fft3dGPU 0.8.2

[BigDid]

Hi,

Just re-read the thread to see if my problem was already identified but doesn't seems so.

Problem: when feeding job control in Vdubmod for a xvid encode (1 or 2 pass) with a complex script using
fft3dgpu(plane=4, precision=1, sigma=2.4) Vdubmod just exits with no error message right after clicking start in the job control

No problem same script feeded to mpc for previewing or in Vdubmod without job control (save directly so 1 operation at a time and no shut down after encoding from job control)

Script is application from this thread: http://forum.doom9.org/showthread.php?p=1283534#post1283534

No problem with fft3dgpu() before using this complex script, no change in config:
XP-SP3, 2gig Ram, GPU is ATI 4670 with 512Ram, driver is 9.2, fft3dgpu is 0.82

From reviewing the thread I suppose the problem may be related to Ram and/or Vram...

What I have tried so far with no change:
- use setmemorymax(512 or 768) at the beginning of the script; process explorer tells me I have 300Mb or more Ram available
- strictly use only MT(""".......""") before and after fft3dgpu with no setmtmode(2,0)
- use setmtmode(2,0) at the begining and setmtmode(5,0) just before fft3dgpu()
- Not use MT or setmtmode()

Only workaround is changing fft3dgpu by fft3dfilter(plane=4, sigma=2.4), which is slower but works.

Any hints or tips?

Thanks in advance.

Did

[G_M_C]

Ha,

Forgot all about this tread, but i still use this filter now and again. But with my QX9650 "regular" fft3dfilter works good enough too ...

But the reason i "butt into" this thread is the following. I've been readin up on the coming version of DirectX 11. With the coming of DX11, and new feature will be added; The compute Shader. And one of the features explicitly named as beeing ideal for the compute shader is .... FFT transformations (but also matrix transformations, onordered memory lookups/writes/multi-treading etc.).

And now offcourse the question: Will there be e update/rewrite/new version of FFT3dGPU, using these new features of DX 11? Cause I think this will be a big chance for great improvements, wich give new life to this great tool/filter!

[BigDid]

Hi,

Update from my previous post: http://forum.doom9.org/showthread.php?p=1285091#post1285091
In case someone wants the script here are some extracts:

setmemorymax(768)
...
Load_Stdcall_plugin("C:\PROGRA~1\AutoGK\filters\yadif.dll")

movie = mpeg2source("G:\FILMS\Episodes Wolverine\agk_tmp\Wolverine_03.d2v")
...
setmtmode(2,0)
yadif(movie)
autocrop(movie,mode=0,wmultof=4,hmultof=4,samples=10,aspect=new_aspect,threshold=34,samplestartframe=0,leftadd=0,rightadd=0,topadd=0,bottomadd=2)
Spline64resize(out_width,out_height)

mt("""
fastlinedarkenMOD(thinning=0, strength=36)
deblock()
fluxsmoothst(5,5)
maa()
#fft3dfilter(plane=4, sigma=2.4)
""")
setmtmode(6,0)
fft3dgpu(plane=4, bt=4, precision=1, sigma=2.8)

mt("""
#cnr2()
tbilateral()
#hybridfupp(preset="anime2")
limitedsharpenfaster(smode=4, strength=60)
#fastlinedarkenMOD(thinning=0, strength=18)
#dehalo_alpha()
gradfun2db(1.51)
""")

What I have tried since last post:
- use bt=2
- use precision=0
to no avail...

Update: Changed ATI drivers from 9.2 to 9.4 and 9.5 no change...

Did

[Arshad07]

whats this error all about?

