Cloud GPU use?

[jamezm79]

Hi. I have a GeForce GTX 1050i and it clearly isn’t up to the task. Performance is woeful (although Task Manager suggests only 7-8% load so maybe something else is wrong).

Is there any viable way to connect with a cloud-based GPU for a fee whilst still using my local webcam?

[jenssalu]

This worked for me:

On my laptop, I installed nginx as rtmp server & OBS to stream my webcam to Cloud machine. Created a VPN for free with Hamachi. (you have to start the rtmp server on the new VPN ip address.) In the videocaptureasync.py file, change self.cap = cv2.VideoCapture(self.src) to self.cap = cv2.VideoCapture(' your rtmp url ').

Good luck!

[E3V3A]

@jenssalu Thanks for sharing your idea, but I find it a little too brief to follow. It's not clear where you are setting up what. So can you please expand your method a little more, and perhaps we can put it in the FAQ? Also, what is the latency and frame rate you get from this?

[E3V3A]

PS. PR #74 is adding remote GPU support, but it's still in testing stage. Maybe some updates from the devs? @alievk ?

[maxxwizard]

I tried the idea written by @jenssalu. The results are very poor - 1 FPS - likely due to all the network latencies. I would suggest using hardwired (ethernet) connections and avoid WiFi to see if you'll get more luck than I did.

Here's what I did for my Windows-only attempt.

Setup I created an NVIDIA Quadro Virtual Workstation on Microsoft Azure.

I installed Hamachi on both my video capture machine and on my NVIDIA VM. I created a VPN to join the 2 computers together.

On video capture machine (instructions adapted from here):

1. Install a version of nginx that packages the RTMP module. I extracted http://nginx-win.ecsds.eu/download/nginx%201.7.11.3%20Gryphon.zip to c:\nginx.
2. Put this nginx.conf file into c:\nginx\conf.
3. Install OBS.
4. Launch OBS and set your stream target as:

• Service: Custom
• Server: rtmp://127.0.0.1/live
• Stream Key: test

5. In OBS, add your webcam as a source.
6. In a command prompt, run c:\nginx\nginx.exe to launch the server.
7. Use a browser to navigate to http://[Hamachi VPN IP address]:8080 to verify server is working.
8. In OBS, click Start Streaming button.
9. Verify stream by opening rtmp://[VPN IP address]/live/test in VLC Player as a network stream.
10. Click Stop Streaming button.

On NVIDIA VM:

1. Run through all instructions to install avatarify.
2. In videocaptureasync.py file, make change so: self.cap = cv2.VideoCapture('rtmp://[VPN IP address]/live/test')

Execution

1. Start streaming using OBS.
2. In VM, launch run_windows.bat in a command prompt.

[jenssalu]

@maxxwizard, Nice! That's almost the same setup as I did it! Thank you for completing the setup! Completely forgot it! I just run the rtmp server on my vpn ip address instead of 127.0.0.1!

My laptop (GeForce GTX 1050) got an fps of max 6 fps! On my VM (GPU: P6000), I was able to get a fps of 16. I'm quiet happy with that result, when you consider that my webcam has a fps of 30.

The last days, I was using a poor wifi connection, but still managed to have a latency of max 1.5 seconds!

[E3V3A]

@maxxwizard Great write-up, however, I am wondering why you are using a super outdated nginx? The Gryphon is >5 years old, so why are you not using the latest. Any nginx that old has a bunch of serious security vulnerabilities. I wouldn't give a server like that to my worst enemy.

Same goes for the outdated/abandoned RTMP plugin, it comes packaged with [0]. The better maintained ones seem to be [1,2] and the best one seem to be the nginx-http-flv-module in [3]. But then again,what do I know!

I'm looking to try to get this working with a more recent version nginx.

---

References:

• http://nginx.org/en/download.html
• http://nginx-win.ecsds.eu/download/
• http://nginx-rtmp.blogspot.com/
• https://obsproject.com/forum/resources/how-to-set-up-your-own-private-rtmp-server-using-nginx.50/
• https://github.com/arut/nginx-rtmp-module/wiki/Directives
• [0] https://github.com/arut/nginx-rtmp-module
• [1] https://github.com/sergey-dryabzhinsky/nginx-rtmp-module
• [2] https://github.com/Cooking-University/nginx-rtmp-module
• [3] https://github.com/winshining/nginx-http-flv-module

[E3V3A]

Ok, I now get it. From Readme nginx-win version.txt:

Commercial subscription only modules: = Streaming with nginx-rtmp-module ... rtmp, 1.7.12.1 is the last free version with rtmp, we do have a rtmp special offer for the 1.9 branch (which without rtmp you could use to tcp load balance 1.7.12.1 with rtmp)

I tried to install nginx 1.18.0, which are showing that it's supposedly compiled with --with-http_flv_module. However, including the config from their readme doesn't work. (It's not recognized.)

configure arguments:
 --with-cc=cl
 --builddir=objs.msvc8
 --with-debug
 --prefix=
 --conf-path=conf/nginx.conf
 --pid-path=logs/nginx.pid
 --http-log-path=logs/access.log
 --error-log-path=logs/error.log
 --sbin-path=nginx.exe
 --http-client-body-temp-path=temp/client_body_temp
 --http-proxy-temp-path=temp/proxy_temp
 --http-fastcgi-temp-path=temp/fastcgi_temp
 --http-scgi-temp-path=temp/scgi_temp
 --http-uwsgi-temp-path=temp/uwsgi_temp
 --with-cc-opt=-DFD_SETSIZE=1024
 --with-pcre=objs.msvc8/lib/pcre-8.44
 --with-zlib=objs.msvc8/lib/zlib-1.2.11
 --with-http_v2_module
 --with-http_realip_module
 --with-http_addition_module
 --with-http_sub_module
 --with-http_dav_module
 --with-http_stub_status_module
 --with-http_flv_module                        <--- THIS !
 --with-http_mp4_module
 --with-http_gunzip_module
 --with-http_gzip_static_module
 --with-http_auth_request_module
 --with-http_random_index_module
 --with-http_secure_link_module
 --with-http_slice_module
 --with-mail
 --with-stream
 --with-openssl=objs.msvc8/lib/openssl-1.1.1f
 --with-openssl-opt='no-asm no-tests -D_WIN32_WINNT=0x0501'
 --with-http_ssl_module
 --with-mail_ssl_module
 --with-stream_ssl_module

Any idea how to implement and configure this?

[E3V3A]

Filed an issue at the module page:

• https://github.com/winshining/nginx-http-flv-module/issues/165

[E3V3A]

Apparently the module included above with --with-http_flv_module is not the same as the one in link above...

[alievk]

Now you can run on Mac with a good speed using Google Colab

[18845045369]

Can i use clould GPU now,because i havn't a NAVID GPU.