Hikvision NVR Load Tested

By Ethan Ace, Published Dec 14, 2017, 11:57am EST

IPVM members recently debated Hikvision NVR's performance under load in Hikvision 30+ Cameras On NVR - Apps And Client Really Slow Down And CPU Usage Jumps - Any Other Experiences Or Tests?

To see how Hikvision's NVRs and iVMS-4200 software handled heavy use, we tested a Hikvision DS-9632NI-I8 NVR, viewing 32 cameras using over 400 Mb/s bandwidth on multiple client PCs to see:

  • How does iVMS-4200 handle 32 cameras across multiple monitors?
  • How do settings such as resolution, frame rate, etc. impact this performance?
  • How much impact do CPU and memory have on iVMS performance?
  • How does the NVR stand up to heavy load? Are there issues with frame drop or lost recordings?

*******

** *** *****, **** client ******** ******* *********'* minimum ****-**** ***** *** few ****** ******* ** cameras ******** **** *** Mb/s *********. ********** ***** issues ******** (******** *******, long *********, *** *******) but ***** ******** ********** of ***, ***, *** memory *****. 

*******, ** *** ** load ****** ** *** DS-9632 ******, **** ** issues ***** *** ***** UI, ** **** ** recording, ******* ******, ** other ****** ***** ***** indicate *** *** *** overloaded, **** *** ****** of **** **** * week ** *******.

Key ********

**** *** *** *** findings **** **** ****:

  • ******** ****** ******* ******:** *** ***** ** experienced ******** ****-**** ******, with ******* ********, **** latency, *** ********* ******, which *** *** **** related ** ********** ******, number ** ******* ******, or **********, ******* ********* random.
  • ********** ****** *****:************, ****-**** ************ ******* outright, ********* **** ********** to **** **** ********** streams ** *** ******* and ********* **** ****** viewing ******** ******* ****.
  • ****** *** **** ****** widely:*** **** ******* ** cameras ****** **** ~*-*% on **** *** ******** to **** **-**% ** higher ** ******* ****** meeting *********'* ******* ****** requirements.
  • ****** ****** ***** ******:** ********, ****** ***** was ********** ****** ****** all ******** ******, ***** about ***-*,*** ** *** when ******* ** *******.
  • ** ***** *** ** web ** ******:****** ****-****, ** *** none ** *** ***** issues **** ******* *** cameras ******* ** *** NVR.
  • ** ********* ******:** ***** ** ******* frames, **** ** *********, stream *******, ** ***** issues ** ******** ***** which *** ***** **** NVRs ***** ***** ******* throughput *****, ******* *** NVR ********* *** ******* near *** *** ********** specifications (*******/*******).

iVMS ****** ****** ************

** ****** ****-**** ** multiple ****** ********, **** differing ***, ******, *** GPU ***** ** *** how ** ********* ** various ********. *** *** were ******* *** ****** versions ** ******* ** Professional.

**** **** *** **** end ******* ****** **************'* "**** ***********" ***** for ****, *** *** *** end ******* ******* *** minimum. **** **** **** Hikvision ***** ********* ****** requirements *** **** ** different *********, **** **** ******* **** * ******* IV ** ******** ***. The ****** ************ ***** on ******* ******** **** ** ** date, *** ** ***** is *** *********, ** recommend ********** ********* *******.

Beware: ******** **** *******

********* **** *** ******* which ******** ***** *** supported ** ****, **** ********** ***** HD ******** *** ****** cards ** ***** *************. However, ** **** ****** told ** ********* **** him ****** ***** *** not *********, ******* **** high *** ****, ****** freezing, *** ***** ******. Users ****** ******* ***** selected ******** **** ** supported **** ********* *******.

CPU **** **********

** *** *****, *** load ****** ******, **** a *** ** ~*% during *** ***, *** motion ** *** **** spec ******* ** **%+ at *****/**** ****** ** the *** *** *******. 