http://i37.tinypic.com/rjfdky.jpg

Script

SetMtMode(2)
import("E:\Encode\LSFmod.v1.8.avsi")
import("E:\Encode\MCTemporalDenoise.v1.3.00.avsi")
Import("E:\Encode\YLevels_mt.avsi")
Import("E:\Encode\plugins\plugins\MCBob.avsi")
LoadPlugin("E:\Encode\mvtools-v1.11.4.5\mvtools.dll")
LoadPlugin("E:\Encode\fft3dfilter211\FFT3DFilter.dll")
LoadPlugin("E:\Encode\MaskTools.dll")
LoadPlugin("E:\Encode\masktools-v2.0a36\mt_masktools-25.dll")
LoadPlugin("E:\Encode\mvtools-v2.5.3\mvtools2.dll")
LoadPlugin("E:\Encode\gradfun2db_25_dll_20060315\gradfun2db\gradfun2db.dll")
LoadPlugin("E:\Encode\medianblur084\medianblur.dll")
LoadPlugin("E:\Encode\EEDI2v092\EEDI2\EEDI2.dll")
LoadPlugin("E:\Encode\mvtools2.dll")
LoadPlugin("E:\Encode\Hdragc-1.8.7\AGC.dll")
AVISource("D:\Encoding Folder\KNPh\lag.avi", audio=false)
Spline64Resize(640,248) # Spline64 (Sharp)
ConvertToYV12(matrix="Rec601", interlaced=false)
YLevels(6, 1, 255, 0, 255)
Tweak(hue=0.0, sat=1.2, bright=0, cont=1.0, coring=true, sse=false)
SetMtMode(5)
fft3dgpu(sigma=.5, sigma2=1.5, sigma3=2, sigma4=1, bt=3, bw=30, bh=30, ow=8, oh=8, wintype=2)
SetMtMode(2)
LSFmod(Smode=3,Lmode=1,ss_x=1.0,ss_y=1.0,strength=50)

[Chikuzen]

LoadPlugin("E:\Encode\fft3dfilter211\FFT3DFilter.dll")

FFT3DFilter and FFT3DGPU are different.

[Leak]

whats this error all about?
If I were you I'd take a close look at the forum rules (especially #8) then either delete this post or the one in your other thread before one of the mods notices... *hint hint*

[Arshad07]

FFT3DFilter and FFT3DGPU are different.

I've already installed fft3dgpu.

[Maitreya]

LoadPlugin("E:\Encode\fft3dfilter211\FFT3DFilter.dll")

should be
LoadPlugin("E:\Encode\fft3dfilter211\FFT3dGPU.dll")
i guess

[Zach]

It really sucks that nVidia's built-in auto-downclocking mechanism has made this filter useless (in terms of speed gain vs. the original FFT3DFilter filter). I've now switched back to the original FFT3DFilter because it's actually faster. :)

If it weren't for games, I'd actually consider switching back to my old 8800 GTX and ditch my newer GTX 275. Turning off this new "3D Performance" feature requires a convoluted and complicated series of registry hacks and then, of course, a reboot, all of which is annoying to do whenever the mood strikes me to encode a video.

[Arshad07]

Anyone knows whats causing these problems:-

http://i37.tinypic.com/iwsky8.jpg
http://i33.tinypic.com/wl5di1.jpg
http://i33.tinypic.com/2nteh44.jpg

Script:-


SetMtMode(2)
DGDecode_mpeg2source("D:\Encoding Folder\W.d2v", cpu=4, info=3)
ColorMatrix(hints=true, interlaced=true, threads=0)
tfm(order=1).tdecimate()
crop( 0, 58, -2, -58)

Spline64Resize(848,360) # Spline64 (Sharp)
YLevels(14, 1, 255, 0, 255)
LSFmod(Smode=3,Lmode=1,ss_x=1.0,ss_y=1.0,strength=30)
fft3dgpu(sigma=.5, sigma2=1.5, sigma3=2, sigma4=1, bt=3, bw=30, bh=30, ow=8, oh=8, sharpen=.5,wintype=2)

[Atak_Snajpera]

Anyone knows whats causing these problems:-
This SetMtMode(2). You canot use SetMTmode with GPU plugin!

[Mystery Keeper]

If it weren't for games, I'd actually consider switching back to my old 8800 GTX and ditch my newer GTX 275. Turning off this new "3D Performance" feature requires a convoluted and complicated series of registry hacks and then, of course, a reboot, all of which is annoying to do whenever the mood strikes me to encode a video.

Consider waiting a little and getting (terribly expensive) NVidia card with shader model 5 support. They are designed for parallel computing.

[VincAlastor]

And now offcourse the question: Will there be e update/rewrite/new version of FFT3dGPU, using these new features of DX 11? Cause I think this will be a big chance for great improvements, wich give new life to this great tool/filter!

yes, direct compute is out! nvidia cuda gpu's work with it. ati releases the first dx11 gpu's and no 3ddenoiser support until now. maybe dfttest is dead, but i were very grateful, when tsp update a dx11 fft3dgpu, too!

TSP please say us, can we count with a new fft3dgpu? (eehhhmmm... a banding protection were great, too ;) )

[aegisofrime]

This SetMtMode(2). You canot use SetMTmode with GPU plugin!

Additionally, I'm pretty sure you cannot use x264 with it. As in, for me to be able to use FFT3DGPU properly, I have to use the pre-rendering pass in MeGUI. Create a raw file first, before encoding with x264.

[cogman]

Additionally, I'm pretty sure you cannot use x264 with it. As in, for me to be able to use FFT3DGPU properly, I have to use the pre-rendering pass in MeGUI. Create a raw file first, before encoding with x264.

Ummm... No, this is just wrong. I've used FFT3dGPU with x264 all the time, Why would you think otherwise?

[aegisofrime]

Ummm... No, this is just wrong. I've used FFT3dGPU with x264 all the time, Why would you think otherwise?

In that case there's something wrong with my system, because I get the same errors if I encode with x264.

[TheProfosist]

i am currently having issues implementing fft3dGPU on MPEG2 content this is the code im using for anime:

FFT3dGPU(sigma=4,sharpen=1.0,plane=4,precision=2,degrid=1.0)

when i load i into MeGUI it crashes but when loading vc-1 material it does not

[TheProfosist]

may have been an issue with the content i will experiment a bit more

[TheProfosist]

ok i got it working but with my current setting i get a ton of blocking. current setting are as follows FFT3dGPU(sigma=4,sharpen=1.0,plane=4,precision=2,degrid=1.0) and i would like to use if for anime

[Hiritsuki]

hey guys
I had use fft3dgpu test qulity of anime video
Toaru Kagaku no Railgun (とある科学の超電磁砲)
and to compare with other filter

Source
http://www.Pixsup.com/uploads/fa97fcf9eb.png

WarpSharp()
http://www.Pixsup.com/uploads/97410b0733.png

WarpSharp()
16-235 => 0-255
http://www.Pixsup.com/uploads/8b2cd9d5a0.png

WarpSharp()
TNLMeans()
16-235 => 0-255
even if this filter is highest qulity but it's too slow to use
http://www.Pixsup.com/uploads/7b42c27bf6.png

WarpSharp()
fft3dgpu(bt=1,sigma=2)
16-235 => 0-255
http://www.Pixsup.com/uploads/b1c922f5d6.png

[Konrad Klar]

This SetMtMode(2). You canot use SetMTmode with GPU plugin!

But not all is lost. SetMTMode(5,0) may be inserted before line with FFT3DGPU and then script should be working.

[rehny]

Hellow.
FFT3dGPU doesn't work well.

Photo 1
http://tomorrow01.netfirms.com/fft3gpu_error.JPG
Photo 2
http://tomorrow01.netfirms.com/fft3gpu_error2.JPG

I don't understand why this error happens.

I'm Jpanese, so not good at English.
But I want to do my best cooperation wiht you.

My FFT3dGPU setting is FFT3DGPU(sigma=1.7 ,beta=1 ,plane=4 ,bw=16 ,bh=16 ,ow=8 ,oh=8 ,bt=4 ,mode=0 ,interlaced=false ,wintype=1 ),
however, the error happens on other settings

[PC 1]
Operating System: Windows 7 Enterprise 32-bit (6.1, Build 7600) (7600.win7_gdr.091207-1941)
Language: Japanese (Regional Setting: Japanese)
Processor: Intel(R) Core(TM)2 Quad CPU Q9550 @ 2.83GHz (4 CPUs), ~2.8GHz
Memory: 4096MB RAM
Available OS Memory: 3328MB RAM
DirectX Version: DirectX 11
Card name: NVIDIA GeForce GTS 250
Manufacturer: NVIDIA
Chip type: GeForce GTS 250
Display Memory: 1905 MB
Dedicated Memory: 497 MB
Shared Memory: 1407 MB
Driver Name: nvd3dum.dll,nvwgf2um.dll,nvwgf2um.dll
Driver File Version: 8.16.0011.9107 (English)
Driver Version: 8.16.11.9107
DDI Version: 10
Driver Model: WDDM 1.1
Driver Attributes: Final Retail
Driver Date/Size: 9/28/2009 07:12:20, 7614056 bytes


