Hardware acceleration (QSV/NVENC) vs direct to disk for lower CPU load?
-
- Posts: 3
- Joined: Fri May 29, 2020 2:19 am
Hardware acceleration (QSV/NVENC) vs direct to disk for lower CPU load?
To squeeze the most from the CPU(s) is it better to have Hardware acceleration of Direct To Disk?
My feeling is direct to disk is the move here and then hardware acceleration is unnecessary.
What you think?
This is for a new build, trying to decide if I need an I7 of can get away with AMD.
new build for: 18 cameras, 1080, motion detection recording.
My feeling is direct to disk is the move here and then hardware acceleration is unnecessary.
What you think?
This is for a new build, trying to decide if I need an I7 of can get away with AMD.
new build for: 18 cameras, 1080, motion detection recording.
Re: Hardware acceleration (QSV/NVENC) vs direct to disk for lower CPU load?
I may not have some of the experience of some of the others here but my Xeon processors are tool old to have QSV and my server doesn't have any sort of added GPU for processing.... and when I turned on Direct-to-Disk it was like night and day. Along with 'Limit decoding unless required'... I've gone from 70%+ (and my server sounding like a jet engine) to around 10% most of the time.
That said, 18 cameras is a lot. With that large of an investment, I wouldn't skimp on your server side hardware.
That said, 18 cameras is a lot. With that large of an investment, I wouldn't skimp on your server side hardware.
Blue Iris 5.9.4.x | Server 2022 VM | Xeon E5-2660 v3 @ 2.60GHz - 16 Cores | 24GB RAM | 8TB RAID | Sophos UTM WAF | Mostly various SV3C Cameras
Re: Hardware acceleration (QSV/NVENC) vs direct to disk for lower CPU load?
Hardware acceleration should be a non-conversation now a days. That is like asking if it better to use a typewriter or word processor to write a term paper.
Asking your main CPU to do all of that video decoding is just not fair. Intel has had QSV since Sandy Bridge which is eons ago in computer time.
Direct-to-Disk is ALMOST a no-lose option and the only thing you give up by going to D2D is the BI generated overlays being recorded to disk. Everything else works fine. If you want the overlays and skip D2D then you essentially double the amount of encode/decode power you need. Not worth it in my view. Just have the cameras put the time stamps on and forget saving BI overlays.
Limit Decoding is the worst option as you are now limited to doing event detection only on key-frames and not all frames. I suppose if you had to, then sure. But it is the worst option.
Asking your main CPU to do all of that video decoding is just not fair. Intel has had QSV since Sandy Bridge which is eons ago in computer time.
Direct-to-Disk is ALMOST a no-lose option and the only thing you give up by going to D2D is the BI generated overlays being recorded to disk. Everything else works fine. If you want the overlays and skip D2D then you essentially double the amount of encode/decode power you need. Not worth it in my view. Just have the cameras put the time stamps on and forget saving BI overlays.
Limit Decoding is the worst option as you are now limited to doing event detection only on key-frames and not all frames. I suppose if you had to, then sure. But it is the worst option.
Re: Hardware acceleration (QSV/NVENC) vs direct to disk for lower CPU load?
If your key frames are 1 per second (as recommended), what kind of motion are you missing that is less than one second? More false positives? I suppose if you're running cinema quality cameras and want to see insects fly by, then sure. But I have never missed a human or vehicle, I'm fairly good at catching some random cat that comes by every night at 3am (not mine), and often catch my chickens - which I don't particularly want to detect. I'm sure there are use cases where Limit Decoding isn't ideal - but I'm yet to run into it in a practical normal sense. Also, watching a typical security camera at 1fps isn't really all that bad - generally NOTHING happens. My cameras still record at a much more respectable 15fps but really even that is a bit much. I'm not making a movie.
To each their own.
To each their own.
Blue Iris 5.9.4.x | Server 2022 VM | Xeon E5-2660 v3 @ 2.60GHz - 16 Cores | 24GB RAM | 8TB RAID | Sophos UTM WAF | Mostly various SV3C Cameras
Re: Hardware acceleration (QSV/NVENC) vs direct to disk for lower CPU load?
Starting with update 5.2.8.1, BI now implements BVR metadata recording. If you use D2D, you can use overlays AND play them back when you play back your recordings. When playing back, make sure that you right click the playback screen and enable 'Add text/graphic overlays from direct-to-disc BVR'.HeneryH wrote: ↑Fri May 29, 2020 4:49 pm ...
Direct-to-Disk is ALMOST a no-lose option and the only thing you give up by going to D2D is the BI generated overlays being recorded to disk. Everything else works fine. If you want the overlays and skip D2D then you essentially double the amount of encode/decode power you need. Not worth it in my view. Just have the cameras put the time stamps on and forget saving BI overlays.
...
Re: Hardware acceleration (QSV/NVENC) vs direct to disk for lower CPU load?
Didn't know that Mike, good news then. I'm on BI4.
What then, could POSSIBLY be any reason for NOT using D2D?
What then, could POSSIBLY be any reason for NOT using D2D?
Re: Hardware acceleration (QSV/NVENC) vs direct to disk for lower CPU load?
killerbotics: Now what's all this HA vs D2D talk ? Surely for lowest cpu load we should be using both
HeneryH: Take the RED pill. Nno wait, I meant the BLUE pill.
MikeBwca: Does this mean I can stop trying to sort out that darn Foscam local NTP server issue ? Will this load up the cpu ? Testing...
HeneryH: Take the RED pill. Nno wait, I meant the BLUE pill.
MikeBwca: Does this mean I can stop trying to sort out that darn Foscam local NTP server issue ? Will this load up the cpu ? Testing...
Forum Moderator.
Problem ? Ask and we will try to assist, but please check the Help file.
Problem ? Ask and we will try to assist, but please check the Help file.
Re: Hardware acceleration (QSV/NVENC) vs direct to disk for lower CPU load?
D2D will defiantly use less cpu as it doesn't re-encode the video.
I don't think the new bvr metadata uses any significant amount of cpu.
I don't think the new bvr metadata uses any significant amount of cpu.
Re: Hardware acceleration (QSV/NVENC) vs direct to disk for lower CPU load?
Well I can see where I stand... left out Guess I should go to the other community site ****KIDDING - and thats something that should never be joked aboutTimG wrote: ↑Sat May 30, 2020 9:34 pm killerbotics: Now what's all this HA vs D2D talk ? Surely for lowest cpu load we should be using both
HeneryH: Take the RED pill. Nno wait, I meant the BLUE pill.
MikeBwca: Does this mean I can stop trying to sort out that darn Foscam local NTP server issue ? Will this load up the cpu ? Testing...
Blue Iris 5.9.4.x | Server 2022 VM | Xeon E5-2660 v3 @ 2.60GHz - 16 Cores | 24GB RAM | 8TB RAID | Sophos UTM WAF | Mostly various SV3C Cameras
Re: Hardware acceleration (QSV/NVENC) vs direct to disk for lower CPU load?
If you use anything other than a business surplus HP or Dell with a consumer Intel chip and a fresh install of Win-10 and use this machine ANYTHING other than BI and you don't run behind a VPN then you are an .... per the other site admin.