DVA3221 loader development thread

[Orphée]

3 hours ago, efla83 said:

hello my firend

I installed DSM DVA 3221, I have a problem with facial recognition.

When I want to import a picture of a person an error message appears. this indicates :

 

"Impossible to perform this operation because the connection to the network is unstable or the system is busy. please try again later"

 

Has anyone had this problem before and know how to resolve it?

Thank you in advance and good day

You probably need a real SN/MAC DVA key pair to have AI features working.

[efla83]

@Orphée hello from France :)

 

thank you for your answer, do you know how to do this?
 

I'm new here and I enjoy the work. I have a real DS218+ at home

 

 

[Orphée]

1 hour ago, efla83 said:

@Orphée hello from France

 

thank you for your answer, do you know how to do this?
 

I'm new here and I enjoy the work. I have a real DS218+ at home

 

 

You're on your own for this.

 

I forgot to ask as it seems so mandatory for me :

 

You've of course read all the thread, and you of course have a Nvidia compatible GPU (like GTX 1650) working and detected in the DSM system, right ?

[efla83]

@Orphée yes i have nvidia card and i see in the information tab.

[Orphée]

A 1060 may not be compatible with DVA3221 AI features, you should look inside logs (SSH).

I may be wrong but DVA3219 may be metter for this GPU.

 

There is a lot commands available in the topic here to check if GPU is actually working.

Edited October 11, 2023 by Orphée

[blinderix]

Hello,

Anyone managed to get AI with MSI GeForce GTX 1650 D6 VENTUS XS OCV1, 4GB GDDR6X ?  

Thank you!

[iansc]

On 10/22/2023 at 6:22 PM, blinderix said:

Hello,

Anyone managed to get AI with MSI GeForce GTX 1650 D6 VENTUS XS OCV1, 4GB GDDR6X ?  

Thank you!

Not compatible.

[incer]

Hi everyone, i have this issue while trying to test face recognition.

SN/MAC generated.

GPU detected.

DSM 7.2-64570 Update 1

 

 

 

 

 

Can some one tell me what im doing wrong, or bad GPU?

 

Update:

2023-11-03T14:40:07+03:00 NVR kernel: [  622.500011] synofaced[21436]: segfault at 0 ip 00007f36b522b350 sp 00007f3648032e58 error 4 in libdvacore-dl.so[7f36b5200000+52000]
2023-11-03T14:40:07+03:00 NVR coredump[32168]: Process synofaced[21425](/volume1/@appstore/SurveillanceStation/synoface/bin/synofaced) dumped core on signal [11]. Core file [/volume1/@synofaced.synology_denverton_dva3221.64570.core.gz]. Cmdline [/var/packages/SurveillanceStation/target/synoface/bin/synofaced ]
 

Meaning my cardis not supported?

Edited November 3, 2023 by incer

[incer]

2 hours ago, incer said:

Hi everyone, i have this issue while trying to test face recognition.

SN/MAC generated.

GPU detected.

DSM 7.2-64570 Update 1

 

 

 

 

 

Can some one tell me what im doing wrong, or bad GPU?

 

Update:

2023-11-03T14:40:07+03:00 NVR kernel: [  622.500011] synofaced[21436]: segfault at 0 ip 00007f36b522b350 sp 00007f3648032e58 error 4 in libdvacore-dl.so[7f36b5200000+52000]
2023-11-03T14:40:07+03:00 NVR coredump[32168]: Process synofaced[21425](/volume1/@appstore/SurveillanceStation/synoface/bin/synofaced) dumped core on signal [11]. Core file [/volume1/@synofaced.synology_denverton_dva3221.64570.core.gz]. Cmdline [/var/packages/SurveillanceStation/target/synoface/bin/synofaced ]
 

Meaning my cardis not supported?

Update^

On DVA3219 all fine.

Is there any chance to get it work on 3221 by replacing gpu rom?

[FastNumber]

First time getting an Xpenology box running but having an issue with the Push Service ("Unable to load system settings, possibly because the network connection is unstable or the system is busy. Please try again later.")

 

Using https://github.com/AuxXxilium/arc loader

Everything else works including AI. (Baremetal Xeon-1541 with a NVIDIA 1660 GPU)

 

I want to get mobile push notifications but can't use it with DS Cam and my Iphone. The email notifications does work. Any ideas?

 

Edited December 9, 2023 by FastNumber

[Orphée]

I don't think this is particularly related to DVA3221 but more Synology DSM in general.

 

You probably should create a thread in the help section to request some.

Or find some ARC dedicated thread maybe as you are using this loader.

 

You could even create an issue on his github to request help as you already found his link. He may close it as in my opinion, it may not be related to the loader too.

[ZeroToHero]

On 7/11/2023 at 8:27 PM, intakes_tourers_0l said:

After encountering "no device detected" from "nvidia-smi" under ESXI passthrough, I resolved the issue by resetting the BIOS settings to default. This allowed both my GDDR5 and GDDR6 versions of GTX 1650 cards to function properly.

