WELCOME TO OUR BLOG

We're sharing knowledge in the areas which fascinate us the most
click

Why GMSL3?

By EyeCloud October 28th, 2026 2 views

Why GMSL3?

If you work with cameras — bringing up modules, running road tests, building vision systems — you have watched the numbers climb year after year: 2 MP sensors giving way to 8 MP, 30 fps giving way to 45 or 60, HDR becoming the default. The cable between camera and processor has to keep up, and GMSL, the single-coax link that carries video, control and even power through one thin cable, has grown through generations to stay ahead of the sensors it serves. GMSL3 is the generation built for today’s high-resolution, high-frame-rate cameras. In this post we look at why GMSL3 appeared, how it differs from GMSL2, what it actually buys you, and how our ECFG-R2-96792 development platform puts a 12 Gbps GMSL3 link on your bench.

1. GMSL2 vs GMSL3: The Differences That Matter

Feature GMSL2 GMSL3
Forward-link data rate Up to 6 Gbps Up to 12 Gbps
Comfortable payloads 2–4 MP sensors, 4K30-class video 8 MP-class sensors at full frame rate, 4K90-class video, multi-sensor aggregation
Video transport Pixel mode: the link decomposes video into pixels plus control data Adds tunnel mode: the native MIPI CSI-2 stream is transported transparently
Backward compatibility — GMSL3 devices commonly also run GMSL2 rates (12/6/3 Gbps)
Link robustness Error detection on control data FEC and CRC with equalization engineered for 12 Gbps operation
Cable and connector Single coax (or STP), FAKRA, POC Unchanged — same single coax, same connector philosophy
Typical chip pairing MAX9295 / MAX96717 + MAX9296 MAX96793 + MAX96792A

The architectural change is tunnel mode. In GMSL2’s pixel mode, the deserializer unpacks the incoming stream into pixels and control data and reassembles them for the host interface. GMSL3’s tunnel mode instead wraps the camera’s native MIPI CSI-2 stream and carries it through the link untouched: embedded metadata, HDR sub-frame structure and sensor-specific payload formats all arrive exactly as the sensor produced them. For engineering teams this is more than convenience — you develop against the sensor’s true output, not a link-processed reconstruction of it.

GMSL3 silicon is also pragmatic. A GMSL3 serializer such as the MAX96793 operates at 12 Gbps (GMSL3) as well as at 6 Gbps and 3 Gbps (GMSL2) rates, and the MAX96792A deserializer accepts both GMSL3 and GMSL2 links. A GMSL3-capable platform therefore covers both generations: your legacy GMSL2 modules and your new GMSL3 designs run on the same bench.

2. What GMSL3 Is Good At

  • Headroom for modern sensors. 8 MP-class HDR sensors stream at full frame rate with room to spare — no forced compromise between resolution, frame rate and exposure scheme.
  • Raw video stays raw. There is no need to compress or thin out data to squeeze through the link; your ISP and algorithms work from the sensor’s true output.
  • Tunnel-mode fidelity. Native CSI-2 transport keeps embedded metadata intact end to end — cleaner input for bring-up, tuning and datasets.
  • No re-wiring. The single-coax, FAKRA, power-over-coax discipline carries over from GMSL2, so harness architecture and mechanical design do not change.
  • One platform, two generations. GMSL3 devices speaking GMSL2 rates mean a GMSL3-ready bench also covers your installed GMSL2 modules.

3. Meet the ECFG-R2-96792

ECFG-R2 is EyeCloud’s two-channel open platform for camera module development and road testing, built on the NVIDIA Jetson Orin Nano. The -96792 variant carries the MAX96792A GMSL 3 deserializer (up to 12.0 Gbps) onboard, making it the model to pick when your camera modules speak GMSL3.

EyeCloud ECFG-R2-96792 GMSL3 camera development platform

As part of the ECFG series, it is built around three ideas: low-code bring-up of your own camera modules, high-bandwidth data transfer, and multi-brand deserializer adaptation (GMSL 2/3 and FPD-Link). You keep the module design in your hands; the platform lights it up.

