Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

RED + NVIDIA SDK and updated REDCINE-X now available

Stacey Spears said:
It was in the beta 4 thread. Phil mentions running dual Titans in post 3 of that thread.
Click to expand...

Thank you. However here is what we are experiencing.

CHANGES IN THIS BUILD (http://www.reduser.net/forum/showthread.php?174600-New-Beta-4-Redcine-X-with-NVIDIA-CUDA-decode)
Added: Memory improvements to perform GPU decoding. Memory requirement has been reduced to 4GB for full GPU decoding and playback.
Added: Speed improvements on GPU decoding and the IPP2 Image Pipeline.
Added: Multi-GPU support for windowed display. (Full screen support is not fully implemented.)

We enacted a dual GPU setup with nVlink and it does NOT work well as it is supposed to be. You don't get the speed, scalability or the stability that is expected on this software update.

Maybe in a future update.
 
Neil Abeynayake said:
We enacted a dual GPU setup with nVlink and it does NOT work well as it is supposed to be. You don't get the speed, scalability or the stability that is expected on this software update.
Click to expand...

Hi Neil,

With Dual Titan RTX cards w/ NVlink I am getting beyond 72fps 8K FF 5:1 at full debayer in realtime (tested via changing the timebase of the clip). Double check those GPU drivers and ensure NVlink is correctly configured.

Some interesting observations. NVlink does indeed work better than dual GPU without the bridge.

This is on an 18-core X299 system, but at this point it's more about what the footage is being read from as that's about 780MB/s.
 
Thank you Phil. I'll try a clean install of nVidia drivers and give it a try. We are waiting for more GPUs to arrive within the next few days. One more question about the nvlink. Do you have the nvlink that is specially made for the Titan and Quadro cards or the "regular" bridge? Is there even a difference between the two (other than the price)?
 
Neil Abeynayake said:
Thank you Phil. I'll try a clean install of nVidia drivers and give it a try. We are waiting for more GPUs to arrive within the next few days. One more question about the nvlink. Do you have the nvlink that is specially made for the Titan and Quadro cards or the "regular" bridge? Is there even a difference between the two (other than the price)?
Click to expand...

I think it's a standard connection, mine is the NVlink for the Titans. SLI will perform differently on older GPUs as that's not doing the same thing. The NVlink is a hardware bridge and not just a cable.
 
Neil Abeynayake said:
Thank you Phil. I'll try a clean install of nVidia drivers and give it a try. We are waiting for more GPUs to arrive within the next few days. One more question about the nvlink. Do you have the nvlink that is specially made for the Titan and Quadro cards or the "regular" bridge? Is there even a difference between the two (other than the price)?
Click to expand...

There are differences between the dual link and single link NVLINK connectors. I think Puget did an article on it. So yes it is recommended to use the correct one.
 
Anyone try this with entry-level 4gb mobile GPUs (specifically MX150)?

Wouldn't mind getting some 6k at 1/2 if possible.
 
How much memory usage do you guys have for 8K video full debayer?
We are planning on upgrading our 1080 Ti cards, but I'm worried that 2080 Ti does not have enough VRAM. In Redcine X it should be fine, but when the SDK is implemented in Davinci I will be more worried since Davinci by itself uses a lot of VRAM.

Cheers,
Andreas
 
1080 ti and the 2080 ti both have the same amount of VRAM 11 GB (2080 ti has more features and is faster).
Rumors are that NVidia is going to refresh their RTX GPU's with 16 Gbit/s GDDR6 which would mean an increase of ~ 15% in memory bandwidth equals 15% more performance in programs like resolve.
 
AndreasOberg said:
How much memory usage do you guys have for 8K video full debayer?
We are planning on upgrading our 1080 Ti cards, but I'm worried that 2080 Ti does not have enough VRAM. In Redcine X it should be fine, but when the SDK is implemented in Davinci I will be more worried since Davinci by itself uses a lot of VRAM.

Cheers,
Andreas
Click to expand...

My choice would be RTX 2080 Ti. If you can afford, buy 2 "RTX 2080 Ti AMP Extreme Core" made by ZOTAC. They are overclocked with a boost clock of 1815 MHz. You could keep it at that speed to whatever duration you need. They require a beefy power supply. A 1300 Watts is OK. For REDCine-X Pro work I'd recommend connecting them with a "nvlink" bridge and select 1st GPU only (in the REDCine-X preferences panel). When connected with the nvlink bridge, you get 22GBs of GPU memory since it combines and shares memory of both cards into one. You can use nvlink bridge made by any manufacturer. Make sure to order the right width to fit your card separation (placement on the motherboard - 3 or 4 slot wide). With this setup, you'll get 8k real-time performance with 40 ~ 50% GPU usage.

When you are using this setup with Resolve Studio, select "BOTH" cards under the preferences panel. DaVinci Resolve uses GPUs independently (according to their techsupport guys). If you have a 18 core workstation, you can get 8k real-time or near real-time depending on your system RAM capacity (quality and quantity of RAM would be a factor here).

