GPU processing does really work!

SpeedyVV

Supreme [H]ardness
Joined
Sep 14, 2007
Messages
4,210
I always thought this whole GPU processing was a bunch of marketing crap with no real potential.

But today when my daugher asked me to encode a video for her iPod, i used the standard CPU encoder that got the job done in about 5min and 15sec.

Since we got a HD cam for x'mas i figured I would try this Badaboom software.

Well I was totally blown away. that GPU (CUDA) app encoded the same video in 35secs!!!!

That is insane speed. I mean, I thought this thing would give me maybe 20%-50% improvement, But 9 times the speed!?!?!

Many of you are already saying, dah!!! Well sorry I thought a software called Badaboom using "CUDA" would just be crap.

I am now going to have to find more apps that use CUDA. Somehow I have a feeling that it is a rare thing, which is a shame.

To think that I was looking at i7 for better encoding times :p For what? 100% improvement if that much. At what cost? A minimum of $1000.00 (mobo, cpu, ddr3).

So now my wife is all happy I spent the $450 on that stupid gaming thing :D, It litterally tooks days on her Pentium D 2.8GHz pc to encode the AVCHD files to something she could use. My Q6600 was not much better.

Now, we are talking much much faster.

Anyone else have this experience? Other software? Any that take advantage of multi gpus? Hell I am thinking of getting another GTX 280 now if it improves my encoding time.
 
At this point, there's not a whole lot of software out there for it. Mostly it is used in research and such. Have to remember, the technology is real new, it'll take time for people to start really using it.

Also only some things are good candidates for GPU processing. They have to be highly parallel, single precision floating point, with a problem set that'll fit in video RAM. So things like video encoding, rendering and such are perfect candidates.

Just keep an eye out, I'm sure more software will come out as time goes on. You can always check on nVidia's page. They are extremely pleased with this and pimp things that use CUDA. Also you'll probably see more uptake when OpenCL, which is kinda like cross platform CUDA, gets finished and implemented.

However do remember that it won't make everything faster. If GPUs could do everything CPUs did faster... Well then we'd use them for CPUs. GPUs are specialty vector floating point processors. When a task is something that falls in that category, they'll rip it apart.

As for the usefulness of a second one, it depends. Pretty much anything that can use one, can use two, in theory, since you are talking about highly parallel problems. However it isn't automatic, they'd have to make their program recognize and use both cards.

For now, probably stick with your single card and just keep an eye out for new stuff. There's lots on the CUDA page, but most of it is professional/research apps, not things for consumers.
 
I use CUDA professionally within my company. It is indeed an amazing technology for the right applications, with speed increases ranging from 5x to 250x and beyond. This compared to a quad-core CPU.

For heavily parallel tasks like video encoding, GPGPU is incredibly useful, same with applying filters to video, audio and photos. In the scientific community it has turned things upside-down, as it allows a single PC to run simulations and such in a fraction of the time it used to take a small cluster of PCs.

Also only some things are good candidates for GPU processing. They have to be highly parallel, single precision floating point, with a problem set that'll fit in video RAM. So things like video encoding, rendering and such are perfect candidates.
Note that CUDA 1.3 capability cards (GTX 200+) also support double-precision FP :)
 
IIRC only a small fraction (24?) of the SPs are actually double precision capable.
 
I tried the trial of it. I got nowhere. Every single file type i tried to use was unsupported. So yeah. Looks like im stickign with CPU renderings for a while.

Edit: That was version 1.0, version 1.1 works, and i have to say wow o_o...

Did a full length movie, to iPhone size in about 12 minutes, compared to a CPU which usually does it in 40-50 odd minutes.
 
IIRC only a small fraction (24?) of the SPs are actually double precision capable.
Each SM group of 8 SPs has one double precision FPU. There are 24-30 double precision FPUs in the GTX models (192 SPs = 24, 216 SPs = 27, 240 SPs = 30), each DP FPU capable of 2.6 billion peak FLOPS. That's about a 3x higher peak rating on a GTX 280 than a 3GHz Core 2 Quad for double precision.
----

CUDA is mostly derided by 1) people who don't have cards that support it or 2) people who don't understand what it is and can be used for.
 
Thanks for posting this, I'll definitely have to investigate. At first glance it seems to be NVidia only? Anyone need an EAH3850 in perfect shape? :rolleyes:

I googled a bit and found this thread:
http://www.avsforum.com/avs-vb/showthread.php?t=1048248
You are welcome, and yes it is NVIDIA. Actually all the green of the app is not only a give away that it is NVIDIA only, it is really annoying. ;)

I would not change card right now. Unless you have a need right now. I have a feeling this technology will be GPU agnostic very soon.