*** **** ****** **** throughput *** ********* ** night, **** *** *** range ******* **** **** doubling, ***** *** *** end ******* ******* *** CPU *****. ** ********, the **** *** ******* rose **** **** ~*% to ~*%. 

Periodic ****** ******

****** *******, ** *** occasional video ****** **** ******* streams **** ****-****.

*********/***** "******** **"

**** ********, ***** ******** for ******** ******* ** time, **** "****** **" quickly, ********** **** ***** at **** **** ****** speed, ***** *****:

**** ***** ******** ********** of ****** ********, ********* streams ***** ****, ** current **********.

******** *******

******, ******* ************ *****, typically *** * *** seconds, ****** ********** *** displaying **** ***** **** again. **** ** **** in *** ******* *****, happening ***** ****** * few *******.

**** *******

*******, ** ************ *********** very **** ******* ****** testing, **** **** ******* ** ** 30 ******* ** ** much ** * ****** ****** real ****. **** ***** issues, **** *** *** correlate ** ********** ** motion ** *** ******, nor ** *********/******* ******* being ****.

** *** ******* *****, the ****** ** *** right ** ****** * full ****** ****** *** camera ** *** ****, displaying *** ******* ****.

Client *******

****** *******, ** **** experienced ******* ************ (***** below). ***** ******** **** the ****** *** ****, just ******* ******** ******* with ** ***** ********* (shown *****), ** **** attempting ** *** **** resolution ****** *** *** cameras. ***** ******* **** less ********, *** ***** present ** ******* ******** of *** ****** (*.*.*.*), but **** ****** ** older ********.

No ****** ** ***** **

****** *******, ** **** periodically ******* ***** ** the **** ***** ********* for *** ****** ******* live ** ******** *****, but *** ****. *** example, in **** ****, *** 32 ******* **** ******* issue ** ******* **** bandwidth ** ~*** **/*:

No ******* ******/********* ****/***** ******

****** *** *****, ** saw ** ****** **** dropped ******, ******* ******** up, **** ** *********, or ***** ****** ***** ***** indicate *** *** ** overloaded. *** ** ******* recorded ******* *****, **** no **** ** *** timeline, ******* ******** ********* near *** **** *** (320 **/*) *** *-* clients ******* **** ***** and/or ********.

*** *******, *** ******** below ** *****, **** no ****, *** ** lost ***** ** ***** issues ** ********.

Test *****

** ****** **** ***** a **-******-** ** ******* NVR, ***** ****** **** 32 *******. ******* **** set ** *****, ** FPS, **** ******* ***** settings ********* ** ********** level:

  • ***/*** ******:*** *** ****** *** bandwidth ******, ******* **** set ** ***, ~**-***, and ******* ******* ** fluctuate ********.
  • *****/**** ******:** ***** ** ***** the ***'* *** ********** specs (*** **/*), ** set ******* ** *** to ******* **** ****** bitrates. ** ***** * DJ ***** *** **** used ** ****** **** levels ** ******.

** *** **** *** tested ***** ******* **** Manager/Performance *******, **** ** other ***** ********* ************ running ** *** *******. All ******* ******* *** GPU ******* **** ******* prior ** *******.

Versions ****

** ****** ***** *** following ******* ** ********/********: 

  • ********* **-******-**:**.*.** ***** ******
  • ****-**** *******:**.*.*.* ***** ********
  • ****-**** *** ***:**.**.**.**

Comments (32)

I have experienced similar issues to this, albeit not loading the system heavily by any stretch of the means.  iVMS-4200 seems to be a resource hog.  Im sure they are not pouring resources  into development given it's cost of Free ninety nine....

What were the CODECS used vis a vis h.264/h.265, smart or not?

H.264, smart was on/off depending on the bitrate we were looking for. For high load, it was off. For normal loads it was on.

Was the NVR set with RAID enabled on the HDD?