 

Here are the steps I took:

1. Installed a GTX GDDR6 GPU and enabled ESXI passthrough using the following parameters:

• hypervisor.cpuid.v0=FALSE
• pciPassthru0.msiEnabled=FALSE

2. Despite seeing the device in the "lspci" output, "nvidia-smi" continued to display the "no device detected" error. I suspected that the older 440.44 driver might not support the GDDR6 version. To test this, I replaced the GDDR6 card with a GDDR5 card, but encountered the same error.

3. Before giving up, I decided to reset the BIOS settings to default. After rebooting, everything started working as expected. The GPU was detected, and deep video analysis functioned normally.

4. I switched back to the GDDR6 card, and it worked perfectly as well.

 

It turned out that the driver did support my GDDR6 card, which was slightly newer than the driver version. I have yet to identify which BIOS setting caused the issue, as I had made multiple changes to reduce power consumption.

 

 

After many failed attempts to get Deep Learning Analytics and Face Recognition working on ESXI, P620. Your parameters worked for me! thank you so much!

 

System: R330

GPU: P620

ESXI: 8.0 Update 2

Loader: ARC 23.12.4a

Loader Model: DVA3219

Wouldn't see the GPU until I added: VM Options>Advanced>Edit Configuration, Add:

• hypervisor.cpuid.v0=FALSE
• pciPassthru0.msiEnabled=FALSE

 

Deep Learning Analytics started working right away. I had to reboot the host to get Face Recognition to work. 

 

 

 

[Robert Baxter]

В 21.03.2022 в 17:03, Orphée сказал:

I don't think it is very usefull to build another loader while DVA3221 works well.
It make another loader to maintain and fix with no real benefit.
DVA3221 is better than DVA3219.

I understand some of you has a gtx1050 ans would like to use it, but gtx1650 is better for HW transcoding, and for Surveillance Station too.

I may be wrong but if it is only for GPU HW transcoding, i think even a stronger Gtx GPU would work on DVA3221. Surveillance Station is probably hardcoded with gtx1650, but I bet NvidiaRuntimeLibrary can handle every Gtx GPUs...

It is just my personal point of view, maybe devs will have a different fealing.
 

The suggestion that a more powerful GTX GPU might also work on DVA3221 for hardware transcoding, with the possibility of NvidiaRuntimeLibrary handling various GTX GPUs, is an interesting thought. It demonstrates a forward-thinking approach and an openness to exploring potential alternatives.

[Orphée]

1 hour ago, Robert Baxter said:

The suggestion that a more powerful GTX GPU might also work on DVA3221 for hardware transcoding, with the possibility of NvidiaRuntimeLibrary handling various GTX GPUs, is an interesting thought. It demonstrates a forward-thinking approach and an openness to exploring potential alternatives.

Hello !

 

Welcome !

 

Thanks to quote a post that as near 2 years old...

Did you read other recent topics regarding NvidiaRuntimeLibrary ?

Don't you think my almost 2 years old message is probably irrelevant today ?

 

What's the point as your very first message ? bot/spam ?

[Robert Baxter]

 

11 часов назад, Orphée сказал:

Hello !

 

Welcome !

 

Thanks to quote a post that as near 2 years old...

Did you read other recent topics regarding NvidiaRuntimeLibrary ?

Don't you think my almost 2 years old message is probably irrelevant today ?

 

What's the point as your very first message ? bot/spam ?

Hi!

Thanks!

I think that after 2 years this will be relevant.

[Robert Baxter]

В 25.12.2023 в 11:40, Robert Baxter сказал:

 

Hi!

Thanks!

I think that after 2 years this will be relevant.

Why can’t useful advice exist after 2 years?
Are these the forum rules?
And why was my message deleted?

[Orphée]

Forum search function:

 

 

This is why your message is irrelevant after 2 years.

 

Someone already made a driver working...

 

[FastNumber]

 So far I have tested these cards to see which ones worked with the AI functions:

QUADRO P4000 8GB  > didn't work

GTX 1650 4GB  > works perfectly

GTX 1660 6GB > works perfectly

GTX 2060 12gb > didn't work

 

The 1660 can run 1 or 2 more tasks than the 1650 that's why I wanted to try a 2xxx gen card but sadly it didn't work. 

Can anyone please confirm if your 2060/2080 card works with AI functions? I want to get the best one possible so I can run more tasks at the same time.

 

 

 

 

Edited January 26 by FastNumber

[FastNumber]

A little update:

There is a hard lock on the tasks you can run with any GPU. The difference is that with any other gpu besides the 1650 you can run the tasks smoothly instead of it being run 90-100% gpu at all times. There is a way with a combination of tasks where you can put the 1650 to run at 100% but there is a sort of bug where the usage drops to less than 5% with all tasks running and while it does that the AI won't trigger at all. The 1660 and other cards stay below 100% with max amount of tasks and they don't miss a beat triggering the alerts.

 

So if you planning to use this for real I recommend you get a 1660 as they are marginally more expensive than a 1650 and produce better results.

