Jump to content
XPEnology Community

DVA3221 loader development thread


Recommended Posts

Good afternoon! I have a PC with an I5-4690 processor, an Nvidia 1070 video card, 16 Gb RAM. Is it possible to use AI functions based on Xpenology DVA3221 with this graphics card? And another question - if AI functions are supported by this video card, is it possible to install Proxmox on the basis of this PC and create 2 virtual machines in it: 1) DSM based on DVA3221 with a forwarded Nvidia GTX 1070 video card and 2) Windows 10 (as an option, if the resources of video card 1070 cannot be used in this virtual machine (because it is forwarded to DSM - forward the internal video card of the Intel HD4600 processor to it). Thank you for your answers!

Link to comment
Share on other sites

Hello,

 

We can't know for you, just try and let us know.

DVA3219 may be an alternative for your GPU if DVA3221 does not work.

 

I gave my findings regarding Proxmox GPU passthrough, feel free to test and experiment.

Edited by Orphée
Link to comment
Share on other sites

14 hours ago, sandro_66 said:

Good afternoon! I have a PC with an I5-4690 processor, an Nvidia 1070 video card, 16 Gb RAM. Is it possible to use AI functions based on Xpenology DVA3221 with this graphics card? And another question - if AI functions are supported by this video card, is it possible to install Proxmox on the basis of this PC and create 2 virtual machines in it: 1) DSM based on DVA3221 with a forwarded Nvidia GTX 1070 video card and 2) Windows 10 (as an option, if the resources of video card 1070 cannot be used in this virtual machine (because it is forwarded to DSM - forward the internal video card of the Intel HD4600 processor to it). Thank you for your answers!

I can tell you from experience, igpu passthrough in proxmox is tricky at best. Discrete gpu passthrough is much easier (ie: dedicated video card).  iGPU passthrough is not impossible but very finicky. At least that was my personal experience, you may have better luck. I personally tried with an i7 4790 (if I remember right) and simply could not make windows work with the igpu. It would see it, and BSOD trying to install drivers for it. Again, my experience, not necessarily your experience.

 

As to the 1070 card, someone earlier in this thread said there was a way to install an earlier version of SS, then un-install and re-install newer version to get drivers for older/other nvidia cards to work, but I never tried so - you would have to look back through the pages and see if that works, or as @Orphée said, maybe 3219 model may be better solution. Be sure to keep us posted what you try and what works and what doesn't  -  that's how we all learn. :)

Link to comment
Share on other sites

Try fix for Pascal gpus from post #54 use SS version SurveillanceStation-x86_64-9.0.1-6873_DVA_3219.spk because newer versions are encrypted.
Now there is a loader for DVA3219, from my point of view; it would be wiser to use it instead of a Frankenstein SS in DVA3221.
If it works out of the box with 3219 and must be tweaked with old 3219 SS release on DVA3221, i would personally not bother with modded release.
Link to comment
Share on other sites

Good afternoon! I tried to install on bare metal on a laptop with arpl 0.4a1 bootloader (DVA3221). Unfortunately, the Geforce GTX940M video card is not visible (in the system information, the PCI-E line is empty). Accordingly, AI and face recognition do not work.

Edited by sandro_66
Link to comment
Share on other sites

8 hours ago, dimakv2014 said:

Try fix for Pascal gpus from post #54 use SS version SurveillanceStation-x86_64-9.0.1-6873_DVA_3219.spk because newer versions are encrypted.

So i need downgrade all system to 7.0.1 ? because any pack that i want instal give error with wrong file format :(

 

With fix i have:

https://pastebin.pl/view/d200bf13

Edited by Bullseye
Link to comment
Share on other sites

10 часов назад, Orphée сказал:

Now there is a loader for DVA3219, from my point of view; it would be wiser to use it instead of a Frankenstein SS in DVA3221.
If it works out of the box with 3219 and must be tweaked with old 3219 SS release on DVA3221, i would personally not bother with modded release.

Dva3219 has less tasks and tell me please which loader still has dva3219?

Edited by dimakv2014
Link to comment
Share on other sites

19 минут назад, Orphée сказал:

It probably has tasks matching GPU capability...

 

 

Well, sandro_66 found a trick which he shared in dva1622 topic that its possible to change tasks limit and create even 5 or more face recognition tasks on dva1622 (1 is supported), so I guess its not gpu capable, more like software locked, that explains why Synology used older 440.44 drivers in Nvidia Runtime Library package, so yes you are right, using dva3219 for Pascal is not bad idea at all. At least to see that Pascal gpu works )

Edited by dimakv2014
Link to comment
Share on other sites

The trick can be usefull, yes, but I think "IF" DVA3219 works fine with pascal GPUs out of the box, it will be very very easier for the guys using these GPU to maintain/update SS package & Nvidia Runtime Library matching there architecture without needs to patch/mod package after every update from GUI.

