Myths Vs Facts: comparison between 10GiGE Vision and CoaXPress v2.0
Recently, we receive more and more questions regarding the comparison between the 10GiGE Vision interface vs the CoaXPress v2.0, so we have decided to publish our opinion.
There are many white papers and articles available that are comparing these two protocols. Most of them, however, are biased towards the products offered by the company writing said articles. A company that promotes
10GiGE cameras will naturally try to prove the 10G to be superior to the CoaXPress v2.0 and vice versa. Since KAYA Instruments supports all possible high-end video protocols, we do not have preferred protocol and therefore believe our opinion to be very reliable.
Historically, the GiGE Vision interface was (and still is, in many applications) a great success. An actual bandwidth of approximately 800Mbit/sec and the lack of need for a dedicated frame grabber card in the PC made this
protocol “the king of protocols” for a long time. The ability to work without a frame grabber in the PC was the major advantage of the GiGE protocol over the others. Recently, however, sensor resolutions became larger, frame rates increased, and customers are becoming more performance demanding.
To answer these needs, three new protocols emerged: 10GiGE Vision, CoaXPress and CLHS (which we will not cover in this article).
Naturally, the industry accepted 10GiGE Vision as the successor of the GiGE Vision protocol, but here’s the thing: what was correct for GIGE Vision did not work anymore for 10GiGE Vision: the bandwidth of the 10GiGE Vision has increased 10 times over the GiGE Vision protocol whilst CPU performances didn’t improve that much. This meant that CPUs could not handle the 10Gbps traffic by themselves and started loosing frames. Using a smart NIC (Networking Interface Card) or a frame grabber with 10GiGe Vision protocol offloading capabilities solved the problem but canceled the largest advantage of the GiGE Vision protocol.
In our opinion, the only advantage of 10GiGE over CoaXPress v2.0 is ability of 10GiGE Vision to transfer video data over standard Ethernet infrastructure including Ethernet switches (To support multicasts) and cables that are already installed inside the walls. That being said, most application transmit uncompressed video streams peer-to-peer or compressed streams when multicasts are required. As for the existing cable infrastructures, the 10GiGe vision requires a cable upgrade from the one used in GiGE vision systems.
There is an additional minor cable length advantage in favor of the 10GiGE over CoaXPress that allows installing the camera far away from the PC, but most of machine vision application operate at cable lengths shorter than 10 meters. Once longer distance is required, range extender over fiber may be used for other protocols. Such extenders are available today for a reasonable price for any protocol, such as Camera Link, CoaXPress, USB3 etc.
As for the disadvantages of 10GiGE over CoaXPress, the list is much longer:
1. 10GiGE Vision protocol does not operate in real time. The frame latency is not deterministic and the frames can arrive with a significant time jitter. In some conditions, especially with switches, one can receive frames
in opposite order or even loss them (see next clause).
2. The 10GiGE Vision protocol is unreliable. The Ethernet infrastructure allows packet loss and the CPU can start dropping packets when it is overloaded. 10GiGe Vision protocol does define a packet retransmit option
but it is not feasible due to long turnaround time compared to the bandwidth. This will require huge memory buffer on camera and will affect the frame latency and jitter even more.
3. The 10GiGE Protocol is very power hungry, especially when using RJ45 cables. This makes camera very hot. High temperature of the camera impacts picture quality exponentially: the higher the temperature of the
sensor, the more noise you get, exponentially. Some 10GiGE vision vendors rate maximal environmental temperature of only 30°C!
4. The protocol is very heavy for streams. All the data de-encapsulation is done in the CPU, making it busy with unpacking the stream data (UDP packets) instead of doing what it is intended to do for the application.
There are several ways to resolve this, e.g. by using dedicated NICs, such as Myracom. Such NICs can slightly reduce CPU utilization but lack video processing features usually available in Frame Grabbers. If you have
to pay for a special card either way, what’s the point of going with the 10 GiGE over CoaXPress?
5. It is impossible to connect multiple cameras to the same PC due to 4 (see above. CPU will be overloaded).
6. Real time triggers cannot be sent over the 10GiGE protocol. There are possible extensions to the protocol, such as PTP, but they requires special care and not are natively integrated.
7. Current prices of Frame Grabbers are lower or comparable with prices of 10GiGE NICS, especially for multiple channel support.
8. CoaXPress is more widely adopted in the industry compare to 10GiGE Vision based products. Most camera vendors started with 10GiGE protocol-based cameras now move to CoaXPress v2.0
9. Image processing pipeline features do not exist when using NICs, (as NICs do not care about video streams).
10.While cable prices are pretty much the same for Ethernet (CAT6 or CAT7) and coaxial cables for CoaXPress, the coaxial cable is much easier to handle and assemble in field, if required.
Is everything perfect with CoaXPress v2.0? Of course not. One disadvantage, for example, is the protocol overhead of the interfaces with CoaXPress at about 22% vs only 4% for the 10GiGE. This is mainly due to 8b/10b encoding of the CoaXPress link and means that 12.5Gbps link transfers are “just” 9.8Gbps of actual video data. Still, this is slightly higher or equal to the approximate 9.6Gbps of the 10GiGE Vision protocol.
For any questions, suggestions, or comments, please contact us at https://kayacameras.com/contact-us/
Tel: +1 (786) 296-2721
KAYA Instruments Team