There is a CPU burden with RAID if it's a software implementation, and will effect the bitrate throughput. I come across a few so called high end NVRs supporting RAID.

Update:

I just notice clicking to the link Hikvision spec even say this, from 320Mb/s to 200Mb/s is somewhat major hit on performance a 63% from the original.

I would feel cheated calling RAID which in the world of PCs would give mostly a performance enhancement in throughput! not a decrease..

PC based NVRs would use a raid controller, unless they are brave to use a AMD threadripper CPU with a X399 chipset and a NVME raid configuration!

I wonder if RAID can be enabled to see what differences it makes to playback performances as well as image quality considering we are hitting the max already.

I had to rip out 2 separate systems because of iVMS last year. One 16 channel system and one 6 channels. Put in Exacq and it cleared all my problems. 

Ethan, thank you for performing this series of tests.  It confirmed a few of my own experiences.  I'd love to further explore a couple of quirky asides that I have come up on.  It seems that the more frustrating issues are on multi camera live display;  recording seems to work smoothly.  

1. Could you reproduce a known IVMS lag set up but this time, have the IVMS connect directly to the cameras for live view, bypassing the NVR, and see if that made a difference?  (I posted an experience with this when using it for static live view monitoring on wall mounted displays).  My experience was that it didn't seem to lock up. 

 

2. The 96xx NVR supports multiple HDMI outputs.  Would connecting multiple displays for a wall mount live view perform better than running the live view through a Windows PC running IVMS?

3. I haven't noticed this lag issue running some of the "Turbo" HD-TVI" DVR's and IVMS nearly as severe as the IP NVR's.  These camera feeds are both 1080p, so it's confusing as to why. Any thoughts?

Thanks again

 

Hey John good feedback. Let me respond in order here:

1. Could you reproduce a known IVMS lag set up but this time, have the IVMS connect directly to the cameras for live view, bypassing the NVR, and see if that made a difference? (I posted an experience with this when using it for static live view monitoring on wall mounted displays). My experience was that it didn't seem to lock up.

We're going to try this out quickly. We need to set some things up to get it ready but we'll report back later.

2. The 96xx NVR supports multiple HDMI outputs. Would connecting multiple displays for a wall mount live view perform better than running the live view through a Windows PC running IVMS?

It's possible. We didn't see issues on the local interface of the NVR like we did in iVMS. No latency, freezing, etc.

3. I haven't noticed this lag issue running some of the "Turbo" HD-TVI" DVR's and IVMS nearly as severe as the IP NVR's. These camera feeds are both 1080p, so it's confusing as to why. Any thoughts?

I'm not exactly sure why that is. There are likely hardware differences which impact this. It may be "easier" to encode the TVI cameras due to dedicated hardware than it is to write IP streams. We have some TVI testing coming up and we will look into this then.

 Have you tried the Web App with same setup?

Several times I have found the Web App behaved netter for Live Viewing (Same with re branded Wbox)

Better, not netter  :)  Where is the edit link?

Hi

 

We're experiencing these issues with iVMS 4200 too. One thing is clear the Hikvision NVRs 9600-i8 NVR are mature and perform extremely well. iVMS 4200 is barely adequate. 

Another issue is when you save several views on iVMS 4200.. If you have several views, they seem to load all the saved views at the same time. This at least, slows the best PCs to a crawl and would crash average but Hik-minimum compliance PCs.

Truth to be told, we are looking forward for a better VMS from Hik. The 5200 doesn't seem to be widely available in the US.

It would be good if the usual suspect (Milestone, Exacq, etc...) would support Hik' NVRs in a native fashion . IOW they retain their recording capabilities and the VMS controls and manage these and the monitoring, searching, etc .. The usual VMS stuff :)

Agreed with the occasional client crashes - long standing Hik issue that has improved in recent years - but is still annoying.

Other annoying features is the lack of proportionate zoom on digital zoom and having to resort to native video in order to view corridor mode cameras.

