Blue iris, why not virtualized?
- Thread starter bestbeast
- Start date
Hello guys,
So I have been trying on a dedicated machine with HA on/off, and found the following data with 6 CAMS:
Is it normal that only intel quick sync reduces the CPU load only just a 10% or so?
So I have been trying on a dedicated machine with HA on/off, and found the following data with 6 CAMS:
| BI | HA | CAM | CPU | GPU |
| open | off | 6 | 82 | 0 |
| open | on | 6 | 70 | 17 |
| close | off | 6 | 60 | 0 |
| close | on | 6 | 46 | 17 |
Is it normal that only intel quick sync reduces the CPU load only just a 10% or so?
In BI (bottom of Status section, Cameras tab), could you please share what the Totals from those 6 cameras are?
In BI (bottom of Status section, Cameras tab), could you please share what the Totals from those 6 cameras are?
This is funny, because it seems that the CPU stats I took before are not valid, they were taken from the same IP camera, now I placed 6 different IP cameras and my CPU (i5-4460) is going to explode.
If I run those 6 cameras with BL without HA, the CPU gets on 100% and everything starts to lag , and if I enable HA, it goes down to 70-80% if I leave BI minimized.
But for my i7-7820X 8c-16t even if the HA is dissabled, as it has no GPU, it does not get above 20% of usage, and if I enable the Nvidia GTX1080 I have, the CPU goes down to 15% more or less.
Anything I might be doing wrong for the I5? It seems to me very high CPU usage just for 6 IP cameras
I will be trying tomorrow though some experiments to passthrough GPU to some VMs. But at the first beginning lets sort out this high CPU increase.
If I run those 6 cameras with BL without HA, the CPU gets on 100% and everything starts to lag , and if I enable HA, it goes down to 70-80% if I leave BI minimized.
But for my i7-7820X 8c-16t even if the HA is dissabled, as it has no GPU, it does not get above 20% of usage, and if I enable the Nvidia GTX1080 I have, the CPU goes down to 15% more or less.
Anything I might be doing wrong for the I5? It seems to me very high CPU usage just for 6 IP cameras
I will be trying tomorrow though some experiments to passthrough GPU to some VMs. But at the first beginning lets sort out this high CPU increase.
Just another QQ, how is possible, that some NVR around 100-200$ can handle this amount of cameras without issues, and it is so overkill for an intel processor?
I believe NVR devices basically offload everything they can to the cameras (i.e. motion detection etc), and so they aren't trying to do as much as Blue Iris is trying to do. It would be more like if you configured Blue Iris to limit-decode, direct-to-disk, and disabled all motion detection in the software.
Regarding the: CPU (i5-4460 4c/4t) vs CPU (i7-7820X 8c/16t) question.
- i5 is 4 cores @ 3.4Ghz max clock versus i7-8 cores @ 4.5GHz max clock -- for simple threaded workloads the i5 is 25% as capable and operating with a 30% speed and 3 generations of IPC improvement handicaps. We have in some other discussions talked about how threads are not the same as cores (2T < 2C) and you have HALF the cores on the i5.
- i5 is Dual-Channel DDR3 memory config with 25GiB/s bandwidth. i7-7820x is Quad Channel DDR4 with ~80GiB/s bandwidth. The memory bandwidth is more than DOUBLE because both memory speed and controller memory channels are handicaps for the i5.
- see @bp2008 recent R9-3950x discussion (here: Ryzen 3950x for Blue Iris) where he tried to explore the upper limits of Blue Iris on a CPU where cores were not the limiting factor (much newer than both your processors).
Last edited:
Hello,
Regarding the first NVRs, yes, I can understand that those are just little crappy devices, that do not offer, same advance settings as blue iris can offer. Was also just wondering, about them for home user people, that only wants to storage their records at low price and low Watt cost.
Regarding the CPU comparison is quite interesting, how much memory speed can affect the CPU performance in this case. Will continue making some tests regarding this new info
The only thing I was quite dissapointed on all this, was about HA, people claiming some difference, but when I turned it on, it just reduced 10% the CPU load. Taking this into account, isn't it really better to buy an AMD CPU that has more cores, than trying with intel and quick sync?
Post back with your memory findings, but I definitely found @bp2008 findings VERY interesting because most home setups aren't expected to be near the top-end requirements, but in his case he was able to discover there is a memory-oriented limit we might need to be aware of for Blue Iris. Anything can become a bottleneck and make an otherwise very good system run sluggish, memory bandwidth (as far as I knew) hadn't previously been discussed much. However, some broader study might illustrate how the memory bandwidth/memory controller impacts Blue Iris performance, as like anything else (hard drive, CPU, GPU) once it becomes saturated it becomes a bottleneck device.
Regarding: HA only giving you 10% performance boost, I think it's more like that performance is basically "free" because it's just silicon sitting idle on the CPU die. Interesting to me would be, how much does average power draw change reducing the CPU by 10% (but maybe consuming slightly more power in the iGPU), if you can save 5-8% CPU load & lower resulting power draw over the long haul it's likely well worth.
