H.265 IP Cameras Tested vs H.264By Ethan Ace, Published Jan 18, 2016, 12:00am EST (Research)
For more than 2 years, anticipation has been building for H.265 IP cameras. Finally, H.265 cameras are starting to ship from multiple manufacturers in North America.
The promise is ~50% bandwidth reduction which, if true, would be a huge saver for storage costs as well as making remote monitoring easier and large scale VSaaS more feasible. In the mean time, 'smart codecs' for H.264, have come out, which also offer significant bandwidth savings (e.g., see our Axis Zipstream test).
We tested H.265 cameras from Samsung and Vivotek to understand:
- Bandwidth tradeoff / savings compared to current H.264 camera models widely used (e.g., IP Camera Bandwidth / Storage Shootout).
- CPU Load tradeoffs vs H.264 for viewing
- VMS support for H.265
- Finally, H.265 vs Smart Codecs such as Axis Zipstream
We performed both technical stream analysis and image quality comparisons to understand quality and bandwidth tradeoffs.
Net / net, these two H.265 cameras offered minimal to no bandwidth savings compared to modern / current H.264 cameras. Worse, Axis Zipstream (a smart codec H.264 implementation) performed substantially better than H.265. Equally important, VMS support right now is very limited and we did find a non-trivial level of increase in CPU load when using H.265.
So if you are looking for a buzzword or a tech spec advantage, these H.265 cameras will work well as they are fine as general purpose cameras. However, they offer no real benefits over typical professional H.264 IP cameras and have notable weaknesses in VMS support and CPU load.
Warning: H.265 will develop
Just like H.264 compression efficiency / effectiveness got better over the years (e.g., 2008 H.264 cameras vs today's H.264 cameras), we expect the same for H.265. It is still very early in the lifecycle for H.265 with, for example, one of the largest chip / compression providers, Ambarella still not shipping their H.265 yet (Our understanding is that both camera tested here are using HiSilicon). Despite our disappointing findings with these cameras, it is possible / likely that in the next few years, H.265 will offer significant improvements over the current market offerings.
June 2016 Update: The Samsung camera we tested is their 'original' H.265 offering. Samsung is now releasing new H.265 cameras that now include smart codecs (their marketing term: SmartStream) which they claim significantly reduces bandwidth. We plan to test them and then will update this report.
H.265 Stream Analysis Overviewed
To analyze the H.265 streams from these cameras, we used Elecard StreamEye. Similar to AVInaptic, this software provides detailed quantization information for streams (though AVInaptic does not support H.265/HEVC), with a graphical display of I/B/P frame structure, as well as the ability to view detailed decoding information for each frame. Using this tool we were able to compare the quantization levels of H.265 and H.264 streams from the two cameras, shown in our results below.
We review the basics of Elecard use in this video:
Samsung H.265 Setup / Options
Samsung's SNV-8081R's web interface is mostly unchanged from previous H.264 models, with two key changes:
- First, most obviously, the ability to create an H.265 encoding profile is present, with options much the same as H.264. However, note that H.265 may not be set as primary, viewed directly in the web interface, even with ActiveX plugins, nor may it be used for recording to SD. It may be viewed and locally recording in Samsung's SmartViewer client.
- Second, this camera provides no option to fix compression to a specific level when using VBR. Past Samsung models provided a 1-20 scale of compression levels. Without this option, there is no way to directly compare quality between H.265 and H.264 at a specific quantization.
We briefly review Samsung configuration in this video:
Vivotek H.265 Setup / Options
Similar to Samsung, the bulk of the Vivotek IP9171-HP's web interface is unchanged from previous models, aside from new options for H.265 (simply located beneath H.264 and MJPEG in CODEC settings) and Smart Stream II. Similar to Axis Zipstream, Smart Stream II varies I frame interval and dynamically adjusts compression on moving objects and/or fixed ROIs. We will cover Smart Stream II performance in more detail in a separate report.