Although the IVMS does support multiple images well and can make use of the auxiliary screens, we've had massively better results using the fantastic DS-6908UDI HDMI decoders. Bear in mind  the units only support 4k images on HALF their inputs with standard HD on the other half.

Slap on the DS1600KI keyboard and you have a great front end with minimal usage of the IVMS.

It would be nice to see IPVM put together this combination of kit and then see what Hik competitors can do to replicate, as it's a strong offering hat most others can't get close to, in terms of hardware. 

 

 

 

Ethan, if you still have that setup it would be nice to know how much GPU is getting used when doing the local viewing.   I am sure you have a copy of GPUz for that.

This will let users know if a GTX1060 is needed or can a lesser card do the task.

I posted my testing results below

 

That's why I opened this article and I couldn't find the most important thing about iVMS testing – testing the GPU load with different types of graphical processors.

Recommendations for the testing team: you should describe your test setup and not just say "it meets specs". Hardware model names, OS version and build, drivers' versions, etc. Open tom's hw guide, overclockers or any other PC hardware testing site. Those guys do it for decades and you don't need to invent the wheel.

So, I opened the latest version of iVMS-4200 on my PC, which is i5-6600/16GB/NVMe 960EVO SSD/Nvidia GTX 1060 6GB, updated drivers, updated windows. GPU-Z 2.5.0.

Opened 64 cameras, most of those are 4 mpx, some mix of 3mpx and 8 mpx as well. CPU load jumped 40% up (decodeprocess.exe), GPU just 10%, RAM usage is close to 500MB, disk usage is close to 0%, network load 7-12 mbps. iVMS became unresponsive but still showed some movement for most of the cameras. Around 10 cameras showed artifacts.

After that, I repeated with just 36 cameras loaded. iVMS became fully functional: no glitches, no artifacts, no frame drops or freezes. All the numbers were the same, except CPU load - decoder process took just 2-10% of the CPU time.

Tried Aux screen preview (25 cams) and main view (25 cams). iVMS worked fine, same numbers as I had with 36 cameras. 

Having the same amount of cameras (36) and switching the view from 36 to 64 makes something to decoder process, so it consumes more CPU time (10% - 20% jump).  Adding cameras to 64cam view makes iVMS extremely slow and not usable. Apparently, there's a problem with a software optimization.

System Configuration – Image – Hardware decoding preferred setting makes no difference to the PC components load.

I have to add some more details: NVR's are 76XX and 77XX series, cameras are distributed thru them so NVR performance is not a bottleneck (up to 10 cams per 7X16 NVR), all the connections are 100 mbps+ to the access level switches.

That's why I opened this article and I couldn't find the most important thing about iVMS testing – testing the GPU load with different types of graphical processors.

Recommendations for the testing team: you should describe your test setup and not just say "it meets specs". Hardware model names, OS version and build, drivers' versions, etc. Open tom's hw guide, overclockers or any other PC hardware testing site.

Dennis, are you sure you read the article when you were looking for the purported “most important thing”? 

 

We tested load using a DS-9632NI-I8 32 channel NVR, fully loaded with 32 cameras. Cameras were set to 1080p, 30 FPS, with varying codec settings depending on throughput level:

Day/low motion: For low motion low bandwidth scenes, cameras were set to VBR, ~28-30Q, and bitrate allowed to fluctuate normally.
Night/high motion: In order to reach the NVR's max throughput specs (320 Mb/s), we set cameras to CBR to provide much higher bitrates. At night a DJ laser was also used to create high levels of motion.
PC CPU load was tested using Windows Task Manager/Performance Monitor, with no other video intensive applications running on the machine. All Windows updates and GPU drivers were updated prior to testing.

