[RESOLVED] DGIndexNV.ini GTX 1070 XTREME best cuda use

[pacoor]

Hello
DGIndexNV.ini to put in, to use the full potential of the gpu 1070 xtreme gtx?
I'm using bdrebuilder and see that mostly uses the cpu and gpu uses only.
I would like to use the gpu, but the cpu and 1920 Cudas
Thank you


1b81 GP104 [GeForce GTX 1070]

[admin]

Testing on my GT 620 shows CUVID and CUDA to be indistinguishable and they give the following performance for a 2160p AVC stream (GT 620 is pretty crappy nowadays ):

GPU 47%
VPU 95%
CPU 9%
FPS 32

DXVA mode gives:

GPU 72%
VPU 95%
CPU 5%
FPS 25

CPU 5% is pretty good for this high-resolution stream. You'll have lots of CPU to spare for encoding.

Your decision is do you want to trade CPU for frame rate. If your bottleneck is not the decoding frame rate, then you might consider DXVA mode. DXVA mode, however, limits you to one instance of DGSource() in your scripts, but that shouldn't be a big deal because multi-instance capability is rarely used in practice. Beware, though, not all video cards may behave the same, so you really should do your own testing.

I see that the List GPUs dialog is not displaying the GPU type correctly. Are you using build 2052 with the updated gpus.txt file included therein? If so, probably I am not taking into account case for the a-f hex digits. I'll check and slipstream if needed.

[pacoor]

yes i use version DGDecNV build 2052 , but I have to put in DGIndexNV.ini to force GP104 appears 1b81 [GeForce GTX 1070]

[admin]

yes i use version DGDecNV build 2052 , but I have to put in DGIndexNV.ini to force GP104 appears 1b81 [GeForce GTX 1070]

I don't understand that. What are you putting in the INI file? I don't know of anything in there that could affect this, other than the GPU selection, but that should be correct by default because you have only one GPU. Did you mean you edited the gpus.txt file to make it uppercase (1B81)? If so, that's a hack and won't fix all the other entries with hex digits.

Anyway, I made a slipstream to do the comparison correctly. Please download it and try it.

[pacoor]

yes i use version DGDecNV build 2052 , but I have to put in DGIndexNV.ini to force GP104 appears 1b81 [GeForce GTX 1070]

I don't understand that. What are you putting in the INI file? I don't know of anything in there that could affect this, other than the GPU selection, but that should be correct by default because you have only one GPU.

Anyway, I made a slipstream to do the comparison correctly. Please download it and try it.

i try to force of 1070 gtx in .ini

[admin]

What is the file name of the file you changed and what did you change in there? In any case, your screenshot is not showing the GPU core correctly, which is what I am talking about.

With CUDA_Device=255 or CUDA_Device=0 in DGIndexNV.ini and and the slipstreamed DGIndexNV.exe, everything should display correctly. Please let me know if it doesn't.

[pacoor]

What is the file name of the file you changed and what did you change in there? In any case, your screenshot is not showing the GPU core correctly, which is what I am talking about.

With CUDA_Device=255 or CUDA_Device=0 in DGIndexNV.ini and and the slipstreamed gpus.txt file, everthing should display correctly. Please let me know if it doesn't.

no understand, try to dowload dgdecnv2052.zip other?

[admin]

Get new build 2052 again from my site and use the new DGIndexNV.exe in there. Then show screenshot of List GPUs dialog. Thanks!

[pacoor]

Get new build 2052 again from my site and use the new DGIndexNV.exe in there. Then show screenshot of List GPUs dialog. Thanks!

[admin]

Working great now! Thanks for your testing, Sir.

[pacoor]

Working great now! Thanks for your testing, Sir.

thanks!

Could you help me to get the most GPU process, instead of the CPU?

[admin]

I already explained it. Didn't you read it?

Use DXVA mode as described in the manual. You'll lose some FPS, however. You'll also be limited to one instance of DGSource() in your script.

Set these in the DGIndexNV.ini file:

UseD3D=1

Decode_Modes=2,2,2,2

[pacoor]

I already explained it. Didn't you read it?

Use DXVA mode as described in the manual. You'll lose some FPS, however. You'll also be limited to one instance of DGSource() in your script.

