logo

Live Production Software Forums


Welcome Guest! To enable all features please Login or Register.

Notification

Icon
Error

Options
Go to last post Go to first unread
daniel514  
#1 Posted : Friday, August 14, 2020 9:20:33 AM(UTC)
daniel514

Rank: Advanced Member

Groups: Registered
Joined: 2/26/2014(UTC)
Posts: 81
Canada
Location: Canada

Thanks: 18 times
Was thanked: 15 time(s) in 15 post(s)
Hi

I got two brand new workstations dedicated to vMix:

i7-9800x
ASUS WS X299 SAGE
32 Gb Ram
1TB NVME
RTX 2070 8 GB
Some mix of BMD DL 8K and duo... lot of SDI inputs!

So i an quite happy with my builds...

But stumbled on some questionings about the mysterious GPU Mem...

https://www.vmix.com/knowledgebase/article.aspx/181/what-does-the-gpu-mem-indicator-mean-in-the-vmix-status-bar

So i felt i should benchmark one of my old 3770K GTX 1060 6GB.... my new PCs should be so more powerful!...

....YES BUT :

The CPU is big difference, but the GPU MEM Available is the same! about 50% with the scene you can DL and test here:
https://drive.google.com...08AKuK6hzXkB?usp=sharing

I am questioning the situation!

  • Is it the GPU L2 Cache memory available we see in vMix?

  • Does vMix adapt to different cards or it is a fixed amount to be fine with most cards?


If i refer to this:
https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units

GTX 1060 have 1.5 MB L2 Cache
RTX 2070 have 4 MB L2 Cache

Please try my little benchmark project and let us know your results!

Thanks!

Daniel


hellokitty  
#2 Posted : Friday, August 28, 2020 10:39:59 AM(UTC)
hellokitty

Rank: Advanced Member

Groups: Registered
Joined: 8/29/2015(UTC)
Posts: 135
Location: Seattle Wa

Thanks: 16 times
Was thanked: 14 time(s) in 13 post(s)
wait for the 3090 supposedly 24 gig? i have a 1080ti, i could maybe test something?
daniel514  
#3 Posted : Friday, August 28, 2020 10:53:44 AM(UTC)
daniel514

Rank: Advanced Member

Groups: Registered
Joined: 2/26/2014(UTC)
Posts: 81
Canada
Location: Canada

Thanks: 18 times
Was thanked: 15 time(s) in 15 post(s)
Originally Posted by: hellokitty Go to Quoted Post
wait for the 3090 supposedly 24 gig? i have a 1080ti, i could maybe test something?

Try my little test project ant post the result here...
https://drive.google.com/drive/folders/1DOkrUtMnHJzOhxdtmfQs08AKuK6hzXkB
DWAM  
#4 Posted : Friday, August 28, 2020 4:24:38 PM(UTC)
DWAM

Rank: Advanced Member

Groups: Registered
Joined: 3/20/2014(UTC)
Posts: 2,721
Man
France
Location: Bordeaux, France

Thanks: 243 times
Was thanked: 794 time(s) in 589 post(s)
Perhaps a tool like GPU-Z is appropriate
https://www.techpowerup....nload/techpowerup-gpu-z/

As I understand it, there is a dedicated MEM and a shared MEM. vMix reports the former (at least for my GTX)
mtone  
#5 Posted : Friday, August 28, 2020 5:05:56 PM(UTC)
mtone

Rank: Advanced Member

Groups: Registered
Joined: 8/9/2020(UTC)
Posts: 156

Thanks: 3 times
Was thanked: 20 time(s) in 19 post(s)
Originally Posted by: daniel514 Go to Quoted Post
Hi

I got two brand new workstations dedicated to vMix:

i7-9800x
ASUS WS X299 SAGE
32 Gb Ram
1TB NVME
RTX 2070 8 GB
Some mix of BMD DL 8K and duo... lot of SDI inputs!

So i an quite happy with my builds...





what are your thoughts about the 2070 and the blackmagic decklink 8k ? does the decklink work reliably or any problems with it ? have you run four 4k streams into vMix with DL 8k ?

do you know if the 2070 can encode a 1080 stream and a 4k recording at same time without any problems ?
daniel514  
#6 Posted : Saturday, August 29, 2020 3:01:45 AM(UTC)
daniel514

Rank: Advanced Member

Groups: Registered
Joined: 2/26/2014(UTC)
Posts: 81
Canada
Location: Canada

Thanks: 18 times
Was thanked: 15 time(s) in 15 post(s)
Originally Posted by: mtone Go to Quoted Post
Originally Posted by: daniel514 Go to Quoted Post
Hi

I got two brand new workstations dedicated to vMix:

i7-9800x
ASUS WS X299 SAGE
32 Gb Ram
1TB NVME
RTX 2070 8 GB
Some mix of BMD DL 8K and duo... lot of SDI inputs!

