search

nihui / dain-ncnn-vulkan

DAIN, Depth-Aware Video Frame Interpolation implemented with ncnn library
MIT License
518 stars 45 forks source link

using 2 grafikcards of different type #22

Open reeneex opened 4 years ago

reeneex commented 4 years ago

hey,

i´m running right now a test on a nvidia 1660i with 8GB ram and i had to go down to 480 resolution. i´m just wondering if it is possible to run a multi GPU system with this. Since direct12 a multi GPU system is possible on games. could it work here too? i have here a extr sapphire nitro+ with 8GB ram for example.

ghost commented 4 years ago

Multi-GPU is possible. but saves time by dividing the given task rather than the concept of performance improvement. Use the -g 0,1 command. If there is a built-in graphic, for example, in the task manager, if the built-in graphic number is GPU 1, you must enter -g 0,2.

reeneex commented 4 years ago

that sound promissing. just to be sure, you mean the graphic from the iGPU on the CPU? I´m using Windows10 with a i7-10700 on an AsRock Z490 PG Velocita. I checked the Task-Manager, but ther is only one GPU, my GTX1660Ti visible. I cant finde a tutorial for this purpose. Maybe you can help me out.

If i need a different OS, just tell me, i have a currently unused NVMe SSD on board.

ghost commented 4 years ago

If only 1 separate GPU was installed and 2 GPUs did not appear in the task manager, iGPU was already disabled. In order to use multi-GPU, in addition to GeForce GPU, did you install sapphire nitro+ with 8GB GPU on your PC?

reeneex commented 4 years ago

Right now i have only the GTX installed. So I have to enable the iGPU? But I could Install the Nitro+ also. Then i have 3 GPU's 😁

What do you suggest is the most promissing way?

ghost commented 4 years ago

You do not need to turn on the iGPU. Using iGPU as a DAIN multi-GPU can rather increase RAM usage and create a work bottleneck. After installing your Radeon GPU use the -g 0,1 command.

However, if black frames occur in some frames during interpolation, it is due to GPU acceleration, so you may need to turn on iGPU for display output. In this case, you must enter the -g command without the built-in graphic number. The way you turn on the built-in GPU varies from motherboard to motherboard, so you need to find the setting yourself and enable it.

reeneex commented 4 years ago

ok, so i try first the Nitro+ You mentioned before the -g command. Could you explain that a little more? I´m not familiar with that.

I was for the internal GPU already very happy, because i found this view on the taskmanager. Thats a lot of V-RAM.

From the 64GB RAM the DAINApp is using only 7 oh so.

I attach a view screenshots.

iGPU RAM Task Manager

Mar2ck commented 4 years ago

Dain-App and Dain-ncnn are two different programs

reeneex commented 4 years ago

@Mar2ck

thats not really correct "ncnn implementation of DAIN, Depth-Aware Video Frame Interpolation." It is the same DAIN App, but this version is also for Windows/Linux/MacOS Executable for Intel/AMD/Nvidia GPU

Mar2ck commented 4 years ago

Theyre based on the same thing but they function differently. Dain-ncnn is a CLI program so you'd need to run it from the command line and use the -g argument