Set these in the DGIndexNV.ini file:

UseD3D=1

Decode_Modes=2,2,2,2

I have no idea to handle your program, so.
ok I will try their config in .ini

[admin]

I have no idea to handle your program, so.

It's all explained in my brilliantly written users manuals.

I suppose the English is an obstacle for you. OK, no problem, just post here and we'll all help you. You seem like a great guy.

Thank you for your interest in my tools.

[pacoor]

I have no idea to handle your program, so.

It's all explained in my brilliantly written users manuals.

I suppose the English is an obstacle for you. OK, no problem, just post here and we'll all help you. You seem like a great guy.

Thank you for your interest in my tools.

yes english ......

I do not understand, or even nvidia drivers are not very compatible with 1070 gtx, or not explain.

[admin]

Forgot to tell you that DXVA mode works only on Win 7 and Win 8/8.1. Sorry but it is an nVidia limitation. 9% CPU for 2160p is pretty damn good, IMHO, so don't be sad.

[pacoor]

DXVA mode works only on Win 7 and Win 8. Sorry but it is an nVidia limitation.

try to reinstall the previous version

[admin]

It works fine on 7/8/8.1. On my main development machine I use 8.1. I hope to be able to ditch my GWX defenses any day now.

Before going to all the trouble of reverting your OS, however, I would run AVSMeter tests to see if you can live with CUVID/CUDA mode on your system. Tell us about your use case; maybe we can have other suggestions for you. Maybe some multi-threading, etc.

Good luck, my friend.

[pacoor]

It works fine on 7/8/8.1. On my main development machine I use 8.1. I hope to be able to ditch my GWX defenses any day now.

Before going to all the trouble of reverting your OS, however, I would run AVSMeter tests to see if you can live with CUVID/CUDA mode on your system. Tell us about your use case; maybe we can have other suggestions for you. Maybe some multi-threading, etc.

Good luck, my friend.

I tried to download AVS Meter, but I have problems with x64 avisynth

thanks, tomorrow I will try other things.

[pacoor]

yes i new in new to Avisynth processing
try to reinstall Avisynth tomorrow. Thanks Donald!

[pacoor]

i have installed the last package Anon17 r2085 binaries https://www.dropbox.com/s/t6yg2cc900tkc ... 85.7z?dl=0 and avsmeter 2.3.4 from http://forum.doom9.org/showthread.php?t=165528

[pacoor]

hello|



I try to use mediacoder and see the gpu is working very well, accelerating the process almost 16x








with bdrebuilder and DGIndexNV 2052 only 6% load gpu

[aceado]

the first thing you need to understand is the first picture is showing that its using the NVENC which is the builtin hardware encoder so yes it will be faster....
best way to find out if it is using gpu to the fullist is to load up a video directly into dgnv itself goto video > disable display then click play if your fps is around ~700 fps then its all fine since this is showing you how fast and able your 1070 is at decoding however in the pictures above its showing your only encoding at 32 fps which is why you would be seeing ~6% video useage

[admin]

aceado is correct. DGDecNV accelerates only the decoding, not the encoding. Try his suggestion while monitoring GPU usage.

What I typically do is run my Avisynth script with AVSMeter. Either way will give you useful statistics.

Perhaps BDRebuilder can also be configured to use NVENC, I don't know. If it uses only a SW encoder then it is expected that the GPU utilization will be much lower, because most of the real time is spent doing software things, not GPU things.

It seems to me that you should focus on maximizing the transcoding frame rate (which determines how long the job will take) rather than the GPU usage. They may be related but it is not a straightforward relation.

[aceado]

thank you admin! both a avisynth script and using avsmeter and this this method should yeilf the same result depeg how the use wants to go about it... for me this way is more to the point (quicker lol) the latter would be the true decoding speed before any filters are added

here's a quick small video showing you that it does fully work just fine. the first video is h264 full hd (1920x1080) and the second video is mpeg also full hd (1920x1080) you'll see if i play it i also see about ~6% video useage

the decoding speeds in the video shows its able to decode h264 at ~756fps and mpeg2 ~854fps in this example

http://www29.zippyshare.com/v/ykcaJml6/file.html