So i an quite happy with my builds...





what are your thoughts about the 2070 and the blackmagic decklink 8k ? does the decklink work reliably or any problems with it ? have you run four 4k streams into vMix with DL 8k ?

do you know if the 2070 can encode a 1080 stream and a 4k recording at same time without any problems ?


To me 2070 is quite enough for my use!
I did run 8x 4K 30p over two DL 8k with no problems testing it for hours! (i use a DL video hub 40 to share a Z Cam)
To me DL were reliable 100% of use over years with many models i own ...:
1x old quad
4x DUO 2
2 mini 1080
1 mini 4K
Plus recent 2x DL 8k

For the HW NVENC encoding, i do record 2X 4K 30p pushing the card encoder to 50% zero drop frames!

I you want to scratch your head about NVENC... look at: https://www.elpamsoft.com/?p=Plex-Hardware-Transcoding
For what i understand 2070 and 2080 have the same NVENC Power!

daniel514  
#7 Posted : Saturday, August 29, 2020 5:41:05 AM(UTC)
daniel514

Rank: Advanced Member

Groups: Registered
Joined: 2/26/2014(UTC)
Posts: 81
Canada
Location: Canada

Thanks: 18 times
Was thanked: 15 time(s) in 15 post(s)
Originally Posted by: DWAM Go to Quoted Post
Perhaps a tool like GPU-Z is appropriate
https://www.techpowerup....nload/techpowerup-gpu-z/

As I understand it, there is a dedicated MEM and a shared MEM. vMix reports the former (at least for my GTX)



I don't think it as much to do to any readings availables from Task manager or gpu-z, it seems to be an amount of directly accessible cache memory available on the card, from what i see it is the same % on the test i do on a 1060 6GB, 2070 8GB or a EC2 g4dn.4xlarge with a NVIDIA T4 16GB!
Maybe vMix dev decided of a minimal amount to accommodates all situations and will tweak it in some update and all will change!

Her is where i doubt if it worth any big investment into a RTX 2080 or RTX 3090 or any Quadro , if at the end vMix is not using the added 3D power or RAM...

i know Quadro "unlimited" number of NVENC... it may work for 8x 1080p30 streams but it wont go so far in 4K streams!

that is my position, buy RTX 2070, solid affordable card on a good X299 platform for multi SDI cards and if you want to go crazy on NDI invest in CPU and network!

I am looking to understand... i am not certain of anything!

Daniel
mtone  
#8 Posted : Saturday, August 29, 2020 2:19:24 PM(UTC)
mtone

Rank: Advanced Member

Groups: Registered
Joined: 8/9/2020(UTC)
Posts: 156

Thanks: 3 times
Was thanked: 20 time(s) in 19 post(s)
Originally Posted by: daniel514 Go to Quoted Post

To me 2070 is quite enough for my use!
I did run 8x 4K 30p over two DL 8k with no problems testing it for hours! (i use a DL video hub 40 to share a Z Cam)
To me DL were reliable 100% of use over years with many models i own ...:
1x old quad
4x DUO 2
2 mini 1080
1 mini 4K
Plus recent 2x DL 8k

For the HW NVENC encoding, i do record 2X 4K 30p pushing the card encoder to 50% zero drop frames!

I you want to scratch your head about NVENC... look at: https://www.elpamsoft.com/?p=Plex-Hardware-Transcoding
For what i understand 2070 and 2080 have the same NVENC Power!



thanks.. i find graphics cards a bit tricky to guage without being able to run them in a machine and see what they can do relative to what you need.. from what ive seen though i think 2070 would be ok.

when you had the eight 4k cameras running into the two DL 8k cards was that on the system you listed in the first post with the 9800x and the 2070 ? if so, did it show any signs of struggling under that load ? good to hear the black magic cards have been reliable for you..
daniel514  
#9 Posted : Sunday, August 30, 2020 1:19:02 AM(UTC)
daniel514

Rank: Advanced Member

Groups: Registered
Joined: 2/26/2014(UTC)
Posts: 81
Canada
Location: Canada

Thanks: 18 times
Was thanked: 15 time(s) in 15 post(s)
Originally Posted by: mtone Go to Quoted Post
Originally Posted by: daniel514 Go to Quoted Post

To me 2070 is quite enough for my use!
I did run 8x 4K 30p over two DL 8k with no problems testing it for hours! (i use a DL video hub 40 to share a Z Cam)
To me DL were reliable 100% of use over years with many models i own ...:
1x old quad
4x DUO 2
2 mini 1080
1 mini 4K
Plus recent 2x DL 8k

For the HW NVENC encoding, i do record 2X 4K 30p pushing the card encoder to 50% zero drop frames!

I you want to scratch your head about NVENC... look at: https://www.elpamsoft.com/?p=Plex-Hardware-Transcoding
For what i understand 2070 and 2080 have the same NVENC Power!



thanks.. i find graphics cards a bit tricky to guage without being able to run them in a machine and see what they can do relative to what you need.. from what ive seen though i think 2070 would be ok.

when you had the eight 4k cameras running into the two DL 8k cards was that on the system you listed in the first post with the 9800x and the 2070 ? if so, did it show any signs of struggling under that load ? good to hear the black magic cards have been reliable for you..