I can't believe you thought it was just marketing crap when there are actual review sites that reviewed this

Sorry, I am a [H[ fanboi. If it has not been reviewed positively by [H] it is crap :p

Anandtech did a comparison of Badaboom and ATI Avivo, its probably worth checking out for more info.
http://www.anandtech.com/video/showdoc.aspx?i=3475

I knew of the Anand article, which said that there was a version 1.1, and that is why I downloaded antried it.

And I can confirm that version 1.1 is a lot more stable. I have yet to get it to crash. Mind you, I have not done a lot of testing.

I dont think the product is really 1.1. it is more like beta, and that is why I am reluctant to buy it.

On the other hand I can get value out of it with the functionality that I know works for me. and that is to encode movies and videos to my kids iPods.

It can actually get them done onDemand timelines. So I might just get the product. I certainly have spent more than $30 before for games that sucked. :mad:

Oh yeah, is anyone else going to try it? That was my original question. I really would like to get the opinion of ot[H]ers.
 
I would not change card right now. Unless you have a need right now. I have a feeling this technology will be GPU agnostic very soon.

It already is GPU agnostic. The problem is that the tech is so new that there are many different ways to do the same thing. Nvidia has CUDA, ATI has the Stream SDK (I think that is the name, anyway), everyone has BrookGPU (which supports anything from a Radeon 9700 Pro or GeForce 5200 and up - basically all DX9 cards), and soon there will be the new standard OpenCL (which afaik both camps will support)

Which variant of the tech wins out no one knows :)
 
Does Badaboom or any of the rest support Microsoft MCE/VMC DVR-MS files?
 
i think the problem is having your cpu use all the cores on one video clip.. if it did this then it prob be as fast or close.. my cpu always does 1 video per core and does ipod full video in about 10 mins.. to bad i have ati in most my pc's ill buy a gfx295 when price drops for faster encodes with out i7
 
I tried Badaboom a while back and it was pretty cool. I probably will buy it for reasons listed above.. $30 isn't bad. :)

BTW, that review is from August, and apparently one or two updates have been released, so I'm not sure how reliable it is, as a guy in the AVS thread mentioned he could stick BD movies through it.

EDIT: 20% off. Day 8.

EDIT2: Reviews of Badaboom 1.1? http://anandtech.com/video/showdoc.aspx?i=3475

http://www.pcper.com/article.php?aid=650

http://www.extremetech.com/article2/0,2845,2337062,00.asp

http://www.pcinpact.com/actu/news/48014-test-badaboom-11-perfectionne-gpu-encodage.htm?vc=1
 
i think the problem is having your cpu use all the cores on one video clip.. if it did this then it prob be as fast or close.. my cpu always does 1 video per core and does ipod full video in about 10 mins.. to bad i have ati in most my pc's ill buy a gfx295 when price drops for faster encodes with out i7

Not even close.
 
Guess what day I started using the dnetc CUDA client...

rc5cuda.png


Now they just need to support priority on GPU operations, so I can leave it running as a service all the time without killing my games.
 
Best part is that GPUs are the #1 contributors of work units with F@H, while they also contribute the most work units per GPU. PS3 is next, with CPUs somewhere in the back, even though most people run the CPU client. This means that a single PC crammed full with GPUs will contribute more to F@H than a whole cluster of PCs running the CPU client. To me this is the beauty of using the right tool for the right job :)
 
This encoder is great... for converting iPod/youtube videos....

Its still slower then x264 at worse quality. :confused:

OP, it IS marketing hype.

http://www.anandtech.com/video/showdoc.aspx?i=3374&p=7

My Q6600 encodes a 20min 1080i .ts file faster on INSANE quality with x264+MeGUI, then this does on its max setting.

No, it isn't marketing hype. The tech is real, the tech works. The software (Badaboom) is what sucks (at the moment). Once the software gets straightened out, GPU encoding times will plummet. x264 was buggy and sucked ass when it first came out as well. Give it a couple of years.
 
No, it isn't marketing hype. The tech is real, the tech works. The software (Badaboom) is what sucks (at the moment). Once the software gets straightened out, GPU encoding times will plummet. x264 was buggy and sucked ass when it first came out as well. Give it a couple of years.

I did specifically say this encoder. No doubt there will be better implementations. A GPU enabled x264, now that's something to look forward to.
 
I did specifically say this encoder. No doubt there will be better implementations. A GPU enabled x264, now that's something to look forward to.

Ditto. Get CUDA support in ffmpeg and x264 and then we'll be talking....
 
I did specifically say this encoder. No doubt there will be better implementations. A GPU enabled x264, now that's something to look forward to.

