As per Jetson Tx2 TRM, in nvcsi controller, it supports the three 4 lanes configurations.
Whether nvcsi controller can support one 12 lanes configurations or not?
Please provide your input on this.
In one device tree configurations, in vi controller device node, nvcsi controller device node, specific device node, bus width is configured as “bus-width = 8”, i.e. it supports one 8 lanes configurations.
How nvcsi controller supports this configuration?
there’s HDMI to CSI bridge driver that use 8-lanes, you could check kernel driver tc358840.c for more details.
about your 12-lane configuration. may i know what’s your hardware device, and what’s the use-case?
thanks
We need to give CSI output to Jetson Tx2 board. This is for Ultra HDMI to CSI conversion.
In TC358840 chip datasheet, it is given as:
Dual links CSI-2 (CSI0 and CSI1), each link supports 4 data lanes @ 1 Gbps/data lane
CSI0 carries the left half data of HDMI Rx video stream and CSI1 carries the right one at the
default configuration.
Left or right data can be assigned/programmed to either CSI-2 Tx link
I am looking into how this works in CSI controller. What kind of configuration we have to do for this?
Default configuration is enough or need to configure in different way to perform this??
Please provide your inputs on this.
Whether we need to look for MIPI specification or nvcsi controller details to get the idea on this.
Need the help on this.
for the 8-lane use-case, there’s configuration in the device tree to specify 8-lanes and VI kernel driver will choose gang mode to handle the streaming.
still curious about more details for your use-case.
for example, is this single chip HDMI-CSI bridge? may i know what’s the resolution?
thanks
each CSI lane of TX2 can support up-to 2.5Gbps,
consider your use-case, 3840x2160@60Hz should be 3840x2160x16(2byte)x60= 7.9Gbps
may i know what’s the “units” of your 12G streaming?
thanks
The input from FPGA to each CSI lane of Tx2 is 1Gbps. This limitation is due to FPGA device. Since this FPGA device comes under our product’s cost. Other FPGA devices (high data bandwidth supports ~1.5Gbps) are not in our product’s cost.
Use case is:
12-G SDI requires - 12Gbps
3840x2160p@60fps with 24 bits per pixel - 3840x2160x60x24=11.89 Gbps
We are looking for this FPGA device. Thanks for the inputs.
Why do you want to convert the 10bit 422 video to 8bit 444?
That just increases the data rate from 20bpp to 24bpp.
We have worked with SDI using PCIe. The 4x3G capture device is right on the edge of what Tegra PCIe can do and FPGA’s in reasonable costs. 12G is just a question of adding a gearbox IC. If you are interested, contact me.
For CSI you have to use either a faster I/O to fit into 8lane CSI, or to use software postprocessing to merge the three CSI inputs into one. AFAIK the nvidia drivers can not share CSI buffers (we would like that as well, for synchronized camera capture).
There is no 12 lane decoding mode (Parker TRM is very clear on this).
FPGA outputs 12-G SDI in the form of 12 lanes (three CSI ports) of MIPI CSI outputs each capable of 1Gbps line rate.
Can Nvcsi Controller be configured in 1x8 and 1x4 lanes and can be applied to merge in software application or software post-processing to merge the three CSI inputs into one?
If FPGA interfaces to NVIDIA TX2 in PCIe the data bandwidth issue is resolved. However, there is only memory coming up with PCIe unlike VI for CSI-2.