Which new Intel NUC?

Which Intel NUC CPU would be sufficient for these needs?


  • Total voters
    4
  • Poll closed .
The hardware wiki is a little dated and doesn't truly account for the substreams.

With the substreams, you can turn an older slower machine into a capable unit. But now with DeepStack added, that adds a new wrinkle that could then hinder the older slower machine.

But most stay away from NUC because they are optimized for low power. Not really designed for a 24/7 constant load that a VMS requires.

Some computers to stay in low power will slow the CPU down to keep the temperature lower. Now you start missing motion triggers and other things.
 
I think you're not getting a direct answer because hardly anybody or maybe nobody who's active here is running a NUC. You'll be the test pilot and find that the common wisdom is right, or perhaps it's wrong. I've been on the receiving end of the same thing with chinese market Dahua cams. "Everybody" here says they're horrible and must be avoided at all costs, yet I've had great success with a bunch of them for ~5 years. It does take dealing with their idiosyncrasies and limitations and using the right models to make it work, and perhaps you'll be in a similar situation with the NUC. It's very well something I could try, knowing up front that I might have to do it over after learning the hard way. I've left a large heap of ruined or unused hardware in my wake over the years, but learned a lot while creating the heap. (My chinese market camera comments aren't meant to encourage their use. The situation and models have changed since 5 years ago, and with Andy as a source for the international models, I see no point in messing with new ones now).
 
I think you're not getting a direct answer because hardly anybody or maybe nobody who's active here is running a NUC.
This is an important point. However, there are numerous entries in threads, and maybe actual threads, that discuss NUCs being used for BI. Do a search for them and maybe ask those specific users the questions?
 
  • Like
Reactions: looney2ns
Build it, Build it, Build it!

mob1.jpg
 
Great comments, everyone!

The NUC being a sticking point is that my dad is 100% sold on the form factor and silence of it.

Just in case I could convince him to compromise a bit, can anyone suggest a SFF machine that's as close as possible to the size of the Intel NUC?

I've showed him some SFF cases, and due to the fact that they're significantly bigger than the NUC, he wasn't interested. Although I feel like if I can find something that's just a 'little' bigger, he may be convinced.
 
Fenderman ( known around here) suggested awhile back that the HP Elitedesk SFF had really everything you need to make an efficient NVR with BI.
It's light. It's quiet. its efficent. and no tools required to open it up and upgrade stuff.
It's amazing that Dell can have the same Form Factor, and yet cannot allow for a 2nd 3.5" HDD.
The biggest factor to me was 2 bays for hard drives. 2 full sized HDD's.
I have an 8TB and a 5 TB splitting the writing load for 15-16 Cameras.
I guess the industry is trending away from Big drives. here is what I have, but I got one used on ebay. $600-700 cheaper.
 
Last edited:
  • Like
Reactions: looney2ns
As an eBay Associate IPCamTalk earns from qualifying purchases.
s-l1600 (1).jpg

The 2 HDD's can lay end to end, and windows runs off an M.2 drive down on the motherboard.
 
Christ, I ought a buy that at that price! but wait i have 7 computers already....

That's an i5-8500 though. Would it be powerful enough to run my 8x 4K cameras + Deepstack?
Wouldn't an 11th gen i7 be much more powerful than this, even if it is 'low power'?
 
As an eBay Associate IPCamTalk earns from qualifying purchases.
OK, I HAVE run BI on a few Intel NUC's, from a celeron, YEAH !! for testing, but mainly 2 x i7 5th and 7th gen machines.

I had the machines and decided to test with them, I have the full height units units, so I have 2 drives (SSD), I used Windows 10 professional for the OS.

Operation is acceptable, they both have 16gb of memory and pretty fast drives, I also pass storage to a mapped NAS when needed.

BUT and this is the issue as 'I' see it, the CPU is an i7, but think of it as a cut down version, or maybe a laptop variant if you need to, it is NOT the full meat eating i7 in any way shape or form, and as such you will without doubt...

Hear the FAN spinning up and down a LOT, I repeat A LOT, so the initial remit of the quiet, cool and power saving functions will be pissed out the window very quickly.

And before the shitstorm start, I strip the units down, thoroughly clean and prepare the CPU with new high quality thermal paste, I also check and replace if needed the 'gum' tabs on the VRM/Memory chips, I clean and lubricate the fan unit and make sure the BIOS settings are optimized for noise/performance, I test each setting in the BIOS to see if it will reduce the spinning up and down on the fan, NADA, it appears to me on the units I have to make no difference.

As an aside, I also have an old Dell 9020 i3 SFF unit that I ended up using instead, it stays cooler, WAY more silent and just works, THAT is what people are eluding to when 'pushing' you towards the SFF and not the NUC devices.

As I have said, I have done some extensive testing on at least 3 NUC devices as well as a SFF unit, my aim was for a quiet unit, I have no issues with power consumption, but I do not want to have some big ugly PC sat around, and my honest opinion is that a NUC is not a long term solution.

You also mentioned running additional process driven apps as WELL as BI, this will only make things worse, too may times people want to drop the likes of PLEX server on as well, thinking the machine is sat around doing little, and yup did that too during testing, and some people will get away with doing all these things and say that they never see any performance issues, my testing showed otherwise, and as always YMMV as YOUR expectations may/will be different to mine, you (or your dad) may well be happy with the outcome, the unit may be out of hearing/sight and not be of concern and the performance acceptable.

To answer the original ?, Both will work, but not without compromises.
 
Thanks so much for your informative post, Silas. Very well-researched.

I still think this leaves a HUGE jump in size between the NUC and a Dell/HP SFF PC.

Another sticking point is that most of these units have an optical drive - which is completely useless for his purposes. And If I just remove it, it will just leave a gaping hole in the chassis, right?

Also another question that nobody seems to have answered yet - does Deepstack take advantage of Tensor core CPUs? I have a spare Google Coral USB that I can donate to him if it helps take load off the CPU.
 
I am aware that DeepStack has a CPU and GPU version.. Your test with one camera is not realistic at all. Your proposed setup is eight 4K cameras. analyzing first for motion detection to determine what DS analyzes, then the DS analysis kicking in will push any CPU to probably close to 100%, except possibly the latest and greatest ones and even those will get pushed well into the 80% area or greater, especially with multiple simultaneous triggers. Remember when BI is triggered it switches from sub stream to main stream adding more load as well. In a NUC that will result in excessive heat plus slow detection times to the point where DS starts timing out. Trust me, been there and done that which is why I say the right way is to use a GPU for DeepStack. I'm trying to save you time, trouble, wasted effort and wasted money by trying to steer you away from the NUC form factor. It's just too limiting.

On another note if your buddy ever tries another camera besides those 4K ones he might be very pleasantly surprised at how the video quality improves at night. The point being that chasing megapixels is another loosing strategy. Chasing sensor size is the right approach.
 
And this isn't like a computer that needs to be on the desk or close by. Stick a SFF anywhere there is an internet connection, or add one to a closet, basement, behind the TV, etc. and then Remote Desktop or TeamViewer into when needed. The thing doesn't need to be visible.

Most of us here run the BI server without a monitor connected to it.
 
That's an i5-8500 though. Would it be powerful enough to run my 8x 4K cameras + Deepstack?
Wouldn't an 11th gen i7 be much more powerful than this, even if it is 'low power'?
Im running 15 Cameras. (Mixed 4 Mp & 2 Mp) on a 6 core i5 8th generation CPU. I can run the whole system with no substreams. ( but i dont) At a commercial property. 1 car entering the property causes Motion triggers on 5 cameras. Couple that with 100 residents milling around. 24-7 recording on 14 cams. And its very common to have 8-10 motion triggers
Nearly simultaneously.
I feel confident that it could run 4-5 more cams.
 
  • Like
Reactions: sebastiantombs
No deepstack needed for my system.
 
Thanks so much for your informative post, Silas. Very well-researched.

I still think this leaves a HUGE jump in size between the NUC and a Dell/HP SFF PC.

Another sticking point is that most of these units have an optical drive - which is completely useless for his purposes. And If I just remove it, it will just leave a gaping hole in the chassis, right?

Also another question that nobody seems to have answered yet - does Deepstack take advantage of Tensor core CPUs? I have a spare Google Coral USB that I can donate to him if it helps take load off the CPU.
So no matter what folks post, you resist the guidance given. So just buy one and see. Your dad wants a NUC, Just do it.

You do not have to remove the optical drive. Just don't use it. If you are worried about it taking up power, disconnect it. You could buy one of those blank panels to fill out the open slot, or put an HD in there.

You are getting no answer on the Tensor core CPU probably because no on knows what one is and the answer. Send a question to BI support to get the proper answer.
 
  • Like
Reactions: looney2ns
I am aware that DeepStack has a CPU and GPU version.. Your test with one camera is not realistic at all. Your proposed setup is eight 4K cameras. analyzing first for motion detection to determine what DS analyzes, then the DS analysis kicking in will push any CPU to probably close to 100%, except possibly the latest and greatest ones and even those will get pushed well into the 80% area or greater, especially with multiple simultaneous triggers. Remember when BI is triggered it switches from sub stream to main stream adding more load as well. In a NUC that will result in excessive heat plus slow detection times to the point where DS starts timing out. Trust me, been there and done that which is why I say the right way is to use a GPU for DeepStack. I'm trying to save you time, trouble, wasted effort and wasted money by trying to steer you away from the NUC form factor. It's just too limiting.

On another note if your buddy ever tries another camera besides those 4K ones he might be very pleasantly surprised at how the video quality improves at night. The point being that chasing megapixels is another loosing strategy. Chasing sensor size is the right approach.

Then why did you say it required a Cuda capable GPU?
Already that was objectively incorrect.

To be fair, the night performance of the Annke c800 has already exceeded our expectations.
Which model do you recommend?
 
They use laptop optical drives in the HP. Takes very little space. Not a 5.25” monster drive you’d find in a Tower Pc.