DeepStack Case Study: Performance from CPU to GPU version

Has anyone checked to see if DS is core intensive? I personally have not. But know from mining crypto core might not be as important as memory clock. I have a p400 which can't be overlooked so no way for me to tell. I do know quadro chips, even tho the same, are the best of the batch. Silicone lottery is really a thing
 
I can't say specifically, but the number of available cores can be important to a lot of other software. Obviously clock speed has an impact as well, but if there's no available processor, core, for the data it won't help as much.
 
Is the 1030 and P400 still the go-to cards? I'm thinking of picking one up. My CPU load is quite a bit higher since enabling DeepStack

I wonder, can I then get OpenALPR/Rekor Scout to use the same GPU? Currently I run it on CPU in a different system
 
IReallyLikePizza2 said:
Is the 1030 and P400 still the go-to cards? I'm thinking of picking one up. My CPU load is quite a bit higher since enabling DeepStack

I wonder, can I then get OpenALPR/Rekor Scout to use the same GPU? Currently I run it on CPU in a different system
Click to expand...

I wouldn't call them the go to cards, but they are the most economical ones right now that will work with a 4th gen CPU and up. If you have a 6th gen or better, you would probably want to invest in a better card.

I have tried to get OpenALPR to run on the 1030 and it keeps timing out. It works great for DeespStack and HA for a few cameras.

Others have got the 1030 to work but it bumps the CPU up by 30% and then stays consistent there, which is weird.

I am currently looking into a Jetson Nano to offload OpenALPR to.
 
Interesting, maybe I'll get one and play with it

I have a spare Nvidia Shield TV Pro, a real shame I can't use it for something like this
 
IReallyLikePizza2 said:
Is the 1030 and P400 still the go-to cards? I'm thinking of picking one up. My CPU load is quite a bit higher since enabling DeepStack

I wonder, can I then get OpenALPR/Rekor Scout to use the same GPU? Currently I run it on CPU in a different system
Click to expand...

I got the 1060 w/3mb RAM and am currently running OpenALPR and DeepStack on the GPU with excellent results. If I were doing it again, I'd buy the 1060 again only with 6mb RAM instead of 3. OpenALPR only has 1 camera feed at this time, soon it will double. DeepStack is analyzing feeds from 10+ cameras.

From everything I've read here and at the DeepStack forum, while there are many versions of the nVidia card that will work, the 1060 offers a very good balance between performance and efficiency. My system runs 24x7 so if I had one of the higher performance cards such as the 1080 or the new 30-series, the GPU would be sucking far more energy and have so much performance headroom it wouldn't be sensible. Yes, it would be faster, but my DS times are already sub 50ms, so the incremental improvement would not be worth the extra power demand, not to mention the initial outlay for those high-end cards right now.

Having said that, my system is pretty robust in that my CPU is an i9-9900k. Not the latest in Intel chips, but still pretty beefy compared to most. Whatever you are running as a CPU will likely play a part in your selection of GPU cards.
 
icpilot said:
I got the 1060 w/3mb RAM and am currently running OpenALPR and DeepStack on the GPU with excellent results. If I were doing it again, I'd buy the 1060 again only with 6mb RAM instead of 3. OpenALPR only has 1 camera feed at this time, soon it will double. DeepStack is analyzing feeds from 10+ cameras.

From everything I've read here and at the DeepStack forum, while there are many versions of the nVidia card that will work, the 1060 offers a very good balance between performance and efficiency. My system runs 24x7 so if I had one of the higher performance cards such as the 1080 or the new 30-series, the GPU would be sucking far more energy and have so much performance headroom it wouldn't be sensible. Yes, it would be faster, but my DS times are already sub 50ms, so the incremental improvement would not be worth the extra power demand, not to mention the initial outlay for those high-end cards right now.

Having said that, my system is pretty robust in that my CPU is an i9-9900k. Not the latest in Intel chips, but still pretty beefy compared to most. Whatever you are running as a CPU will likely play a part in your selection of GPU cards.
Click to expand...

Interesting, that sounds like a pretty good solution

My system is 2u and so I can forced to low profile cards, unless I want to get interesting with a riser card and mount it on its side
 
I have a search for P400, T400, P620, P1000 and 1030, and a seperate one for 1050 and 1060, just incase
 
  • Like
Reactions: sebastiantombs
Oh wow, right you are. Maybe I'll skip my search for the 1050
 
Whats everyones thoughts on the Tesla P4? Looks like they go for around $320 and are pretty powerful
 
Well I ordered one, lets see how it does!

8GB RAM, and 2560 CUDA Cores
 
  • Like
Reactions: MikeLud1 and sebastiantombs
Realized this crappy board doesn't have any more good PCIe slots spare! All I had is the middle 1x slot with the CMOS battery behind it (So I cant cut the back off the slot)

Had to cut the NIC down to a 1x to make a 16x spare

1634840255049.jpeg

Now I have a 16x spare

1634840266702.jpeg

The electrical tape is because Gigabyte in their infinite wisdom put capacitors under where the card goes, and the metal contacts! And on the back it touches the CMOS battery itself
 
  • Like
  • Wow
Reactions: beandip, Pentagano and sebastiantombs
Really weird issue

I had an Intel Pro/1000 Quad Port NIC in there, and I had a spare I350-T4 NIC which is better and newer. I had originally bought it for this system but everything was just weird when I put it in the system when I built it. Blue Iris 100% CPU, half my RAM was gone and performance sucked

I've had the card on a shelf for years, and just tested it in my desktop, works fine on all ports nothing wrong! So I cut it down, throw it in the system and its weird again, yet again half my memory was gone and Blue Iris was running at 100%! WTH!

This was the fix, I put masking tape over the 5th and 6th PCIE Pin, all works well now. Never in my life have I come across such an issue, and never would I think this is the fix

 
  • Like
  • Wow
Reactions: Pentagano and sebastiantombs
I've just played around with my son's asus zenbook as I was curious about gpu.

nvidia geoforce mx350 card and ryzen 4000 series. 8GB memory.
Installed deepstack gpu, blueiris trial and AI tool.

Connected to one of my cameras. OMG just this cheap gpu is soo fast. It runs through the snapshots at less than 100ms, much faster than my i7 7700 with 32gb memory.
Deepstack on High mode also

cpu hardly rises, gpu15-18%.

Ok now I am interested in getting a decent gpu for my intel desktop!
 
  • Like
Reactions: sebastiantombs
Well I'm much more excited to get my GPU now! It gets here Monday
 
  • Like
Reactions: sebastiantombs and Pentagano
You must log in or register to reply here.
Top Bottom