Regarding: solving performance with AMD cores instead of Intel + Quick Sync. Sure maybe this could work, but jury is out due to limited sample data, since high core count AMD aren't currently widely available on the second hand market, and you might be "trading up" to a memory bandwidth issue anyway.
Regarding: HA only giving you 10% performance boost, I think it's more like that performance is basically "free" because it's just silicon sitting idle on the CPU die. Interesting to me would be, how much does average power draw change reducing the CPU by 10% (but maybe consuming slightly more power in the iGPU), if you can save 5-8% CPU load & lower resulting power draw over the long haul it's likely well worth.
Regarding: solving performance with AMD cores instead of Intel + Quick Sync. Sure maybe this could work, but jury is out due to limited sample data, since high core count AMD aren't currently widely available on the second hand market, and you might be "trading up" to a memory bandwidth issue anyway.
- R7-3700X 8-core CPU only has dual-channel (47 GIB/s) memory bandwidth. So (I'm guessing) it would out perform your i5-4460 (since it has more cores & bandwidth), but still fall behind your i7-7820x eventually (since both are 8c/16t but extreme processor has approx twice the memory bandwidth).
- If you would consider doing a test kind of like @bp2008 did, increasing MP/s on the i7-7820X we could at least learn where it reaches max capacity, and infer whether it hits CPU or Memory limits first.
- what would be interesting to me is (if your motherboard supports it) how much more performance an i7-4960X would get you instead of the i5-4660 (but you need a dedicated GPU then so higher energy cost).
It is my experience that when all the advice from here is followed (but with Limit decoding remaining OFF), Intel hardware acceleration reduces CPU usage by closer to half.
If you have 6x 6.3 MP cameras, and you configure those to encode at 15 FPS, then you end up with 567 MP/s which is higher than I would try to run on an i5-4460. Reduce frame rates to 10 FPS to get 378 MP/s and that is much more reasonable. Still maybe a bit heavy for that CPU though!
Yes, I have to try and test the Watt consume that 10% makes, if it goes higher because of the IGPU or goes down. Once I test that I will let you know.
Also I was able to optimize a bit more the system, by lowering to 10 FPS and also a bit the MP. Apart from that I am now running into 23% CPU with HA enabled.
Regarding the ryzen and intel, sure my i7 could deal better because of the 4 channel memory, in case of ryzen, I would be thinking about the new threadripper gen, which also has the same specs, that could be interesting to be tested.
As @bp2008 did, in case I have some little time I will be doing some more tests, but at the moment, I will try to focus on the passthrough to VMs, to see if I can achieve HA working on them successfully.
Nice also to have some sort of CPU list as a basis
Yes, at the moment I have everything running smoothly at 10FPS 250MP/S with just a 20% CPU usage. But lol, I have enabled limited-decode-if-necessary option, and it went down to 5% of CPU usage :/, so need to check what is that option exactly doing.
Kind regards
Hey guys,
So finally I achieved igpu passthrough to my Windows 10 VM though promox and UNRAID systems. Saying that UNRAID was easier, as it worked on the last version, but with proxmox I has to use the previous version of the software to make everything stable.
About the performance inside the VM, I can fairly say that it is pretty much identical as outside of it, maybe a 2% more of CPUglobal usage, because of the hypervisor stuff, but apart from that, I was really impressed.
Another thing to say, is that the last controller of the intel graphics was giving me error 43, so trying some of the old controllers made the trick. Letting you know @bp2008 as it seems you posted elsewhere that you were having trouble with that also.
So at the very end the conclusion now, is that I will be able to use this device for other stuff (80% cpu left) at the same time as recording my CCTV videos
Kind regards
So finally I achieved igpu passthrough to my Windows 10 VM though promox and UNRAID systems. Saying that UNRAID was easier, as it worked on the last version, but with proxmox I has to use the previous version of the software to make everything stable.
About the performance inside the VM, I can fairly say that it is pretty much identical as outside of it, maybe a 2% more of CPUglobal usage, because of the hypervisor stuff, but apart from that, I was really impressed.
Another thing to say, is that the last controller of the intel graphics was giving me error 43, so trying some of the old controllers made the trick. Letting you know @bp2008 as it seems you posted elsewhere that you were having trouble with that also.
So at the very end the conclusion now, is that I will be able to use this device for other stuff (80% cpu left) at the same time as recording my CCTV videos
Kind regards
I have been running Blue Iris in a VM for a year now no issue. I migrated my license and settings to it and have not had an issue.
I was using Proxmox 5.4 and currently am on 6.1 with no issues. I am not passing through a GPU, but have successfully passed through the hard drive to the VM instead of a logical volume. This will come in handy if the situation comes up that I need to hand the hard drive and footage over.
If you have a server with extra resources you should have no issues throwing more at BI if you need to scale up as well.
I was using Proxmox 5.4 and currently am on 6.1 with no issues. I am not passing through a GPU, but have successfully passed through the hard drive to the VM instead of a logical volume. This will come in handy if the situation comes up that I need to hand the hard drive and footage over.