Link to comment
Share on other sites

Только что, Orphée сказал:

The trick can be usefull, yes, but I think "IF" DVA3219 works fine with pascal GPUs out of the box, it will be very very easier for the guys using these GPU to maintain/update SS package & Nvidia Runtime Library matching there architecture without needs to patch/mod package after every update from GUI.

Agree

Link to comment
Share on other sites

4 minutes ago, Orphée said:

The trick can be usefull, yes, but I think "IF" DVA3219 works fine with pascal GPUs out of the box, it will be very very easier for the guys using these GPU to maintain/update SS package & Nvidia Runtime Library matching there architecture without needs to patch/mod package after every update from GUI.

But DVA3219 have only 4 camera liceces yes?

Link to comment
Share on other sites

Good afternoon!  I gave up on the idea of bootloaders DVA3221 and DVA3219 and using arpl 0.4a1, for the sake of experiment, put the DVA1622 assembly on my HP Pavilion x360 14-ba103ur ultrabook with an I5-8250U processor (integrated UHD620 graphics, as well as GeForce GTX 940M).  I bypassed the limitation of 1 task (face recognition) or 2 tasks (Deep AI Analitics) using the "rules of action" method I proposed earlier (see the topic about DVA1622), dimakv2014 mentioned this in a previous post.  I decided to load the ultrabook to the fullest - 10 AI tasks and 5 face recognition tasks at the same time.  The load on the processor was only 50-60%.  At the same time, the graphics processor of the GeForce GTX940M ultrabook is not used at all (since the DVA1622 uses only the central processor and the intel graphics integrated into it)!!!  Miracles!!!  Now my J4125 PC will be used as a second NAS server for storing files, photos, etc., and an ultrabook for video recording with AI functions.

 

But you can, of course, try to use the DVA3219 bootloader in order for the ultrabook to use the graphics processor, but I'm afraid in this case cards of the 10th series (1050, 1060, 1070, 1080) and not 940M will be supported. Although on github, I asked the developer of the arpl loader in the next release to add the ability to build for DVA3219.

  • Like 2
Link to comment
Share on other sites

12 часа назад, sandro_66 сказал:

Good afternoon!  I gave up on the idea of bootloaders DVA3221 and DVA3219 and using arpl 0.4a1, for the sake of experiment, put the DVA1622 assembly on my HP Pavilion x360 14-ba103ur ultrabook with an I5-8250U processor (integrated UHD620 graphics, as well as GeForce GTX 940M).  I bypassed the limitation of 1 task (face recognition) or 2 tasks (Deep AI Analitics) using the "rules of action" method I proposed earlier (see the topic about DVA1622), dimakv2014 mentioned this in a previous post.  I decided to load the ultrabook to the fullest - 10 AI tasks and 5 face recognition tasks at the same time.  The load on the processor was only 50-60%.  At the same time, the graphics processor of the GeForce GTX940M ultrabook is not used at all (since the DVA1622 uses only the central processor and the intel graphics integrated into it)!!!  Miracles!!!  Now my J4125 PC will be used as a second NAS server for storing files, photos, etc., and an ultrabook for video recording with AI functions.

 

But you can, of course, try to use the DVA3219 bootloader in order for the ultrabook to use the graphics processor, but I'm afraid in this case cards of the 10th series (1050, 1060, 1070, 1080) and not 940M will be supported. Although on github, I asked the developer of the arpl loader in the next release to add the ability to build for DVA3219.

Dva3219 would be nice to try, maybe pocopico will add it soon ) since Quadro T400 is too new and Nvidia 440.44 drivers outdated and dont include this Turing gpu, maybe Quadro P400 and other pascal gpus will work with it by using dva3219, since so far it looks like Tinycore Redpill by pocopico works very stable compared to other loaders. 

Btw, I was thinking on upgrading Nvidia drivers to new one, modifying Nvidia Library .spk fails to install because of signature problems I think. So there are two options left:

1. Find actual drivers in dsm via ssh and replace it

2. Find the file of a driver with list of gpus and add unsupported gpu info manually 

Well at least until synology will update it, there must be some alternative way for sure. For two years no updates yet.

 

 

Edited by dimakv2014
  • Like 1
Link to comment
Share on other sites

I have a couple questions about DVA3221. Does ram play a part in how well the AI work or function of the Nas/nvr in general? How much ram are you guys using? What's acceptable? 8gb? 4gb? 32gb?

 

What cameras are acceptable? Does it only work with ip cameras? Is there any certain specific for a camera to look for?

I've got a 5mp system of HD over analog cctv cameras now, the cameras are BnC, so I'm sure there's no way to use these 8 cameras? Not even sure how to interface them to a pc but even if that were possible I'm sure it wouldn't work in a Synology build, but maybe something that might convert them into something addressable on the network? Not sure.