Fair enough. I thought you meant that GPGPU in general was just marking BS, not that Badaboom in its current state was marketing BS. My bad, I apologize :)
 
Ive tried Badaboom multiple times(beta, release 1.0 and 1.1) and im just disappointed.

They have tried to make it so user friendly that it has taken advanced control away.
It does not encode any faster then when using my cpu's.
Sometimes when loading HD video it crashes.

The CUDA/GPGPU technology is great, but IMO the software is just not up to par.
 
I had a chance to use Badaboom beta and was fairly impressed. It made short work of a encoding a video on my laptop. I do look foreward to seeing where this goes.
 
Wow, I do alot of encoding for family videos and whatnot, and I will really have to check this out. Save alot of time! Thanks!
 
Guess what day I started using the dnetc CUDA client...

rc5cuda.png


Now they just need to support priority on GPU operations, so I can leave it running as a service all the time without killing my games.

Dude how long has this been out? I like OGR more but getting to a million blocks is now easier for me now lol Just started on this today after reading this thread and I hope that doing OGR will be a option soon.
 
Dude how long has this been out? I like OGR more but getting to a million blocks is now easier for me now lol Just started on this today after reading this thread and I hope that doing OGR will be a option soon.

Not very long. http://bugs.distributed.net/show_bug.cgi?id=4030 is the tracking bug for the CUDA core. People have been working on it for a while, but I don't think a "real" client has been released until recently (core code is available to the public for testing, but the full network client isn't). http://www.distributed.net/download/prerelease.php has the CUDA client releases from just before Christmas - I'm not sure if they had earlier releases out.
 
I am not very familiar with GPU processing.

Does it work for CAD rendering software like 3dsmax and V-RAY?

Also which does it work better on, ATi or nVidia?
In particular on the 8800GTS 320, or the 3870X2.

K-TRON
 
I am not very familiar with GPU processing.

Does it work for CAD rendering software like 3dsmax and V-RAY?

Also which does it work better on, ATi or nVidia?
In particular on the 8800GTS 320, or the 3870X2.

K-TRON

I think you are a bit confused. This product is for encoding video, thats it. Furthermore, this specific product is NVidia only.
 
GPU processing can be used for anything. Badaboom is video specific, CUDA is nvidia specific, and video-encoding is only one of the things CUDA could do (anyone recalls how WiFi security was cracked recently? it was done using CUDA and a pair of GTX280s).
 
i think the problem is having your cpu use all the cores on one video clip.. if it did this then it prob be as fast or close.. my cpu always does 1 video per core and does ipod full video in about 10 mins.. to bad i have ati in most my pc's ill buy a gfx295 when price drops for faster encodes with out i7

I think that's just the encoding software you use.
I edit and encode videos using Sony Vegas, and most of the codecs I've tried with it, will render with both my cores on a single video.
 
I did specifically say this encoder. No doubt there will be better implementations. A GPU enabled x264, now that's something to look forward to.

While Badaboom may not be perfect, it's quite usable, which is more than can be said for AMD's Avivo at this point. That one is more buggy than any release of Badaboom has ever been. I can't believe AMD released it, it just doesn't seem like a finished product.
 
InvisiBill,do you leave yours running all the time,or shutdown dnetc when you want to play a game?
 
Yeah we have been using GPU processing in the Distributed Computing field for a while now.

Guess what day I started using the dnetc CUDA client...

InvisiBill & OldDeadOne - I didnt know there was a GPU client for RC5-72. Are you guys affiliated with a RC5-72 team? There is a "HardOCP" team for that project if you arent already on that team. Stop down in the DC sub-forum sometime and say hi. We would love to know how the GPU client is working for you guys. :)
 
Best part is that GPUs are the #1 contributors of work units with F@H, while they also contribute the most work units per GPU. PS3 is next, with CPUs somewhere in the back, even though most people run the CPU client. This means that a single PC crammed full with GPUs will contribute more to F@H than a whole cluster of PCs running the CPU client. To me this is the beauty of using the right tool for the right job :)

Well, keep in mind, the GPU has a very narrow path of what data it can crunch but does it BLAZING fast, the CPU can do just about anything but is much slower, and the PS3 is right in the middle of them.

No one should give up their CPU or PS3 to just GPU Fold, they're all the right tools for the job. :)
 
Well, keep in mind, the GPU has a very narrow path of what data it can crunch but does it BLAZING fast, the CPU can do just about anything but is much slower, and the PS3 is right in the middle of them.

No one should give up their CPU or PS3 to just GPU Fold, they're all the right tools for the job. :)

Oh definitely, I wouldn't dream of doing a lot of general purpose stuff and other tasks on a GPU. Sure, it might work, but its performance would be worse than when running it on a 286 :p
 
Back
Top