Key specifications:

  • Link: onboard MAX96792A GMSL 3 deserializer, up to 12.0 Gbps; 2 × high-speed MINI FAKRA connectors with 12 V @ 500 mA POC per port over automotive-grade coax connectivity
  • Compute: NVIDIA Jetson Orin Nano — 6 × Arm Cortex-A78AE v8.2 64-bit CPU @ 1.7 GHz, 4 GB 64-bit LPDDR5 (51 GB/s)
  • Storage: 128 GB M.2 NVMe SSD for onboard data and road-test recording
  • Network: 1 × RJ-45 10/100/1000 Mbps + 1 × RJ-45 10/100/1000/10000 Mbps
  • I/O: 2 × USB 3.2 Gen2 Type-A, 1 × USB 3.2 Gen1 Type-C, 2 × USB 2.0 Type-A, 1 × USB 2.0 Micro-B (debug), 1 × HDMI 1.4
  • Power: USB Type-C PD3.0 (DC 20 V @ 3.25 A); 0–45 °C operating range
  • Software: Ubuntu 22.04 with JetPack 6.2.1, SDK development kit and UGrab interactive software

The development workflow is deliberately low-code: initialize your sensor and SerDes chips from configurable scripts instead of writing a driver stack from scratch, read and write device registers over I2C passthrough, control frame-sync signals, switch and monitor POC power per port, and capture raw data from both linear and HDR streams via virtual channels. UGrab, the interactive software shipped with the platform, gives you a visual workflow for all of it — the same low-code approach we described in this UGrab deep dive.

On the GMSL3 side, ECFG-R2 has been validated with the MAX96793 camera-side GMSL 3 serializer, and the validated sensor list includes Sony’s 8 MP-class IMX728 and IMX735, the HDR IMX490, onsemi AR0830 / AR0820 / AR0234 and SmartSens SC8238. Some customer-provided camera models are not listed — contact us for the latest compatibility list.

4. Where Teams Use It

  • Camera module bring-up: light up your own GMSL3 modules with low code — sensor and serializer initialization, register access, sync control and first-frame capture from one workflow.
  • Road-test data recording: record raw camera streams onto the onboard 128 GB NVMe SSD during vehicle tests, then offload quickly over the 10G-capable Ethernet port.
  • Burn-in testing: long-duration aging under continuous operation before design freeze.
  • Reliability testing: robustness and stability validation over long-run operation.
  • MTF/SFR evaluation: quantitative image-quality evaluation of your modules as part of lens and tuning verification.
  • SerDes link signal-quality monitoring: watch link integrity on the 12 Gbps GMSL3 line during development and road tests, and catch marginal links before they become field issues.

5. One Platform, Four Links

Not on GMSL3 yet? The same ECFG-R2 base is available with three other onboard deserializers: ECFG-R2-9296 (MAX9296, GMSL 2, default), ECFG-R2-954 (DS90UB954-Q1, FPD-Link III) and ECFG-R2-9702 (DS90UB9702-Q1, FPD-Link IV). Same compute, same FAKRA inputs, same software — you simply pick the link your camera speaks. All four options are listed on the ECFG-R2 product page.

6. Put a 12 Gbps GMSL3 Link on Your Bench

If you are designing with 8 MP-class sensors, chasing higher frame rates or capturing HDR raw streams, the 6 Gbps ceiling is exactly the constraint GMSL3 removes — and the ECFG-R2-96792 is the fastest way to start working against it: a complete, Jetson-powered development and road-test platform with the MAX96792A GMSL 3 deserializer onboard, ready for your own camera modules.

  • For full specifications, the validated sensor and SerDes list, and all four model options, visit the ECFG-R2 product page.
  • Ready to order, or want to confirm that the -96792 variant matches your sensor and serializer? Contact us for pricing, availability and a UGrab software demo — our team will help you match the platform to your cameras and workloads.
Differences Between R2 and P53: Which One Should You Choose?, NVIDIA Jetson Orin EdgeAI Boxes, GMSL & FPD-Link Cameras,Frame Grabbers & Camera Test Tools | EyeCloud AI
Previous
Differences Between R2 and P53: Which One Should You Choose?
Read More
Leave a message
FirstName*
LastName*
Email*
Message*
Company Name