[PC 2]
Operating System: Windows XP Home Edition (5.1, Build 2600) Service Pack 2 (2600.xpsp_sp2_rtm.040803-2158)
Language: Japanese (Regional Setting: Japanese)
Processor: AMD Phenom(tm) II X4 955 Processor, MMX, 3DNow (4 CPUs), ~3.2GHz
Memory: 3072MB RAM
DirectX Version: DirectX 9.0c (4.09.0000.0904)
Card name: NVIDIA GeForce GTS 250
Manufacturer: NVIDIA
Chip type: GeForce GTS 250
Display Memory: 512.0 MB
Driver Name: nv4_disp.dll
Driver Version: 6.14.0011.9062 (English)
DDI Version: 9 (or higher)
Driver Attributes: Final Retail
Driver Date/Size: 8/17/2009 00:57:00, 5845760 bytes


[PC 3]
Operating System: Windows 7 Home Premium 64-bit (6.1, Build 7600) (7600.win7_rtm.090713-1255)
Language: Japanese (Regional Setting: Japanese)
Processor: AMD Athlon(tm) II X4 605e Processor (4 CPUs), ~2.3GHz
Memory: 4096MB RAM
Available OS Memory: 3582MB RAM
DirectX Version: DirectX 11
Card name: ATI Radeon HD 4200
Manufacturer: ATI Technologies Inc.
Chip type: ATI display adapter (0x9710)
Display Memory: 2124 MB
Dedicated Memory: 589 MB
Shared Memory: 1534 MB
Driver Name: atiumd64.dll,atidxx64.dll,atiumdag,atidxx32,atiumdva,atiumd6a.cap,atitmm64.dll
Driver File Version: 8.14.0010.0685 (English)
Driver Version: 8.641.0.0
DDI Version: 10.1
Driver Model: WDDM 1.1
Driver Attributes: Final Retail
Driver Date/Size: 7/30/2009 02:14:56, 4383744 bytes


[PC 4]
Operating System: Windows Vista? Ultimate (6.0, Build 6002) Service Pack 2 (6002.vistasp2_gdr.091208-0542)
Language: Japanese (Regional Setting: Japanese)
Processor: Intel(R) Xeon(R) CPU E5450 @ 3.00GHz (8 CPUs), ~3.0GHz
Memory: 4094MB RAM
DirectX Version: DirectX 11
Card name: NVIDIA GeForce 8600 GTS
Manufacturer: NVIDIA
Chip type: GeForce 8600 GTS
Display Memory: 2030 MB
Dedicated Memory: 239 MB
Shared Memory: 1791 MB
Driver Name: nvd3dumx.dll,nvwgf2umx.dll,nvwgf2umx.dll,nvd3dum,nvwgf2um,nvwgf2um
Driver Version: 8.16.0011.9107 (English)
DDI Version: 10
Driver Attributes: Final Retail
Driver Date/Size: 9/27/2009 23:12:22, 9441384 bytes


[PC 5]
Operating System: Windows XP Professional x64 Edition (5.2, Build 3790) Service Pack 2 (3790.srv03_sp2_gdr.091215-1207)
Language: Japanese (Regional Setting: Japanese)
Processor: Dual Core AMD Opteron(tm) Processor 275 (4 CPUs), ~2.2GHz
Memory: 2046MB RAM
DirectX Version: DirectX 9.0c (4.09.0000.0904)
Card name: NVIDIA GeForce 7800 GT
Manufacturer: NVIDIA
Chip type: GeForce 7800 GT
Display Memory: 256.0 MB
Driver Name: nv4_disp.dll
Driver Version: 6.14.0010.7777 (English)
DDI Version: 9 (or higher)
Driver Attributes: Final Retail
Driver Date/Size: 7/20/2005 21:07:00, 5767168 bytes

Only PC 5 has no problem.
If you want to know my PC's other specs, plase tell me.

Thank you.

[osgZach]

You may also want to post your entire AVS script, in case something in it is causing the problem.

[rehny]

I use AviSynth 2.5.8 and don't use MT.

one of my avs scripts.