Note that Vivotek includes options for VBR (with fixed compression level and a cap) and CBR, unlike Samsung, which allowed direct comparison of H.265/H.264 at specific settings.
We briefly review the Vivotek web interface in this video:
Compared to H.264 Mainstream Cameras
In our tests, H.265 bitrates were average or higher when compared to typical H.265 models. For example, the chart below shows Vivotek's 3MP IP9171-HP (both H.265 and H.264) compared to Hikvision and Axis, with all cameras set to ~27 quantization. Vivotek's H.265 stream was lower than Hikvision H.264 only during the day, but higher than both Axis and Hikvision at night.
Measurements for Samsung cameras were taken using ~27 Q, as this is what their VBR defaulted to, with no way to adjust. Again, H.265 bitrates are higher than competitive H.264 bitrates. Note also that the Bosch starlight 5MP is ~5.5MP vs. Samsung's 5MP, but still lower bitrate, especially at night.
Compared to Axis Zipstream
H.265 had far less bitrate reductions than their Smart Stream II smart CODEC, which varies I frame interval from 1-10 seconds depending on motion, and dynamically adjust quantization on moving objects. With this turned on, the Vivotek camera's daytime bitrate was lower than even the Axis Q1615 with Zipstream set to max settings (dynamic GOP 300, Zipstream "high"). However, bitrate still spiked substantially at night, over 8 Mb/s, where Zipstream remained low, about 2.4 Mb/s.
Note that Axis Zipstream numbers below are extrapolated from 1080p measurements, since no 3MP Axis camera with Zipstream was available.
Note that we plan to test Vivotek's Smart Stream II performance in more detail in a future report.
CPU Load Tradeoffs
For low bitrate streams (~4 Mb/s and under), H.265 decoding was slightly more processor intensive than H.264, both on low and high end PCs. This was true when viewing both a single stream as well as multiple simultaneous streams. These measurements were taken viewing streams in VLC, though client softwares showed similar trends:
However, when bitrates spiked at night in either of the H.265 cameras, CPU usage jumped significantly, with H.265 more than doubling H.264 processor usage in a high end PC with dedicated video card. Users should beware of these difference in performance, as client machines which work fine for H.264 may no longer view H.265 properly.
Specs of PCs used:
- Low End PC: AMD Quad Core 3.7 GHz, 8 GB RAM, Integrated Radeon Graphics
- High End PC: Intel Dual Quad Core i7 2.4 GHz, 16 GB RAM, 2GB GeForce GT 650M Graphics
As of publication, H.265 support from third-party VMSes is rare. We were able to integrate H.265 with Genetec but only the Vivotek model. Adding the Samsung 8081R to the VMS provides only H.264/MJPEG options. Genetec users may select the H.265 for live, recording, remote, etc., just as other streams, seen below.
Milestone and Exacq both inform us they are working on adding support, but no dates were given.
Vivotek's own VMS, VAST, support H.265 viewing and recording of their own cameras. Samsung's recorders/VMS software do not support H.265 currently, but their SmartViewer client supports live viewing and local clip recording.
We found few, if any visible differences between H.265 and H.264 using similar quantization levels. For example, this image shows the two CODECs side by side (click image for full size version):
H.265 is on the right, H.264 on the left, both ~Q28 in this comparison. Visual differences are minimal, and only visible upon fine inspection, despite H.265 bitrates being 30% lower than H.264 in this scene.
In some cases, we found the H.265 image contrast (seen below) was slightly different than H.264, though the practical effects are small.
We expect H.265 performance and impact to improve over the next few years, though it remains to be seen how fast it improves. In the meantime, H.264, especially with its broad support and lower CPU utilization, continues to be the easy and right choice for normal deployments. Indeed, if and as smart H.264 advances, H.265 may not make much sense for many years to come (or even ever).
13 reports cite this report:
Back to Top