 

For shits and giggles I tried a 2080ti too and this is what I saw:

Max number of tasks:

1650  - 90-100% gpu load

1660 -  65-75% gpu load

RTX 4000  60-70% gpu load

2080 TI - 35-40% gpu load 

 

I wish the dva3221 didn't have any hard cap of the number tasks you could run because I could use the 2080ti to its full potential even if it consumes a lot more power. 

[tm81]

Hi,

I hope someone can point me to the right direction

I have installed DVA3221 using tinycore redpill bare metal on laptop Acer Nitro 5 that comes with GTX 1650

when i go to dsm terminal and run nvidia-smi its says No devices were found

the nvidia runtime library is installed 

lsmod | grep nvidia shows:

nvidia              20259333  1 nvidia_uvm

 

i am using DSM 7.2-64570 Update 1

 

what might be my issue?

[Captainfingerbang]

so i dont have to spend 4 hours,

 

whats the latest on this?

 

• Is there a stable GPU for DVA3221?      Assuming Real/Serial Mac.

 

• Has someone found a way around the need for using a dedicated card possibly by now?

[Orphée]

6 minutes ago, Captainfingerbang said:

so i dont have to spend 4 hours,

https://www.amazon.fr/Synology-DVA3221-Desktop-NVR-4bay-KI/dp/B08L3QC9VL

 

You're welcome.

[Pantigon]

On 2/4/2024 at 3:18 AM, FastNumber said:

A little update:

There is a hard lock on the tasks you can run with any GPU. The difference is that with any other gpu besides the 1650 you can run the tasks smoothly instead of it being run 90-100% gpu at all times. There is a way with a combination of tasks where you can put the 1650 to run at 100% but there is a sort of bug where the usage drops to less than 5% with all tasks running and while it does that the AI won't trigger at all. The 1660 and other cards stay below 100% with max amount of tasks and they don't miss a beat triggering the alerts.

 

So if you planning to use this for real I recommend you get a 1660 as they are marginally more expensive than a 1650 and produce better results.

 

For shits and giggles I tried a 2080ti too and this is what I saw:

Max number of tasks:

1650  - 90-100% gpu load

1660 -  65-75% gpu load

RTX 4000  60-70% gpu load

2080 TI - 35-40% gpu load 

 

I wish the dva3221 didn't have any hard cap of the number tasks you could run because I could use the 2080ti to its full potential even if it consumes a lot more power. 

"There is a hard lock on the tasks you can run with any GPU" - How to activate it?

[dj_nsk]

В 10.12.2023 в 16:00, ZeroToHero сказал:

 

After many failed attempts to get Deep Learning Analytics and Face Recognition working on ESXI, P620. Your parameters worked for me! thank you so much!

 

System: R330

GPU: P620

ESXI: 8.0 Update 2

Loader: ARC 23.12.4a

Loader Model: DVA3219

Wouldn't see the GPU until I added: VM Options>Advanced>Edit Configuration, Add:

• hypervisor.cpuid.v0=FALSE
• pciPassthru0.msiEnabled=FALSE

 

Deep Learning Analytics started working right away. I had to reboot the host to get Face Recognition to work. 

But it didn't work that way for me

GPU: 1050Ti

ESXI: 8.0.2

Loader: ARC

Model: DVA3219

 

Can you clarify:

- have you changed the file /etc/vmware/passthru.map (ESXi)? What do you have listed for NVIDIA? 10de  ffff  bridge  false ?  link ? d3d0 ?

- is the GPU name displayed in the information center (DVA)?

- which compatibility version is specified for the VM? perhaps there were some other features when creating a virtual machine?

 

[cam0123]

On 2/3/2024 at 3:18 PM, FastNumber said:

A little update:

There is a hard lock on the tasks you can run with any GPU. The difference is that with any other gpu besides the 1650 you can run the tasks smoothly instead of it being run 90-100% gpu at all times. There is a way with a combination of tasks where you can put the 1650 to run at 100% but there is a sort of bug where the usage drops to less than 5% with all tasks running and while it does that the AI won't trigger at all. The 1660 and other cards stay below 100% with max amount of tasks and they don't miss a beat triggering the alerts.

 

So if you planning to use this for real I recommend you get a 1660 as they are marginally more expensive than a 1650 and produce better results.

 

For shits and giggles I tried a 2080ti too and this is what I saw:

Max number of tasks:

1650  - 90-100% gpu load

1660 -  65-75% gpu load

RTX 4000  60-70% gpu load

2080 TI - 35-40% gpu load 

 

I wish the dva3221 didn't have any hard cap of the number tasks you could run because I could use the 2080ti to its full potential even if it consumes a lot more 

 

Hello All,

Given the detailed analysis on GPU performance for the XPEnology DVA3221 project, focusing on the GTX 1650 and 1660, I'm planning to build a test tower with used components. Considering the performance insights, I'm pondering between GDDR5 and GDDR6 versions of the GTX 1650 or 1660. Which would you recommend for balancing cost and performance? Additionally, what processor would complement this setup for optimal functionality? Thank you!