SetMemoryMax(1024)
PPath = "C:\encode\AviSynth\"
TS = "F:\ts\"
ENCODE_NAME = "test"
LoadPlugin(PPath + "FFT3dGPU.dll")
LoadPlugin(PPath + "EEDI2.dll")
LoadPlugin(PPath + "TDeint.dll")
LoadPlugin(PPath + "TIVTC.dll")
LoadPlugin(PPath + "DGDecode.dll")
video = MPEG2Source(TS + ENCODE_NAME +".d2v")
audio = WavSource(TS + ENCODE_NAME + ".wav")
AudioDub(video, audio)
AssumeTFF()
Trim(a,b)++Trim(c,d)++Trim(e,f)
TIVTC24P2()
BlackmanResize(1280,720)
FFT3DGPU(sigma=1.7 ,beta=1 ,plane=4 ,bw=16 ,bh=16 ,ow=8 ,oh=8 ,bt=4 ,mode=0 ,interlaced=false ,wintype=1)
return last

function TIVTC24P2(clip clip){
Interp = clip.AssumeTFF().SeparateFields().SelectEven().EEDI2(field=1)
Deinted=clip.TDeint(order=1,field=1,edeint=Interp)
clip = clip.TFM(mode=6,order=1,PP=7,slow=2,mChroma=true,clip2=Deinted)
clip = clip.TDecimate(mode=1)
return clip
}

[osgZach]

I'm sure someone with the proper knowledge will be along, but have you tested by removing the first AssumeTFF() or putting it higher up the chain (maybe after loading the source) ? and have you tested calling FFT3DGPU before your IVTC instead ?

[rehny]

Thank you, osgZach.

have you tested by removing the first AssumeTFF() or putting it higher up the chain (maybe after loading the source) ?
I hadn't done, so I tried to remove the first AssumeTFF(), but the error happened. :(

have you tested calling FFT3DGPU before your IVTC instead ?
Yes.
For more information, please see below.

FFT3DGPU(sigma=1.7 ,beta=1 ,plane=4 ,bw=16 ,bh=16 ,ow=8 ,oh=8 ,bt=4 ,mode=0 ,interlaced=true ,wintype=1)
TIVTC24P2()
BlackmanResize(1280,720)

However, I was unable to solve the problem.

[osgZach]

In the meantime, I suggest doing the following.

Go through your script line for line, and disable one line at a time until the problem dissapears (or do you know for sure disabling FFT3DGPU will make it go away?).

Hopefully this will get you proper output at some point and you can start turning stuff back on..

If its without a doubt FFT3DGPU, start with the least custom options as possible (standard call " FFT3DGPU()") and add back custom values one at a time to see if you can find the problem.

[rehny]

I'm sure that FFT3DGPU causes the problem.
Unfortunately, I also made sure the error happens with FFT3DGPU().

[osgZach]

Then all you can do is call it without your custom options ( FFT3DGPU() ) and see if it still gives the corruption.

If it does, try redownloading it, or trying different video drivers. If the problems go away, you'll have to add your custom options to the call one by one, until you can isolate which setting, or combination of settings is doing it.. (or until someone who knows hopefully clues you in to what the problem is)

[asarian]

Is the GeForce GTX 260 supported? I'm getting the following error, using MCTemporalDenoise (with 'GPU=true'):


avs [info]: 1920x796p 1:1 @ 24000/1001 fps (cfr)
x264 [info]: using SAR=1/1
x264 [info]: using cpu capabilities: MMX2 SSE2Fast SSSE3 Cache64
x264 [info]: profile High, level 4.1
x264 [error]: malloc of size 6856704 failed/s, eta 19:40:54
x264 [error]: x264_encoder_encode failed


With 'GPU=false' is doesn't fail.

Thanks.

[kemuri-_9]

MCTemporalDenoise (with 'GPU=true'):

x264 [error]: malloc of size 6856704 failed/s, eta 19:40:54
x264 [error]: x264_encoder_encode failed

With 'GPU=false' is doesn't fail.

how is it saying anything about your gpu card being not supported?
all x264 said is "I can't allocate any more memory and thus failed to work".
sounds like using the gpu version uses more memory which is causing x264 to error like that.

[asarian]

how is it saying anything about your gpu card being not supported?
It isn't; which is why I'm asking. :)


all x264 said is "I can't allocate any more memory and thus failed to work".
sounds like using the gpu version uses more memory which is causing x264 to error like that.
Well, isn't malloc specified in bytes? So, x264 is telling me it can't allocate a measly 6MB? (out of 4G, of which it only use ca. 1.4G).

My understanding of fft3dGPU (which is nearly 0, as I'm just using it for the first time), is that it will try and allocate video memory. So, if an entire card weren't supported, I could see a 6MB chunk allocation fail too.

[Adub]

Hmmm...I've had issues with fft3dGPU and HD content. The main problem is that you can only call it once in a script. With a compound script like MCTemporalDenoise, I can see it being called on more than one occasion, which my be your problem.

