Tvnet Driver regression for Jetpack 4.4 vs 4.6?

We are using the tegra_vnet.c driver (source files attached
Makefile (159 Bytes)
tegra_vnet.c (26.4 KB)
tegra_vnet.h (9.2 KB)
) for our PCIe DMA ethernet interface.

When running this driver, we tested that the data rate for JetPack 4.4 was around 6 Gbps. But when we upgraded to Jetpack 4.6 (on the same hardware without any changes), we dropped to around 2 Gbps.

Are we using the most updated version of this tvnet driver for 4.6 (and beyond)? We plan on upgrading to 5.1 sometime in the future, but not until we have the same data rate that we experience with JetPack 4.4.

What could be causing this data rate regression for v4.6?

No idea. Maybe try to pull the clock up first and see if it makes difference.

Thanks for the response. What do you mean ‘pull the clock’? How is this done?

For our tests, we are using Iperf3 for our test code. See below:

THRESHOLD=400000 #in kilobytes

function verify_iperf_output {
NUM_LINES=cat $1 | wc -l
if [ $NUM_LINES -lt 19 ]
echo “ERROR: iperf did not create enough lines - only $NUM_LINES”
exit 1

SEND_BANDWIDTH=cat $1 | tail -4 | head -1 | tr -s ' '| cut -d' ' -f7
RECEIVE_BANDWIDTH=cat $1 | tail -3 | head -1 | tr -s ' '| cut -d' ' -f7
if [ $SEND_BANDWIDTH -lt $2 ]
echo “ERROR: Send bandwidth did not meet threshold of $2”
exit 1

if [ $RECEIVE_BANDWIDTH -lt $2 ]
echo “ERROR: Receive bandwidth did not meet threshold of $2”
exit 1


iperf3 -c -f K | tee $TEMPFILE
if [ $? -ne 0 ];then
echo “ERROR communicating with”
verify_iperf_output $TEMPFILE $THRESHOLD

Run with jetson_clocks.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.