CUDA/CL port of WavPack?

[Gregory S. Chudo...]

I don't think it's possible. FLAC seems to be the only codec properly suitable for parallel encoding.

[Gregory S. Chudo...]

FLACCL 0.2:  flaccl02.rar ( 110.44K ) Number of downloads: 612


Supported devices:
1) NVIDIA Geforce 4XX (Fermi)
Requires fresh drivers (e.g. http://www.nvidia.com/object/win7-winvista...hql-driver.html)
2) ATI Radeon HD 5XXX
Requires fresh drivers (e.g. http://sites.amd.com/us/game/downloads/Pag...on_win7-64.aspx)
Be sure to download "AMD Catalyst™ Accelerated Parallel Processing (APP) Technology Edition", not "Catalyst Software Suite (64 bit) English Only". This contains both display drivers and opencl.
3) Multicore CPU
Requires "AMD Catalyst™ Accelerated Parallel Processing (APP) Technology Edition".
Option to use CPU encoding: --opencl-type cpu
Option to select opencl platform if you have both NVIDIA and AMD installed on single computer: --opencl-platform "ATI Stream"

[Qest]

I see more benchmarks on the horizon.

[bb10]

I don't think it's possible. FLAC seems to be the only codec properly suitable for parallel encoding.

I thought it wouldn't harm to ask:
http://www.hydrogenaudio.org/forums/index....st&p=730471

[Wombat]

Thanks Gregory for your efforts!
I wanted to report my GTX260 does well with FLACCL 0.2 even if it is no Fermi card
Using the recent 260.99 driver it has similar speed as FlaCuda 0.91 here. The encoded files only differ slightly in size. The few files i threw in show the FLACCL files are some kb bigger at -8 but that may be 0 when using a big enough test corpus.

[Metroid]

FLACCL 0.2:  flaccl02.rar ( 110.44K ) Number of downloads: 612

Hi Greg, thanks for all your efforts. Any chance of the next Cuetools release to include this feature?

[Bad Monkey]

Well I have my GTX 460 now, and am trying the new ver above, but still can't get it to run as quickly as others.

Encoding the 13 tracks from Pink Floyd's Final Cut, as before, from WAVS with the settings

CODE

-8 - -o %d

I get speeds of less than 160x

(That's with 4 threads, it is [much] slower if I set Foobar to just 1... what is the story there?)

This vs 122x with regular Flac -8 on my Q6600, which maxes out all four cores.

To try to minimize the HDD factor, I transcoded a single FLAC of Mike Oldfield's Amarok (60 mins long), but only got 86x... can the HDD be slowing it down this much? The Amarok FLAC is 330 MB... Windows will copy the file in mere seconds...

[Wombat]

Well I have my GTX 460 now, and am trying the new ver above, but still can't get it to run as quickly as others.

Encoding the 13 tracks from Pink Floyd's Final Cut, as before, from WAVS with the settings

CODE

-8 - -o %d

I get speeds of less than 160x

(That's with 4 threads, it is [much] slower if I set Foobar to just 1... what is the story there?)

This vs 122x with regular Flac -8 on my Q6600, which maxes out all four cores.

To try to minimize the HDD factor, I transcoded a single FLAC of Mike Oldfield's Amarok (60 mins long), but only got 86x... can the HDD be slowing it down this much? The Amarok FLAC is 330 MB... Windows will copy the file in mere seconds...

try "--cpu-threads 2 -8 - -o %d" It isn´t wise to let 4 threads of these GPU encoders run. Just give one instance of Flaccl use 2 additional threads via --cpu-threads. You may even try 3 threads.
On my HDD i also get more hiccups but encoding is above 200x all the time. My SSD is much faster even if i only have an old 965P, Q9550, GTX260.

Lets see if case finds some time, his graph is the only reference atm.

[Bad Monkey]

I hear you but 4 threads was the fastest result I could get. 1 thread, or 1 thread + CPU threads, give slower results. Go figure.

I don't buy the HDD bottleneck issue - the file sizes are tiny! as above, 330 MB file can be duplicated on the same HDD by Windows in mere seconds. And if it was the HDD, wouldn't all variations of settings hit the same capped speed factor?

Also, these results are faster than my old 8800 GT, and faster now than FLAC on the Q6600 CPU, again suggesting the slow results on both video cards are not being capped by some separate factor.

[Wombat]

Sorry, can´t help you with your PC then. Just tried with Amarok in one single file. On my SSD it comes out mostly between 350x-400x speed, encoded from HDD i get hiccups in encoding and end up mostly ~250x
This is done on the commandline, not in foobar. I think the speed result implementation from Gregory is right.

[Case]

Here's my benchmarks with Core i7 940 + GTX 285 for FlaCUDA and HD 5870 for FlacCL. I used --slow-gpu and --cpu-threads 2 parameters with CL compressor as that seemed to give the fastest speed.

[Metroid]

m:\flac>CUETools.FLACCL.cmd.exe --cpu-threads 2 -8 1.wav -o 1.flac
FLACCL#0.2, Copyright © 2010 Gregory S. Chudov.
This is free software under the GNU GPLv3+ license; There is NO WARRANTY, to
the extent permitted by law. <http://www.gnu.org/licenses/> for details.
Filename : 1.wav
File Info : 44100kHz; 2 channel; 16 bit; 01:01:22.1470000
Results : 496.35x; 377954376 bytes in 00:00:07.4184243 seconds;

C:\flac>CUETools.FLACCL.cmd.exe --cpu-threads 2 -8 1.wav -o 1.flac
FLACCL#0.2, Copyright © 2010 Gregory S. Chudov.
This is free software under the GNU GPLv3+ license; There is NO WARRANTY, to
the extent permitted by law. <http://www.gnu.org/licenses/> for details.
Filename : 1.wav
File Info : 44100kHz; 2 channel; 16 bit; 01:01:22.1470000
Results : 496.28x; 377954376 bytes in 00:00:07.4194243 seconds;

C: = System SSD
M: = Internal Storage HDD Samsung F3 1TB

No difference here. my 8800GT used to give me a maximum of 220. This thing is amazing. I look forward for the Cuetools version.

[Wombat]

No difference here. my 8800GT used to give me a maximum of 220. This thing is amazing. I look forward for the Cuetools version.

Interesting. Having that close numbers between SSD and HDD makes me wonder. Do you have Win7 and some kind of Readyboost kicking in?

This post has been edited by Wombat: Nov 10 2010, 22:49

[Metroid]

Interesting. Having that close numbers between SSD and HDD makes me wonder. Do you have Win7 and some kind of Readyboost kicking in?

I'm using the Windows 7 64 Professional but all HDD features are disabled, the IntelSSDToolbox did the trick and that is one way of disabling all, the other way is manual via registry but I bet you know all this. Well even the pagefile is disabled which makes me wonder if that was the case.