If you have a server with extra resources you should have no issues throwing more at BI if you need to scale up as well.
I've found that a lot of things can make a difference with how BI performs as a VM. One of the biggest things I did was upgrading the CPU on the host. Jumping from a low-end Xeon to the highest one our Poweredge R710s could handle resulted in a significant reduction in CPU usage and we found the program itself as well as the remote web view worked a lot smoother.
We are currently running Blue Iris on a Windows 8 Pro VM under Hyper-V Server 2016. Host computer is a Poweredge R520 with a Xeron E5-2430 2.2ghz CPU. The VM has 16gb of ram (just about 6gb in use) and 8 virtual processors. We have about 34 IP cameras of varying quality, most set to 10 FPS. All 8 virtual processors are nearly maxed out at 90%. The host CPU runs at about 30%. There is only one other VM on the host and it doesn't use up much.
We also found that switching to SSDs made a difference. We have a PERC H310 running 8 512gb Samsung 850 Pros in a RAID 10.
This host also acts as a replica for an identical R520 that runs a few other VMs. If that host were to go down and we have to launch the replica copies on the BI host I imagine we'd have some problems. Or vice versa of course. But I'm a huge proponent of virtualization so unless I have no other choice I'm always going to go that route and recommend that others do that same.
We are currently running Blue Iris on a Windows 8 Pro VM under Hyper-V Server 2016. Host computer is a Poweredge R520 with a Xeron E5-2430 2.2ghz CPU. The VM has 16gb of ram (just about 6gb in use) and 8 virtual processors. We have about 34 IP cameras of varying quality, most set to 10 FPS. All 8 virtual processors are nearly maxed out at 90%. The host CPU runs at about 30%. There is only one other VM on the host and it doesn't use up much.
We also found that switching to SSDs made a difference. We have a PERC H310 running 8 512gb Samsung 850 Pros in a RAID 10.
This host also acts as a replica for an identical R520 that runs a few other VMs. If that host were to go down and we have to launch the replica copies on the BI host I imagine we'd have some problems. Or vice versa of course. But I'm a huge proponent of virtualization so unless I have no other choice I'm always going to go that route and recommend that others do that same.
At the moment I am running also proxmox 5.4 as it works for my case (new proxmox version was not working with iGPU passthrough), with iGPU passthrough (which I highly recommend), just think that the amount of CPU % that you can free up, that amount of cpu you will be able to use for other tasks. The only downside, is that I can not passthrough the iGPU to 2 different VMs at the same (which is impossible, so wouldn't consider it a downside), or that BI is not linux compatible.
I saw some people arguing about why should you like BI to run into linux, why not just creating a windows VM and thats it. Well, for that people, there is just a simple explaination, on intel cpu, you only have 1 iGPU, and to be honest if you have other linux services (as most server services are built for linux) that make use of HA, then you might like to be able to run all those services in the same VM, so that all of them can make use of HA, when needed.
At least, I was lucky enough, that some other linux services I was running on linux, are also compatible with windows, but it is not the preferred way, I wanted things to go. So probably in a near future, I will go and try other CCTV systems like it might be Zoneminder.
Apart from this, I am quite happy about the passthrough option in proxmox.
Ive had Blue iris running in a vm for a few years now and its been flawless.
Started off using unraid, now using esxi and also running unraid in a vm. Quick sync/igpu had no issues being passed through. I certainly cant see any downsides and its allowed me to save some $$$ by having 1 system running pfsense,unraid,blue iris 24/7.
Started off using unraid, now using esxi and also running unraid in a vm. Quick sync/igpu had no issues being passed through. I certainly cant see any downsides and its allowed me to save some $$$ by having 1 system running pfsense,unraid,blue iris 24/7.
Yep, I totally agree with you. Virtualization was designed with that purpose mainly, mostly saving costs, there is no sense to have more than 1 system. According to the passthrough, there are multiple issues when trying to use igpu, indeed I expend like 1 week to achieve mine to work, firstly because of some hypervisors, and secondary because of windows drivers messing around.
Hello,
I have been searching around, and it seems there is also much discussion about using dedicated GPUs for blue iris, as it can be a lost of energy. I have been searching around and found P400 (MAX 30W) and P600 (MAX 40W), don't know if anyone has tried, but it also seems to be a decent solution in case igpu can not be used, or is already is use for other stuff, and not that expensive for decoding h264 and h265.
Has anyone tried?
Kind regards
I have been searching around, and it seems there is also much discussion about using dedicated GPUs for blue iris, as it can be a lost of energy. I have been searching around and found P400 (MAX 30W) and P600 (MAX 40W), don't know if anyone has tried, but it also seems to be a decent solution in case igpu can not be used, or is already is use for other stuff, and not that expensive for decoding h264 and h265.
Has anyone tried?
Kind regards
Another option is Intels new gpu which I hear is going to land this year. If it works and its affordable then it would be a great option to build a system with amd/ryzen cpu's.