Pentagano
Getting comfortable
Purchased! Had a few interested people and nothing else on the market here matched.
I'll let you know know how it goes
Hell Yeah! Direct Deepstack Integration - 5.4.0 - March 31, 2021
- Thread starter fenderman
- Start date
I'll let you know know how it goes
IReallyLikePizza2
Known around here
Got the P4 in there, now to get everything installed!
Attachments
Pentagano
Getting comfortable
Ahhh purchase canceled. Great messed me around.
Tells me after he only offers local pickup even though the publication offered delivery. Sounds odd. Dodgy area also
Last edited:
IReallyLikePizza2
Known around here
Well shit.
GPU load is literally zero, CPU is completely untouched by Deepstack, 10-20x faster analysis. It almost feels like I'm doing something wrong...
Look at the dots under GPU load, thats how much utilization. This is nuts
GPU load is literally zero, CPU is completely untouched by Deepstack, 10-20x faster analysis. It almost feels like I'm doing something wrong...
Look at the dots under GPU load, thats how much utilization. This is nuts
IReallyLikePizza2
Known around here
I ordered a clip on fan for this, as these cards designed to be in servers are expecting a bit of airflow which I do not have
Nvidia Tesla Cooling Fan Shroud Mounting Kit for T4 P4 M4 GPU Accelerator Card | eBay
This is a 3D-printed fan mounting kit for cooling server-grade Nvidia Tesla GPU accelerator cards. The fan shroud slides on top of the GPU and locks into the notch in the heatsink above the PCIE locking tab.
www.ebay.com
As an eBay Associate IPCamTalk earns from qualifying purchases.
Pentagano
Getting comfortable
65ms? Woww.
What gpu is that sorry?
IReallyLikePizza2
Known around here
This is the Nvidia Tesla P4
IReallyLikePizza2
Known around here
IReallyLikePizza2
Known around here
I think the Utilization I showed is not accurate, as that's probably 3D performance maybe? It just can't be that low
For whatever reason I can't see utilization in task manager. Not sure if the Windows Server 2019 Driver maybe doesn't include that support? Unsure. I'll work on it
For whatever reason I can't see utilization in task manager. Not sure if the Windows Server 2019 Driver maybe doesn't include that support? Unsure. I'll work on it
I don't recall exactly, but I think it was the 960 and lower models which don't have the necessary processing power to support DeepStack. Maybe do a search of the forum here for that card and see what it turns up.
sebastiantombs
Known around here
The 9xx series is a power hog when compared to the 1xxx series if that makes any difference. IReallyLikePizza is having excellent results with a Tesla P4 which is eve stingier with power than the 1xxx series.
Last edited:
MikeLud1
IPCT Contributor
What mode do you have DeepStack running Low, Medium, or High with the 65ms results also is DeepStack using main or substream at what resolution?
IReallyLikePizza2
Known around here
Its set to Medium, should I set it higher?
Its using the main stream at 4PM
Its using the main stream at 4PM
MikeLud1
IPCT Contributor
Give High a try to see if you get the same results.
I am running a RTX 3060 with DeepStack set at high and using 1080P image and getting results at 90ms +-5ms. I had to do some hacking to get DeepStack to work with the RTX 3060 so I think DeepStack is not optimized to use the full power of the RTX 3060
MikeLud1
IPCT Contributor
To get my RTX 3060 to work with DeepStack I had to do the below steps
Install
CUDA Toolkit 11.4
cuDNN v8.2.4
DeepStack-Installer-GPU-2021.09.1.exe
Manually update the below Windows Packages
numpy-1.21.2-cp37-cp37m-win_amd64
Pillow-8.3.2-cp37-cp37m-win_amd64
scipy-1.7.1-cp37-cp37m-win_amd64
torch-1.9.1+cu111-cp37-cp37m-win_amd64
Torchvision-0.10.1+cu111-cp37-cp37m-win_amd64
Install
CUDA Toolkit 11.4
cuDNN v8.2.4
DeepStack-Installer-GPU-2021.09.1.exe
Manually update the below Windows Packages
numpy-1.21.2-cp37-cp37m-win_amd64
Pillow-8.3.2-cp37-cp37m-win_amd64
scipy-1.7.1-cp37-cp37m-win_amd64
torch-1.9.1+cu111-cp37-cp37m-win_amd64
Torchvision-0.10.1+cu111-cp37-cp37m-win_amd64
IReallyLikePizza2
Known around here
Okay cool, I set it to high
What do you guys have set for each camera and the number of images to, and how long for each?
What do you guys have set for each camera and the number of images to, and how long for each?
How many images and how long depends on each camera field of view (as well as the computer power you have LOL)
For a field of view that is say just a person going to a front door you can probably get by with the bare minimum.
For a camera that is to tag cars, you probably need more at night as headlight shine on the road may trigger it early and come back with nothing found.
But if you have a screaming enough setup, why not go ahead and tell it to send as many images as you have the trigger set to be.
For a field of view that is say just a person going to a front door you can probably get by with the bare minimum.
For a camera that is to tag cars, you probably need more at night as headlight shine on the road may trigger it early and come back with nothing found.
But if you have a screaming enough setup, why not go ahead and tell it to send as many images as you have the trigger set to be.
MikeLud1
IPCT Contributor
How is the performance set a High, are you still getting about 65ms
I am running 10 images at 500ms
Looks promising, what are your '+ real-time images' settings? default or higher? I've got most of my cameras set above 20 at the moment.
