Video RAM usage on 4k

[tormento]

I currently own a Nvidia 1060 3GB and I am bond to 1080p when using multithreaded AVS+. I can comfortably run 6 threads while denoising with OpenCL and decoding with DGIndexNV.

Can some of you tell me the video RAM requirements to run 4/6/8 or more number of threads using SMDegrain with KNLMeansCL on 4K HDR material?

I am planning to invest some money on newer video card and would like to know how much memory to consider.

:thanks:

[kedautinh12]

I currently own a Nvidia 1060 3GB and I am bond to 1080p when using multithreaded AVS+. I can comfortably run 6 threads while denoising with OpenCL and decoding with DGIndexNV.

Can some of you tell me the video RAM requirements to run 4/6/8 or more number of threads using SMDegrain with KNLMeansCL on 4K HDR material?

I am planning to invest some money on newer video card and would like to know how much memory to consider.

:thanks:

Just buy RTX2080TI, you won't worry about it

[DJATOM]

RTX 3080 is much better choice now.

[lansing]

I don't understand why are people so obsess with SMDegrain, it was known to wipe out details in moving objects.

[tormento]

Thanks for your replies but nobody answered what I asked :p

[Groucho2004]

I currently own a Nvidia 1060 3GB and I am bond to 1080p when using multithreaded AVS+. I can comfortably run 6 threads while denoising with OpenCL and decoding with DGIndexNV.

Can some of you tell me the video RAM requirements to run 4/6/8 or more number of threads using SMDegrain with KNLMeansCL on 4K HDR material?

I am planning to invest some money on newer video card and would like to know how much memory to consider.

:thanks:Well, you can easily test this yourself with AVSMeter:

Script:
LoadPlugin("E:\Apps\VideoTools\DGDecNV\DGDecodeNV.dll")
DGSource("UHD.dgi")
smdegrain(prefilter = 4)
#prefetch(4)


My card is a RTX2060 with 6G, the clip is 8 bit h.264 (3840 x 2160). As you can see, prefetch(8) is pretty much the limit for a 6G card @ 8 bit, not sure about HDR.

No prefetch:
Frames processed: 200 (100 - 299)
FPS (min | max | average): 0.206 | 1.572 | 1.505
Process memory usage (max): 923 MiB
Thread count: 17
CPU usage (average): 25.0%

GPU usage (average): 9%
VPU usage (average): 3%
GPU memory usage: 1198 MiB


Prefetch(4):
Frames processed: 200 (100 - 299)
FPS (min | max | average): 0.185 | 201420 | 5.201
Process memory usage (max): 1871 MiB
Thread count: 49
CPU usage (average): 97.3%

GPU usage (average): 32%
VPU usage (average): 15%
GPU memory usage: 3394 MiB


Prefetch(6):
Frames processed: 200 (100 - 299)
FPS (min | max | average): 0.108 | 189572 | 4.782
Process memory usage (max): 2601 MiB
Thread count: 65
CPU usage (average): 94.5%

GPU usage (average): 31%
VPU usage (average): 16%
GPU memory usage: 4518 MiB


Prefetch(8):
Frames processed: 200 (100 - 299)
FPS (min | max | average): 0.104 | 201420 | 4.772
Process memory usage (max): 2817 MiB
Thread count: 81
CPU usage (average): 93.6%

GPU usage (average): 30%
VPU usage (average): 15%
GPU memory usage: 5641 MiB

[tormento]

Well, you can easily test this yourself with AVSMeter
Thanks!

How could I test, if my video RAM is not enough? :)

AKA

"Hey, I wanna do x, I don't have enough RAM to allocate in my card, how much video RAM do I need to do it, so I can buy an appropriate card for my needs?"

10x FranceBB :D

[gonca]

If you max out your video ram, odds are that you don't have enough

[FranceBB]

If you max out your video ram, odds are that you don't have enough

He's actually saying: "Hey, I wanna do x, I don't have enough RAM to allocate in my card, how much video RAM do I need to do it, so I can buy an appropriate card for my needs?"

He clearly can't test it.
I would happily provide the info if it wasn't for the fact that my NVIDIA Quadro P4000 (8 GB GDDR5) is actually always in use as it's not mine, it's used in our 2 main server, so... there's always something running on it...
Still, I can tell you that while you'll probably be fine with decoding HDR contents and do some filtering on it, you won't be able to do very complex operations with 16bit planar precision. I've been using it with 10bit and 12bit HDR CLog and Slog masterfiles and I can tell you that when you apply 65x65x65 matrices of linear transformation with 16bit planar precision and you do some other filtering, you can easily max out even an 8 GB card... The thing is, of course, how much filtering you wanna do and how precise it's going to be. Besides, take this with a bit of salt, 'cause my source files are generally AppleProRes XQ 4:4:4 so it will probably use less resources on common consumer H.265 HEVC 4:2:0 files...

[gonca]

Since tormento mentioned 4K and HDR I tested a short clip

script

LoadPlugin("C:/Program Files (Portable)/dgdecnv/DGDecodeNV.dll")
DGSource("I:/some clip.dgi", fieldop=0)

Report

[Runtime info]
Frames processed: 7080 (0 - 7079)
FPS (min | max | average): 41.74 | 132.9 | 110.8
Process memory usage (max): 443 MiB
Thread count: 24
CPU usage (average): 7.3%

GPU usage (average): 26%
VPU usage (average): 59%
GPU memory usage: 2180 MiB
GPU Power Consumption (average): 0.5 W

Note that system was being used for an encode when AVSMeter was used for this

[tormento]

He's actually saying: "Hey, I wanna do x, I don't have enough RAM to allocate in my card, how much video RAM do I need to do it, so I can buy an appropriate card for my needs?"
Thanks for your translation and kind reply. :p

I have edited my answer to be clearer.

[tormento]

GPU memory usage: 2180 MiB
We will probably need 10GB to use Prefetch. :eek:

[gonca]

I hope the other encode was not using video RAM, otherwise I think we will probably need 10GB to use Prefetch. :eek:

I can repeat the test when the system is not in use, probably tomorrow

EDIT
Single encode, video card is frame serving and denoising

script
import vapoursynth as vs
core = vs.get_core()
core.std.LoadPlugin("C:/Program Files (Portable)/dgdecnv/DGDecodeNV.dll")
clip = core.dgdecodenv.DGSource(r'I:\movie.dgi', fieldop=0)

#####AVSCOMPAT#####
core.avs.LoadPlugin("C:/Program Files (Portable)/dgdecnv/DGDecodeNV.dll")

#####DENOISE#####
clip = core.avs.DGDenoise(clip, strength=0.200)

clip = core.resize.Point(clip, format=vs.YUV420P10)

clip.set_output()

gpu-z log
Date GPU Core Clock [MHz] GPU Memory Clock [MHz] GPU Temperature [°C] Fan Speed (%) [%] Fan Speed (RPM) [RPM] Memory Used [MB] GPU Load [%] Memory Controller Load [%] Video Engine Load [%] Bus Interface Load [%] Power Consumption [% TDP] PerfCap Reason [] VDDC [V]
2020-09-12 08:07:30 1556.5 1251.5 56 24 880 2117 0 1 0 0 24.7 16 0.8
2020-09-12 08:07:31 1556.5 1251.5 56 24 876 2117 0 1 0 0 24.6 16 0.8
2020-09-12 08:07:32 1556.5 1251.5 56 24 867 2114 0 1 0 0 24.6 16 0.8
2020-09-12 08:07:33 1556.5 1251.5 56 23 866 2114 0 1 0 0 24.7 16 0.8



You might have to open the log as a spreadsheet
Memory usage is over 2100 MB