IP Camera Bandwidth / Storage Shootout
Bandwidth consumption can vary tremendously, even with the same resolution, frame rate, compression and scene, by 50% to 80% in our tests. This is because cameras process video differently (e.g., gain control applied, noise reduction technology, etc.).
We tested 16 cameras head to head to understand who is the best and worst at minimizing bandwidth consumption / storage costs.
In the 1MP category, we set these 6 cameras to 720p / 10fps:
In the 2MP category, we set these 4 cameras to 1080p / 10fps:
Additionally, we tested 6 integrated IR cameras to match up against these.
- Avigilon 2.0W-H3-B02
- Axis M1144-L
- Dahua IPC-HFW2100N
- GeoVision GV-UBX1301
- Hikvision DS-2CD2032-I
- Q-See QCN7001B
The Wrong Way
The simple, naive and wrong way is to put cameras side by side, measuring bandwidth at the same resolution and frame rate.
Why? The default compression settings of cameras vary widely by manufacturers. Though compression is often overlooked, every H.264 camera must choose a Q or quantization level (from 0 to 51). The lower number chosen, the higher the bandwidth, and the greater the compression (to learn more, see our IP Camera Manufacturer Compression Comparison). This does not make one camera 'better' or 'worse' and can be adjusted by the user.
Indeed, in IPVM testing, default compression variances routinely accounted for a massive 50%+ difference in bandwidth consumption.
To truly and fairly measure bandwidth, compression levels must be normalized, that is, they must be set the same. Using a stream analyzer, IPVM determined the actual quantization levels and adjusted the appropriate scales so the quantization levels all would be the same for our tests (using Q=28).
Test Questions and Answers
This IPVM test help identified:
- The 1 cameras that clearly had the lowest bandwidth consumption
- The 1 camera that clearly had the highest bandwidth consumption
- The 4 camera manufacturers whose default setting make them look far worse than they are
- The tradeoff of non IR and IR cameras and how it varied by low light levels
- The impact of VMD
- Lowest bandwidth cameras overall - Bosch Starlight 720p model stood out, especially in low motion and low light scenes where savings was more than 50% over the group.
- Highest bandwidth camera overall - Sony Gen 6 models stood out, especially in low motion and low light scenes where consumption was more than 50% over the group.
- Default settings skewed - Avigilon, Bosch, Samsung and Sony's default compression levels were much lower than Axis, making them appear to be relatively worse than Axis, until normalization occurred. Beware naive bandwidth comparisons.
- IR vs Non-IR variance: At ~2 lux (dim / moderately dark), the non-IR cameras consumed less bandwidth than the IR ones. However, at 0.1 lux (real dark), the IR cameras consumed less as the non-IR ones typically spiked up with high gain control levels.
- 1080p spikes more than 720p - In our tests, 1080p bandwidth was more volatile, with bigger spikes than 720p in low light and high motion scenes.
- The strongest VMD - Axis VMD2.1 (not standard but a free add-on) and Mobotix MxActivitySensor where the two strongest VMDs by far.
- 720p cameras averaged about 500 Kb/s in full light, ~160 lux. The Axis M1114 and Sony SNC-VB600 were both much higher than average, at 700 Kb/s and 900 Kb/s, respectively.
- Three of four 1080p cameras tested consumed between 1-1.5 Mb/s. Only the Sony SNC-VB630 was notably higher, over 3 Mb/s 2-3x others.
- In low light, ~2 lux, average 720p bandwidth was about 1.3 Mb/s, with no outlier either high or low.
- 1080p cameras at ~2 lux average consistently 4.5-4.6 Mb/s, with no drastic highs or lows.
- In the dark, ~0.1 lux, 720p cameras average about 3.7 Mb/s. The only true outlier in this case was the Bosch NBN-733V, which ran at 1-3 Mb/s lower bitrate than all other cameras.
- 1080p cameras varied from about 7.8-13 Mb/s. Note that in this scene the Avigilon H3 3MP camera produced no usable images, nearly black, accounting for its very low bitrate.
- In an outdoor roadway scene with high percentage of motion, 720p bitrate averaged around 1.4 Mb/s. No cameras were extremely high or low in this case, with a swing of just over 1 Mb/s.
- 1080p cameras averaged about 4.3 Mb/s in this scene. The Bosch NBN-932V and Samsung SNB-6004 were notably lower, at 3.0 and 2.6 Mb/s, than the Avigilon H3 3MP and Sony VB630 (6.3 and 5.3 Mb/s respectively).
- 720p integrated IR cameras, with IR on in a dark (0.1 lux) room averaged 1.6 Mb/s, with Geovision and Q-See running 700-900 Kb/s higher than the Axis M1144-L. Average bandwidth was less than half the average of day/night cameras at this light level (3.7 Mb/s).
- Average bandwidth of 1080p integrated IR cameras was about 4 Mb/s, though Avigilon's 2.0W-H3-B02 ran much higher than Dahua and Hikvision (6.3 Mb/s vs. 2.6-3). Cameras averaged ~30% lower bitrate than 1080p day/night cameras in this same scene.