[Zep]

It isn't; which is why I'm asking. :)

My understanding of fft3dGPU (which is nearly 0, as I'm just using it for the first time), is that it will try and allocate video memory. So, if an entire card weren't supported, I could see a 6MB chunk allocation fail too.

Make sure you put a SetMTMode(5,threads) before it. I do this and have never had a problem on 1080 stuff and I only have a 512 8800GT. Like the poster above me said you should only have 1 FFT3DGPU call. just that 1 call uses 90% of the ram on my card. (I use rivatuner to see what is happening)


This works well for me:

threads=8
SetMTMode(3,threads)
mpeg2source(whatever)

SetMTMode(5,threads)
FFT3DGPU(whatever)

SetMTMode(2,threads)
Mode2filters(whatever)

[asarian]

Make sure you put a SetMTMode(5,threads) before it. I do this and have never had a problem on 1080 stuff and I only have a 512 8800GT. Like the poster above me said you should only have 1 FFT3DGPU call. just that 1 call uses 90% of the ram on my card. (I use rivatuner to see what is happening)


This works well for me:

threads=8
SetMTMode(3,threads)
mpeg2source(whatever)

SetMTMode(5,threads)
FFT3DGPU(whatever)

SetMTMode(2,threads)
Mode2filters(whatever)
Thanks, Zep! I'll give that a try after the current 'regular' encoding job finishes.

My card is only a GTX260 (only one of its kind that would still fit on my Commando motherboard after my 8800GTX died). So, I'm not sure I could gain much from the GPU route (except,. maybe, an increase in decibel from the GPU fan); but it's worth finding out.

[asarian]

Make sure you put a SetMTMode(5,threads) before it.
Hmm, just realized I don't use MT AviSynth. :) So, that's not going to work.

[aegisofrime]

Hmm, just realized I don't use MT AviSynth. :) So, that's not going to work.

Is there a reason why you choose not to use it?

[Zep]

Hmm, just realized I don't use MT AviSynth. :) So, that's not going to work.

well you should :D MUCH faster! (though you will need to tweak scripts to find the sweet spot for speed)
GPU version is much faster and it frees up ram and it mean less CPU used so more CPU for filters.

Anyway, my guess is your parms are too memory hungry. You need to change block size and get rid of overlap etc... to be able
to do 1080 (how much ram is on that GTX260? (I just looked 896. that should be enough even for defaults) hmmm...) I suggest you
get rivatuner and take a look at WHAT is going on.


good luck

[asarian]

well you should :D MUCH faster! (though you will need to tweak scripts to find the sweet spot for speed)
GPU version is much faster and it frees up ram and it mean less CPU used so more CPU for filters.
To tell ya the truth, I briefly looked at MT AviSynth when it first came out; it was kindy buggy then. And besides, I figured the vast majority of CPU time is usurped by x264 anyway; so who cares whether the encoding takes 80 hours or 79.55? That thinking may need an overhaul, though. :)

Anyway, my guess is your parms are too memory hungry. You need to change block size and get rid of overlap etc... to be able
to do 1080 (how much ram is on that GTX260? (I just looked 896. that should be enough even for defaults) hmmm...) I suggest you
get rivatuner and take a look at WHAT is going on.
I'm not using it directly myself. It's part of the MCTemporalDenoise script. And to tell you the truth again, much as I love encoding speed, I like quality better. Meaning I'd hate to cripple brilliant scripts like MCTemporalDenoise, just to cater to the needs of a GPU plugin.

From what I currently understand, it actually really just looks to be a x264 memory issue (which is to say, x264 + AviSynth + MCTemporalDenoise all competing for the 1x 2GB-per-process limit on my 32-bit OS). At least, I no longer think it's really graphics memory I'm running out of. So, I'm going to experiment with avs2yuv and pipebuf a bit, and maybe directly try so on a 64-bit OS.

[asarian]

Yep, it was a x264 memory issue. :) Or rather, a Windoze 32-bit 2 GB-per-process limit issue. I solved it by letting avs2yuv.exe pipe its output to x264 (a very handy memory separation trick I learnt recently!).

Can't say 'GPU=true' is faster for me, though. In fact, it seems to make no difference at all, LOL, both memory + speed wise. So, maybe it's not even using GPU after all. Kinda weird, actually.