Make sure you have a hi-performance system cooling (liquid cooling is recommended) with good heat dissipation within the case or tower. We use 200mm Noctua NF-A20 PWM. They also have 140mm, 120mm, 92mm versions at slightly higher decibel ratings. They are the best and quietest fans out there.
 
Last edited:
Neil Abeynayake said:
My choice would be RTX 2080 Ti. If you can afford, buy 2 "RTX 2080 Ti AMP Extreme Core" made by ZOTAC. They are overclocked with a boost clock of 1815 MHz. You could keep it at that speed to whatever duration you need. They require a beefy power supply. A 1300 Watts is OK. For REDCine-X Pro work I'd recommend connecting them with a "nvlink" bridge and select 1st GPU only (in the REDCine-X preferences panel). When connected with the nvlink bridge, you get 22GBs of GPU memory since it combines and shares memory of both cards into one. You can use nvlink bridge made by any manufacturer. Make sure to order the right width to fit your card separation (placement on the motherboard - 3 or 4 slot wide). With this setup, you'll get 8k real-time performance with 40 ~ 50% GPU usage.

When you are using this setup with Resolve Studio, select "BOTH" cards under the preferences panel. DaVinci Resolve uses GPUs independently (according to their techsupport guys). If you have a 18 core workstation, you can get 8k real-time or near real-time depending on your system RAM capacity (quality and quantity of RAM would be a factor here).
Click to expand...

Hiya Neil, thanks for the reply. But Davinci still do not have the new SDK built in, not even in the 16 beta. So right now they use CPU and GPU, but soon I would think they will pick the new SDK which only uses GPU, then the memory needs should be much higher because suddenly all is on the GPU.
If Davinci uses the GPUs independently this means that you still have the memory limit of 11GB per card right?

Andreas
 
If your GPUs are connected with a nvlink bridge and you've selected the 1st GPU only, there would be a total of 22 GBs. However, there is an option to select both GPUs individually. If you select both, there would be 11 GBs per GPU.

Also, when rendering you could select either "native" or nVidia as the engine. The nVidia rendering selection is still buggy if you were to keep default settings on. In this setting you need to increase the fan speed to 80% or so during the rendering session.
 
r3d clip: Monstro 8K HD 23.98fps 6:1 4769 frames
RCX: 51.47074

2x Xeon 6154
384 ECC RAM
Read Volume: Intel VROC RAID 0 Array 1
Write Volume: Intel VROC RAID 0 Array 2
2x Titan RTX. nvidia drive 430.86

RCX Encode to Proress 422 HQ @ 1080p:
Single GPU: 41.88 FPS
Dual GPU: 70.21 FPS
Dual GPU NVLINK: 65.47 FPS

RCX Encode to Proress 4444 XQ @ UHD:
Single GPU: 38.92 FPS
Dual GPU: 43.89 FPS
Dual GPU NVLINK: 44.21 FPS

RCX Encode to DPX UHD
Single GPU: 40.42 FPS
Dual GPU: 68.01 FPS
Dual GPU NVLINK: 53.9 FPS

Adobe Media Encoder to h264 1080p
Dual GPU: 42.98 FPS
Dual GPU NVLINK: 54.19 FPS


Quadro RTX series up next.
 
Just got the auto-update notification on Mac to v51.

Couple of glitches:

-Worst is that almost every adjustment causes the workspace to expand the window, so the controls go off-screen.

-Upon launching, it doesn't seem to take the image pipeline/monitor out settings into consideration and seems to default to the OS's gamma settings... So every thing is hyper gamma/saturated (doubled up). I have to switch to "Primary Only" and then back to "Full IPP2 Graded" for it to click off/engage to the monitor out settings.

-ADD does not export frames properly/as they are in the viewer (it appears to add gamma adjustments on top of the tonemap, even though viewing is set to IPP2-3G10/RWG).

-Also not convinced that IPP2 Chroma Noise is better than the old method + DEB on Dragon footage. Would be nice if there was an option to use the old method.

Still putting it through the wringer. Is there going to be a dedicated thread?
 
Mike P. said:
Just got the auto-update notification on Mac to v51.

Couple of glitches:

-ADD does not export frames properly/as they are in the viewer (it appears to add gamma adjustments on top of the tonemap, even though viewing is set to IPP2-3G10/RWG).
Click to expand...

The ADD button has always used BT.709/BT.1886. To have your own color space and transfer function settings applied, you will need to select ADD during a normal export.
 
Stacey Spears said:
The ADD button has always used BT.709/BT.1886. To have your own color space and transfer function settings applied, you will need to select ADD during a normal export.
Click to expand...

In that case, is there any possibility of getting it to function in the same way the stills/pic button functions (WYSIWYG type thing)?

Conversely, what’s the reason for that?
 
You must log in or register to reply here.
Back
Top