We tested using the following version of firmware/software:

Hikvision DS-9632NI-I8: V3.4.93 build 170808
iVMS-4200 Windows: v2.6.2.7 build 20170826
iVMS-4200 Mac OSX: v1.02.05.02

They (you?) spent a week or so, and all we see is just one diagram? Where are all the CPU/GPU load diagrams with different camera sets, testing the client with 4, 8, 16, 32, 64 streams? Switching between views? That's why I pointed to the computer hardware portals. They would post comprehensive research with tons of information - tables and diagrams.

You said:

...you should describe your test setup and not just say "it meets specs". Hardware model names, OS version and build, drivers' versions, etc.

Which was unfair and sloppy on your part.  Right?

 

what is your point, anonymous?

what is your point, anonymous?

That you wrongly claimed that the test setup was only  described as “it meets specs”.

yes, I exaggerated on purpose to show the lack of information

exaggerated on purpose

You think exaggerating on purpose, and let's call it what it is, distorting what we reported, is a positive for you? That's quite a tactic.

Moreover, what is the point of your grandstanding here? We both agree that hardware load is not a problem for the Hikvision NVRs and, to the extent that there are issues, it is software related.

 

This was not meant to be an exhaustive test of every iVMS-4200 configuration on every hardware combination. It was meant to answer a question asked multiple times in discussions, which is whether iVMS-4200 and/or Hik NVRs break down under heavy load. That is why we tested with 32 cameras and bandwidth nearing max specs. 

Moreover, it was not intended to be a Tom's Hardware style test full of benchmark test graphs. It was intended to show the practical effects of high load on the client and NVR, which is why there are many more examples showing those issues than measurements themselves. Measurements won't tell average users what's going to happen when their client hits high load.

ok, makes sense now. But isn't HW decoding an answer to an installer question about load, configuration, etc.? It's easy with CPU - just buy more expensive and get more power. But when it's about GPU and decoding, we need a research to make a decision.

The low GPU suggests that a Quadro P600 might be a better value and it is the design point made for 24/7 usage.

I replaced Hikvision NVRs on 4 projects due to instability of the iVMS-4200. In every case, I had a loaded i7 processor for the client PC and had unacceptable latency/dropouts and freezing.  NVR as stated worked fine.  The support from Hikvision was (including local reps/techs) was severly lacking.  They blamedthe issues on my client PC. Lesson learned!!

ivms-4200 is crap. If you have small CCTV system, I'd go with Milestone XProtect. For big Hik systems, I'd use Hikcentral as Milestone price tag would be comparable to the whole CCTV hardware.

I have had issues with slower I7s but 3ghz and above I have not seen any. Actually even on a 3ghz+ I5 there where no issues

We have several (10+) facilities with the DS-9664-NI NVR with 50-64 cameras pushing 5-8mp each on stream 1 recording to the NVR with audio and and 1.4mp stream 2 for ivms remote viewing. As long as you use an Intel i7 4770 CPU (that is the oldest one that will perform consistently viewing 32 cameras at a time) or any Intel CPU with quicksync and at least HD4000 graphics you shouldnt have a problem with remote viewing 32 cameras at a time on a single 1080p or 4k display. The HDMI 4K out of the NVR is always your best bet for minimal latency and reliability when cycling views. IVMS crashes often on redraws if you change the layout. The new software seems to be a little more consistent, we saw almost no benefit when using nvidia or amd gpu on remote live view systems.

I noticed that my viewing stations with GTX1060 produce better picture and less CPU load compared to the same system with Intel HD 530. I use Hikcentral client, as iVMS just crashes all the time when you try to watch a lot of cameras

Dennis, does using the HIK Central client require purchasing the entire Central package?  If so, world of difference in price......

if you have up to 8 cameras, it's free. I've got 300 cameras license with Dell server and that was just a fraction of the price compared to Milestone or Genetec, no annual fee.

Different software, for sure, but Hikcentral is developing pretty fast.

Read this IPVM report for free.

This article is part of IPVM's 6,817 reports, 914 tests and is only available to members. To get a one-time preview of our work, enter your work email to access the full article.

Already a member? Login here | Join now
Loading Related Reports