Here are our key points from this test, with all cameras standardized to a framerate of 10 and quantization averaging between 27 and 30:
Full Light Performance
Our first scene was our typical indoor conference room, at about 160 lux, with a field of view seen in this image:
The comparisons below shows image quality of all cameras in full light. All cameras were fairly close in image quality, with small variations in color, saturation, and sharpness, but no drastic differences in image quality.
Average bandwidth of the 720p cameras in this test was about 500 Kb/s (0.5 Mb/s). The Sony SNC-VB600 was about 80% above average, at 900 Kb/s. The Axis P3354 and Q1604, Bosch NBN-733V, and Samsung SNB-5004, all 400 Kb/s or below, have historically had the lowest full light bandwidth consumption in our tests.
The average of our four 1080p cameras was about 1.5 Mb/s, with a single outlier, again the sixth generation Sony, SNC-VB630, with bitrate over 3 Mb/s.
Low Light Performance
We lowered the lights to ~2 lux to check differences in bandwidth as gain and noise increased, seen here:
At this light level, bitrates were quite consistent, ranging only between 1.0 and 1.6 Mb/s, with an average of about 1.3.
The 1080p cameras tested were also much more consistent at 2 lux, with no drastic swings high or low, and an average around 4.5-4.6 Mb/s.
In our final indoor shot, we lowered lights to about 0.1 lux to test bandwidth performance in a truly dark scene:
The comparison below shows all cameras at this light level to display comparative performance, brightness and noise in the scene, first 720p:
720p cameras in this scene averaged about 3.7 Mb/s, with the only true outlier being the NBN-733V, which averaged 1.6 Mb/s, 1-3 Mb/s lower than other cameras. Note that the Axis M1114 was relatively low in this scene, though delivered no usable image, nearly black.
1080p cameras varied from 7.8 to 13 Mb/s in this scene. The Avigilon 3.0W-H3-B2 was nearly black at this light level, accounting for its much lower bitrate than other cameras, which all produced usable images.
Outdoor Roadway Performance
To measure bandwidth in a high-motion scene, we set up the cameras next to a busy roadway, seen in this image:
Here, 720p cameras averaged about 1.4 Mb/s. There were no drastic high or lows in this case, with a swing of just over 1 Mb/s between high (Axis Q1604) and low (Sony SNC-VB600).
The 1080p/3MP cameras tested averaged about 4.3 Mb/s in this scene. The Bosch NBN-932V and Samsung SNB-6004 were notably lower than Avigilon and Sony, using 40-50% of the bandwidth.
Integrated IR Performance
We tested six integrated IR cameras at 720p and 1080p resolutions to see how much average bandwidth varied among them and versus day/night cameras in the same scene. This image shows the FOV in the dark, with IR on:
And this image compares all 720p cameras, showing the relative IR power and noise levels of each:
At 720p, bandwidth averaged about 1.6 Mb/s. The Axis M1144-L was substantially lower than other cameras, at 1.1 Mb/s, versus 1.8-2 for Geovision and Q-See. Notably, this average is much lower than the day/night cameras in our tests, which averaged over 3 Mb/s, with only the Bosch NBN-733V on par with or lower than 720p IR counterparts.
And the 1080p cameras:
The Avigilon 2.0W-H3-BO1, at 6.3 Mb/s was substantially higher than Dahua and Hikvision 1080p integrated IR cameras tested, especially interesting given their superior IR performance. As with 720p, 1080p integrated IR bandwidth is lower in all cases, with the exception of the Avigilon 3.0W-H3-B2. Its bitrate was only 1.4 Mb/s, though it delivered no usable images.
Default vs. Standardized
A few cameras defaulted to much different quantization levels than what we standardized to in this test. The Bosch NBN-932V and Samsung SNB-5004 were the worst examples of this, with reductions of ~80% when changing from default settings to our standardized quantization levels of 28-29. The differences for these and other cameras are shown in the chart below.
These figures further drive home the point that non-standardized bandwidth testing is likely worthless. Unboxing cameras from manufacturers below and immediately testing without standardization will drastically skew results, as bandwidth and storage needs will be much greater than some competitors, such as Axis, who default to higher quantization levels.
These are the settings used for each camera in order to achieve an average quantization of 28:
Readers may learn more about manufacturers' varying quantization and bandwidth levels in our IP camera compression report.
Motion Detection Performance Impact
Since the majority of surveillance systems record only on motion, VMD performance, specifically false positive rejection, may have a huge impact on storage needs. In our tests, we found that advanced motion detection testing that packages such as Axis VMD 2.0 and Mobotix MxActivitySensor reduced false positives from unwanted objects, as well as noise caused by increased gain, in nearly all scenes tested.
Users should also see our tests of standard VMD from seven camera manufacturers for more performance information. Performance of each manufacturer varied depending on the scene, so these reports contain our recommendations for each: