CPU / GPU requirements

[Sebastiaan]

At first I was considering using a Synology NAS ad the base for my camera system but hardware plus additional license costs made me look at alternatives. I found ZoneMinder and now want to built my own system.

I’ll be running about 16 cameras. Some full hd and some 4k resolution and most will have event detection enabled for which i’d want to use the AI plugins.

While it’s easy to calculate the storage needs I’m stuck on CPU and GPU requirements. I’ve read some of the topics in this forum and if I understand correctly an i5 type CPU is capable of running about 8 cameras in lower resolution. Now before I run to the store to get the 32 core AMD Threadripper I’d like to check id there’s any way of calculating CPU / GPU requirements or any similar experiences.

Are there any more experienced ZoneMinder users that can share some insight?

[iconnor]

The problem with 4k cameras is that everytime you double the resolution, you are actually multiplying by 4. So 4k res takes a lot more ram.

default buffers is 50, so 16 cameras * 50 * 3840 x 2160 *4bytes per pixel = 26Gb of ram. Which needs to be < half of ram. So you will be wanting around 64Gb of ram.

You will likely need more than an i5. You would need at least 4 real cores. More is better obviously.

[Sebastiaan]

Thanks, so having enough RAM is more important to watch out for than having a beefy CPU. Not all cameras will be 4k but still even 64 GB is doable with current memory prices.

The default buffer of 50, is that the amount of frames?

[snake]

It's not just RAM but CPU as well. You need both. No free lunch. Look at the RPI. The RPI4 and RPI3 are about comparable in performance with ZM. It doesn't matter much even if you add the 4GB of RAM for RPI4, in my experience. CPU is the bottleneck.

GPU is not used with ZM. There is no such thing as GPU requirements.

Basically, you'll want server motherboard hardware if you are going to put any significant number of cameras on board. Some server CPUs are more powerful than others.

[iconnor]

In 1.33.14 you can use gpu for decoding. In 1.35 you will be able to use it for encoding as well.

[CamNewbie]

My situation is exactly the same as the original poster of this question - Synology Surveillance Station plus associated costs are very high so I want to build my own.

I gather from this discussion that the AI plugins need large RAM buffers to perform analysis. A 24x7 surveillance system without detection capability is kind of useful only for post-incident manual analysis so I would like the AI turned on and plan the hardware requirements for it.

1. Core i5 (atleast)
2. Lots of RAM - to keep costs reasonable, buy a motherboard with lots of RAM slots because dense high-capacity RAM sticks are $$$$ compared to less dense RAM sticks but I can put in lots if the motherboard has the slots.
3. GPUs - So, which GPUs are recommended/tested by the devs ? I don't want to run into support issues where the devs cannot debug issues because I got a new-fangled GPU that they don't have.

Any specific requirement on the HDD speed ? Will a mirrored software RAID 1:1 5400 RPM pair work ?

[iconnor]

we have tested with a variety of nvidia/amd gpus. So far nvidia with cuda has been the best.

You may want the db on something faster, like ssd.