[G_M_C]

From this thread: SEt's Avisynth 2.5.8 MT compiled for *X86_64*, Latest Build 4/16/2010 (http://forum.doom9.org/showthread.php?t=152800)

I found that there is a 64 bit verion available of fft3dgpu. But it is also mentioned that (quote)

note:The hlsl (shader program) file is edited from the original to adhere to pixel shader 3.0 syntax rules.


I've take this PS3.0 HSL file, and i have been using it with fft3Dgpu version 0.82 (i installed 0.82, and copied the "new" hsl file over the original one). It seems to work, and i have the idea that it is much faster than the original HSL. At least on my HD5770.

Can someone confirm you can simply "update" the HSL on the original 32-bit fft3dgpu, without errors or problems ?

[Hiritsuki]

Wow!! thanks G_M_C, I will test it on 32-bit fft3dgpu tomorow
if it's usable, it's help me alot of encode time.

[Hiritsuki]

To G_M_C
it's can be used on 32-bit fft3dgpu
but I don't know it's faster or slower than original hlsl.

[G_M_C]

To G_M_C
it's can be used on 32-bit fft3dgpu
but I don't know it's faster or slower than original hlsl.

I've tested and you are right;

AviSynth test file

DirectShowSource("ïnput_1080p.grf",fps=23.976,audio=false) \ (grf opens blu-ray m2ts with CoreAVC2.0, output YV12)

fft3dgpu(precision=2,plane=???,bw=32,bh=32,ow=16,oh=16)

AssumeFPS (24000,1001)


Plane=4
HSL2.0 12.8~12.9 fps
HSL3.0 13.0~13.1 fps

Plane 1 & 3
HSL2.0 and HSL3.0 scored almost equal; 23.8 fps

This on my machine;
QX9650 @ non-OC 9x333 for fair comparison with other users
Ati HD5770 @ non-OC
XP 32-SP3

[Boulder]

Could you please run a comparison against FFT3DFilter with same settings?

[Hiritsuki]

I've tested and you are right;

AviSynth test file

DirectShowSource("ïnput_1080p.grf",fps=23.976,audio=false) \ (grf opens blu-ray m2ts with CoreAVC2.0, output YV12)

fft3dgpu(precision=2,plane=???,bw=32,bh=32,ow=16,oh=16)

AssumeFPS (24000,1001)


Plane=4
HSL2.0 12.8~12.9 fps
HSL3.0 13.0~13.1 fps

Plane 1 & 3
HSL2.0 and HSL3.0 scored almost equal; 23.8 fps

This on my machine;
QX9650 @ non-OC 9x333 for fair comparison with other users
Ati HD5770 @ non-OC
XP 32-SP3

Wow! I also test it, but I use no Plane to run it.
My video frame size is 1920*1080.

fft3dgpu() # used original to run

HSL2.0 15.7~16.4fps
HSL3.0 16.1~16.6fps

This on my PC
i7-980X @ oc 200*21
nVidia GTX275 @ oc 800-1680-1200
Win7 64bit (use with 32-bit avisynth)

[G_M_C]

Wow! I also test it, but I use no Plane to run it.
My video frame size is 1920*1080.

fft3dgpu() # used original to run

HSL2.0 15.7~16.4fps
HSL3.0 16.1~16.6fps

This on my PC
i7-980X @ oc 200*21
nVidia GTX275 @ oc 800-1680-1200
Win7 64bit (use with 32-bit avisynth)

fft3dgpu(), default uses plane=1 afaik ;)

@Boulder, i'll set it up.

[Hiritsuki]

Tomorow I will use new graphic card to run this test.
Is the MSI N465GTX TwinFrozer Gold Edtion (can update bios to GTX470).
Think that will more difference with HSL2.0 to HSL3.0.

[Hiritsuki]

I had try GTX465 to run this and find a result is HSL2.0 or 3.0 is can't run more than more faster in new Graphic Card than this FFT3DGPU devlop time.
Old Graphic Card that like 6600GT or 7600GT is the FFT3DGPU devlop time.
If HSL 4.0 or 5.0 or framework for renovation maybe run more than more faster.

New test below

My video frame size is 1920*1080.

fft3dgpu() # used original to run

HSL2.0 16.1~16.4fps
HSL3.0 16.5~16.7fps

This on my PC
i7-980X @ oc 200*21
nVidia GTX465 @ oc 800-1680-1000
Win7 64bit (use with 32-bit avisynth)

[Hiritsuki]

Anyone used NVperf?
Who can tell me what this effect is?