 

Well that's it for now. Thanks to anyone for replying. 

Link to comment
Share on other sites

18 hours ago, sandro_66 said:

Good afternoon!  I gave up on the idea of bootloaders DVA3221 and DVA3219 and using arpl 0.4a1, for the sake of experiment, put the DVA1622 assembly on my HP Pavilion x360 14-ba103ur ultrabook with an I5-8250U processor (integrated UHD620 graphics, as well as GeForce GTX 940M).  I bypassed the limitation of 1 task (face recognition) or 2 tasks (Deep AI Analitics) using the "rules of action" method I proposed earlier (see the topic about DVA1622), dimakv2014 mentioned this in a previous post.  I decided to load the ultrabook to the fullest - 10 AI tasks and 5 face recognition tasks at the same time.  The load on the processor was only 50-60%.  At the same time, the graphics processor of the GeForce GTX940M ultrabook is not used at all (since the DVA1622 uses only the central processor and the intel graphics integrated into it)!!!  Miracles!!!  Now my J4125 PC will be used as a second NAS server for storing files, photos, etc., and an ultrabook for video recording with AI functions.

 

But you can, of course, try to use the DVA3219 bootloader in order for the ultrabook to use the graphics processor, but I'm afraid in this case cards of the 10th series (1050, 1060, 1070, 1080) and not 940M will be supported. Although on github, I asked the developer of the arpl loader in the next release to add the ability to build for DVA3219.

totally, dva1622 was a miracle, many AI tasks with iGPU, your method makes it work like a charm

Link to comment
Share on other sites

# free -m
              total        used        free      shared  buff/cache   available
Mem:          32022        3795       13453         185       14773       27597
Swap:         21265           0       21265

 

# nvidia-smi 
Fri Aug 12 10:24:48 2022       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.44       Driver Version: 440.44       CUDA Version: 10.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 1650    On   | 00000000:01:00.0 Off |                  N/A |
| 34%   49C    P0    23W /  75W |   1947MiB /  3903MiB |     24%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0     15181      C   ...anceStation/target/synodva/bin/synodvad  1002MiB |
|    0     15195      C   ...ceStation/target/synoface/bin/synofaced   934MiB |
+-----------------------------------------------------------------------------+

 

g1wF5qx.png

 

r3tkJcc.png

 

JtsG9oQ.png

 

I have one FHD camera running only. with motion detection, and one deep analytics task, and one face detection task.

I don't know if adding more camera would increase GPU and Memory usage or if Surveillance Station cache it.

Edited by Orphée
  • Like 1
Link to comment
Share on other sites

4 часа назад, phone guy сказал:

I have a couple questions about DVA3221. Does ram play a part in how well the AI work or function of the Nas/nvr in general? How much ram are you guys using? What's acceptable? 8gb? 4gb? 32gb?

 

What cameras are acceptable? Does it only work with ip cameras? Is there any certain specific for a camera to look for?

I've got a 5mp system of HD over analog cctv cameras now, the cameras are BnC, so I'm sure there's no way to use these 8 cameras? Not even sure how to interface them to a pc but even if that were possible I'm sure it wouldn't work in a Synology build, but maybe something that might convert them into something addressable on the network? Not sure.

 

Well that's it for now. Thanks to anyone for replying. 

Up to 32 cameras and up to 12 tasks (face rec = 2 tasks) its enough 32gb ram, which is maximum supported by dva3221, retail comes with 8 licenses and 8gb of ram, so I guess its 1gb per cam or less, since you can use it for other things besides surveillance. So my verdict you can easely get away with 16gb or even 8gb of ram 

Link to comment
Share on other sites

3 часа назад, Orphée сказал:
# free -m
              total        used        free      shared  buff/cache   available
Mem:          32022        3795       13453         185       14773       27597
Swap:         21265           0       21265

 

# nvidia-smi 
Fri Aug 12 10:24:48 2022       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.44       Driver Version: 440.44       CUDA Version: 10.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 1650    On   | 00000000:01:00.0 Off |                  N/A |
| 34%   49C    P0    23W /  75W |   1947MiB /  3903MiB |     24%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0     15181      C   ...anceStation/target/synodva/bin/synodvad  1002MiB |
|    0     15195      C   ...ceStation/target/synoface/bin/synofaced   934MiB |
+-----------------------------------------------------------------------------+

 

g1wF5qx.png

 

r3tkJcc.png

 

JtsG9oQ.png

 

I have one FHD camera running only. with motion detection, and one deep analytics task, and one face detection task.

I don't know if adding more camera would increase GPU and Memory usage or if Surveillance Station cache it.

It should increase cpu, gpu, gpu memory and system memory usage upon increasing number of cams. You can check it out by adding .mp4 files instead of cams  

Edited by dimakv2014
Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Restore formatting

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...