Yes :
9800x
RTX 2070
WS x299 SAGE
32 GB RAM

Let's push it a bit...:
8x 1080p60
8x 4Kp30
3 x multicorder 4K30p... NVENC to 70% (NVENC is used only for mp4 recording, without it the card run at 22% 3D)
GPU CPU stats on the screen capture!
i don't expect to push this system so much! I am just benchmarking it...

PS: i ad to put a 120mm fan blowing directly over my decklinks sandwich...
at some point i eared one card fan winning of over spinning due to eat accumulating it was over eating and dropping frames,
now i am confident it will be stable!
With multicorder:
The 4K cards all dropped 18 frames over 3Hrs... maybe caused by the input as they all suffer the same... i feel fine with it!
Screenshot (23).png (1,186kb) downloaded 25 time(s).
Without multicorder:
Screenshot (20).png (2,002kb) downloaded 21 time(s).
thanks 1 user thanked daniel514 for this useful post.
mtone on 8/31/2020(UTC)
mtone  
#10 Posted : Sunday, August 30, 2020 11:38:11 AM(UTC)
mtone

Rank: Advanced Member

Groups: Registered
Joined: 8/9/2020(UTC)
Posts: 156

Thanks: 3 times
Was thanked: 20 time(s) in 19 post(s)
nice test.. I noticed on the windows meter it showed 70% on the GPU but no spikes which is encouraging. I also noticed 3D shows 22%, is that reflecting some titles you have used or something else ? good results, given how much you had in the project..

with the 8x SDI, did you run live inputs into each of them ? or were they blank sources ? or was it one source that was repeated across 8 inputs ? im just wondering, if there arent 8 different sources (showing different images) and if there wasnt lots of changing motion would that make it easier for the hardware ? i think still images or something without much motion can be easier to encode and might not stress the system as much ?
daniel514  
#11 Posted : Monday, August 31, 2020 12:14:13 AM(UTC)
daniel514

Rank: Advanced Member

Groups: Registered
Joined: 2/26/2014(UTC)
Posts: 81
Canada
Location: Canada

Thanks: 18 times
Was thanked: 15 time(s) in 15 post(s)
Originally Posted by: mtone Go to Quoted Post
nice test.. I noticed on the windows meter it showed 70% on the GPU but no spikes which is encouraging. I also noticed 3D shows 22%, is that reflecting some titles you have used or something else ? good results, given how much you had in the project..

with the 8x SDI, did you run live inputs into each of them ? or were they blank sources ? or was it one source that was repeated across 8 inputs ? im just wondering, if there arent 8 different sources (showing different images) and if there wasnt lots of changing motion would that make it easier for the hardware ? i think still images or something without much motion can be easier to encode and might not stress the system as much ?


"All" i ad is 8x 1080p60 + 8x 4Kp30 16 total with over a DL HUB 40x40 feeding 2 cameras with some image feeds.
I see it as a PCIE test... all equal to 48 1080p30 inputs!
No titles... nothing, just testing the video inputs.

the 3 NVENC recordings are not influenced by any camera content change! Stable 70%

You bring me back to the root of this tread... GPU Mem
Recording/streaming NVENC 1080p30
All my SDI inputs GPU Mem: 64% Render time: 12ms
Plus 11 classic titles 1080: GPU Mem: 99% Render time: 17ms

Screenshot (25).png (1,300kb) downloaded 14 time(s).





mtone  
#12 Posted : Monday, August 31, 2020 3:37:28 AM(UTC)
mtone

Rank: Advanced Member

Groups: Registered
Joined: 8/9/2020(UTC)
Posts: 156

Thanks: 3 times
Was thanked: 20 time(s) in 19 post(s)
cool thanks.. its good info to see the tests you have done..
thanks 1 user thanked mtone for this useful post.
daniel514 on 8/31/2020(UTC)
NiBTour  
#13 Posted : Wednesday, September 9, 2020 1:23:47 AM(UTC)
NiBTour

Rank: Advanced Member

Groups: Registered
Joined: 3/28/2016(UTC)
Posts: 159
Man
United States
Location: SACRAMENTO

Thanks: 5 times
Was thanked: 19 time(s) in 18 post(s)
Originally Posted by: mtone Go to Quoted Post
cool thanks.. its good info to see the tests you have done..


Also be sure to check the check-box that says "High Input Performance Mode" in vMix settings to utilize more than 3GB of VRAM on the GPU. Without it your project utilizes 100% in vMix GPUMem and render time was about 22-50ms and with it checked, the GPU Mem dropped to 48% with a render time of about 0-2ms.

Users browsing this topic
Guest (5)
Forum Jump  
You cannot post new topics in this forum.
You cannot reply to topics in this forum.
You cannot delete your posts in this forum.
You cannot edit your posts in this forum.
You cannot create polls in this forum.
You cannot vote in polls in this forum.