SSD, nvme or mechanical?

wittaj said:
However, with substreams being introduced, the CPU% needed to offload video to a GPU is more than the CPU% savings seen by offloading to a GPU. Especially after about 12 cameras, the CPU goes up by using a GPU and hardware acceleration.
Click to expand...

Just to clarify for myself and for the OP, (I haven't set my BI server up yet due to other more important financial committments (it is built)), with BI and DS, is it now better NOT to have a GPU than to have a GPU? I have a P620 v2 quaddro but with energy prices in Europe, I'm seriously thinking I'm better off without the extra 50W if there's no gain or even a loss to adding it.

Does a BI / DS combination work as well / better now on CPU alone rendering a quaddro unecessary or are there still advantages to a quaddro that justfy the power draw?
 
CCTVCam said:
Just to clarify for myself and for the OP, (I haven't set my BI server up yet due to other more important financial committments (it is built)), with BI and DS, is it now better NOT to have a GPU than to have a GPU? I have a P620 v2 quaddro but with energy prices in Europe, I'm seriously thinking I'm better off without the extra 50W if there's no gain or even a loss to adding it.

Does a BI / DS combination work as well / better now on CPU alone rendering a quaddro unecessary or are there still advantages to a quaddro that justfy the power draw?
Click to expand...

That is correct as it relates to hardware acceleration within BI.

However, A GPU used for Deepstack will result in faster response times on object identification over the CPU version.

So it comes down to what the response times are in your system. As an example, if the DS CPU version returns 150ms response times and the GPU version returns 40ms response times, then only you can decide if 110ms savings in time using the GPU is worth the extra energy expense. For many, even 1500ms is sufficient and not worth the extra energy costs.
 
  • Like
Reactions: CCTVCam
The "WD Ultrastar DC SN630" has a U.2 interface, which will not work in your computer. This is a very good NMVE, but it needs over 5W even when idle and over 10W on high load. A nvme like Corsair Force MP510 with 2tb would be good enough for any NVR usage in your system. You´ll run out of CPU cycles or IOPS before you even come close to any TBW limits (+3PB) for a decade. The problem is that you would like to save power. A good M.2 NVME needs around 7-10w depending on your load. A Toshiba NVR 3.5" HDD is half the price and also has half as economical.

An ssd would be more economical, but the market for ssd's with high TBW is practically dead and the selection is getting worse. I´ If you reduce your requirements to 1PB TBW, a simple Samsung EVO 870 would be sufficient.

But if you use proxmox on the host, you must also consider the filesystem used. Proxmox currently supports 3 copy-on-write filesystems which cause significantly more writes for checksums etc. Make sure you format your disk with ext3 or ext4.
 
Jake1979 said:
I'm building a small i5 NUC for my BI NVR and the OS will be installed on a spare NVME or SSD that I have but for the video storage, I can pick SSD or mechanical. For longevity, what is best? I'm leaning towards a 2.5 mechanical, probably a 2TB. Thoughts?

My goal with the NUC is to save on power.
Click to expand...
OS disk: an NVME or SSD will be fine. for video storage, I would use a Toshiba L200 If you were looking for a decent 2.5" hdd for video
 
spuls said:
The "WD Ultrastar DC SN630" has a U.2 interface, which will not work in your computer. This is a very good NMVE, but it needs over 5W even when idle and over 10W on high load. A nvme like Corsair Force MP510 with 2tb would be good enough for any NVR usage in your system. You´ll run out of CPU cycles or IOPS before you even come close to any TBW limits (+3PB) for a decade. The problem is that you would like to save power. A good M.2 NVME needs around 7-10w depending on your load. A Toshiba NVR 3.5" HDD is half the price and also has half as economical.

An ssd would be more economical, but the market for ssd's with high TBW is practically dead and the selection is getting worse. I´ If you reduce your requirements to 1PB TBW, a simple Samsung EVO 870 would be sufficient.

But if you use proxmox on the host, you must also consider the filesystem used. Proxmox currently supports 3 copy-on-write filesystems which cause significantly more writes for checksums etc. Make sure you format your disk with ext3 or ext4.
Click to expand...
Good catch, I didn't even pick up on the U.2... thank you.

I'm still going back and forth between Intel NUC and a small, 2U custom blade "server" with the most efficient PSU and i5. I'm looking at the Corsair suggestion as well as the Seagate Firecuda 520 2TB. Trying to find the heat off put on these as well.
 
Jake1979 said:
Good catch, I didn't even pick up on the U.2... thank you.

I'm still going back and forth between Intel NUC and a small, 2U custom blade "server" with the most efficient PSU and i5. I'm looking at the Corsair suggestion as well as the Seagate Firecuda 520 2TB. Trying to find the heat off put on these as well.
Click to expand...

SSD and NVME uses about the same power as a regular HDD in this application.
 
wittaj said:
That is correct as it relates to hardware acceleration within BI.

However, A GPU used for Deepstack will result in faster response times on object identification over the CPU version.

So it comes down to what the response times are in your system. As an example, if the DS CPU version returns 150ms response times and the GPU version returns 40ms response times, then only you can decide if 110ms savings in time using the GPU is worth the extra energy expense. For many, even 1500ms is sufficient and not worth the extra energy costs.
Click to expand...


Thanks. Processor is 11400 and likely max load 4-5 cameras although 1 initially.
 
You must log in or register to reply here.
Top Bottom