The speed of 5G dongle

Hi

The performance of download speed is still around 275Mbps.
With jetson_clocks enabled and auto suspend disabled.

Plus, does R28.2 support nvpmodel?

see log.

Thank you for any advice,

info_dmesg_debug.log (1.1 KB)

Hi,
Please check the falcon clock:

root@nvidia-desktop:/sys/kernel/debug/clk/xusb_falcon_src# cat clk_rate
336000000

The device node is on r32.3.1. The path could be different on r28. Please grep falcon and see if you can get it and check the clock.

Hi DaneLLL,

Thank you for your prompt support.

Tested the 5G module with L4t R32.3.1 also got the same performance.

And clk_rate is 336000000 on L4T R32.3.1

Thanks,

Hi,
Please also share sudo tegrastats and full dmesg for reference.

Hi HuiW,

We also want to know if it is possible to use other usb device to test the speed on both your host and TX1 too.

For example, some ethernet dongle and see if the iperf result also has large gap between TX1 and host.
Currently, we don’t think this issue is due to usb bandwidth.

You could also try the 5G dongle on ubuntu host as a benchmark.

Hi DaneLLL,

Here is the data of tegrastats and full dmesg.
https://drive.google.com/drive/folders/14uponxnTfVFGRTAQ-NwvYwoYoMHARE2j?usp=sharing

Hi Wayne,

Tested a USB3 drive on TX1.

platform L4T version *USB2.0 port read *USB2.0 port write USB 3.0 drive read USB 3.0 drive write
TX1 R32.3.1 34.3MB 11.3MB 88.8 MB 11.3MB

Thank you,

Hi HuiW,

The USB3 driver test is on custom carrier board, right?
Can you do the test on NV TX1 devkit?

Hi,
From tegrastats shown in IMG_0816.JPG, it looks like the performance is capped by CPU capability. The four CPUs are at max clocks with > 90% loading. Probably there are significant CPU executions in the speed test. Not sure if this is possible, but if there is a mode in the speed test which can take less CPU usage, maybe we can try the mode.

Hi,
We have set up r32.3.1/TX1 + USB3 dongle

  1. Connect r32.3.1/TX1 to www.speedtest.net and get the result as attachment.

    tegrastats:
RAM 1961/3962MB (lfb 183x4MB) SWAP 0/1981MB (cached 0MB) IRAM 0/252kB(lfb 252kB) CPU [100%@1734,90%@1734,90%@1734,91%@1734] EMC_FREQ 18%@1600 GR3D_FREQ 99%@998 APE 25 PLL@42C CPU@44C Tboard@41C Tdiode@45.5C GPU@40.5C AO@50C thermal@42.25C PMIC@100C VDD_IN 8581/7086 VDD_GPU 1631/1060 VDD_CPU 3572/2826
RAM 1961/3962MB (lfb 183x4MB) SWAP 0/1981MB (cached 0MB) IRAM 0/252kB(lfb 252kB) CPU [100%@1734,89%@1734,90%@1734,85%@1734] EMC_FREQ 18%@1600 GR3D_FREQ 99%@998 APE 25 PLL@41.5C CPU@43.5C Tboard@41C Tdiode@45.5C GPU@41C AO@49.5C thermal@42.5C PMIC@100C VDD_IN 8426/7127 VDD_GPU 1553/1075 VDD_CPU 3417/2844
RAM 1962/3962MB (lfb 183x4MB) SWAP 0/1981MB (cached 0MB) IRAM 0/252kB(lfb 252kB) CPU [100%@1734,96%@1734,93%@1734,93%@1734] EMC_FREQ 18%@1600 GR3D_FREQ 86%@998 APE 25 PLL@42.5C CPU@44C Tboard@41C Tdiode@45.75C GPU@40.5C AO@49.5C thermal@42.5C PMIC@100C VDD_IN 8391/7164 VDD_GPU 1476/1087 VDD_CPU 3494/2863
RAM 1963/3962MB (lfb 183x4MB) SWAP 0/1981MB (cached 0MB) IRAM 0/252kB(lfb 252kB) CPU [98%@1734,89%@1734,89%@1734,89%@1734] EMC_FREQ 18%@1600 GR3D_FREQ 9%@998 APE 25 PLL@42.5C CPU@44.5C Tboard@41C Tdiode@45.75C GPU@41.5C AO@50C thermal@42.5C PMIC@100C VDD_IN 8387/7199 VDD_GPU 1320/1094 VDD_CPU 3494/2881
  1. Follow https://vitux.com/test-your-internet-speed-through-ubuntu-command-line
    The result is
Retrieving speedtest.net configuration...
Testing from HiNet (59.124.78.18)...
Retrieving speedtest.net server list...
Selecting best server based on ping...
Hosted by Chief Telecom (Taipei) [1.62 km]: 4.501 ms
Testing download speed................................................................................
Download: 698.39 Mbit/s
Testing upload speed................................................................................................
Upload: 302.57 Mbit/s

tegrastats:

RAM 1941/3962MB (lfb 180x4MB) SWAP 0/1981MB (cached 0MB) IRAM 0/252kB(lfb 252kB) CPU [81%@1734,29%@1734,32%@1734,29%@1734] EMC_FREQ 3%@1600 GR3D_FREQ 0%@998 APE 25 PLL@44.5C CPU@45.5C Tboard@45C Tdiode@48.25C GPU@43C AO@52.5C thermal@44C PMIC@100C VDD_IN 5678/4197 VDD_GPU 233/322 VDD_CPU 2722/1200
RAM 1943/3962MB (lfb 180x4MB) SWAP 0/1981MB (cached 0MB) IRAM 0/252kB(lfb 252kB) CPU [96%@1734,35%@1734,41%@1734,37%@1734] EMC_FREQ 4%@1600 GR3D_FREQ 0%@998 APE 25 PLL@44.5C CPU@45.5C Tboard@45C Tdiode@48.5C GPU@43.5C AO@52.5C thermal@44.5C PMIC@100C VDD_IN 5523/4362 VDD_GPU 233/310 VDD_CPU 2411/1351
RAM 1945/3962MB (lfb 180x4MB) SWAP 0/1981MB (cached 0MB) IRAM 0/252kB(lfb 252kB) CPU [97%@1734,37%@1734,35%@1734,36%@1734] EMC_FREQ 4%@1600 GR3D_FREQ 8%@998 APE 25 PLL@44.5C CPU@46C Tboard@45C Tdiode@48.5C GPU@43.5C AO@53C thermal@44.75C PMIC@100C VDD_IN 5873/4530 VDD_GPU 311/310 VDD_CPU 2644/1495
RAM 1944/3962MB (lfb 181x4MB) SWAP 0/1981MB (cached 0MB) IRAM 0/252kB(lfb 252kB) CPU [96%@1734,38%@1734,27%@1734,30%@1734] EMC_FREQ 4%@1600 GR3D_FREQ 0%@998 APE 25 PLL@44.5C CPU@46C Tboard@45C Tdiode@48.75C GPU@43.5C AO@53C thermal@44.75C PMIC@100C VDD_IN 5367/4614 VDD_GPU 233/303 VDD_CPU 2333/1578
RAM 2059/3962MB (lfb 163x4MB) SWAP 0/1981MB (cached 0MB) IRAM 0/252kB(lfb 252kB) CPU [24%@1734,14%@1734,71%@1734,6%@1734] EMC_FREQ 5%@1600 GR3D_FREQ 0%@998 APE 25 PLL@44.5C CPU@48C Tboard@46C Tdiode@48.25C GPU@44C AO@52.5C thermal@45.75C PMIC@100C VDD_IN 5951/4735 VDD_GPU 233/296 VDD_CPU 2411/1654

Looks like method 1 triggers higher CPU usage. Could you try method 2?

Hi DaneLLL,

The result is form UI testing.


The result is captured from command line.

It seems that using command line is less usage of resources.
However, the download speed is not enhanced.

Hi,
Do you run on TX1 module + devkit? We are able to reach ~700Mbps in TX1 module + devkit + USB3 Ethernet dongle. It is strange only 260Mbps is achieved in 5G network.

Also we don’t set

sudo sysctl -w net.core.rmem_max=25000000

Not sure if this impacts the result.

Hi,
The buffer size is default setting and the test result is showed below.


I try to enlarge the buffer size.
The result of download speed is about 200Mbps.

rmem max size doesn’t seem to make much difference so far.
Tuning a network stack may not be easy. A user recently shared a good link for general understanding, you may have a look to it, unless you’re already familiar with this.

Hi,
We have run the USB3 ethernet dongle with Jetson Nano and Xavier in executing speedtest-cli:

[Nano]
Retrieving speedtest.net configuration...
Testing from HiNet (59.124.78.18)...
Retrieving speedtest.net server list...
Selecting best server based on ping...
Hosted by Chief Telecom (Taipei) [1.62 km]: 6.551 ms
Testing download speed................................................................................
Download: 376.26 Mbit/s
Testing upload speed................................................................................................
Upload: 144.18 Mbit/s
[Xavier]
Retrieving speedtest.net configuration...
Testing from HiNet (59.124.78.18)...
Retrieving speedtest.net server list...
Selecting best server based on ping...
Hosted by Chief Telecom (Taipei) [1.62 km]: 6.4 ms
Testing download speed................................................................................
Download: 748.17 Mbit/s
Testing upload speed................................................................................................
Upload: 597.57 Mbit/s

Looks like the CPU capability still dominates the result.
On Jetson Nano, not all four cores running at 100%, but always one core runs at 100%. Max CPU clock of Jetson Nano is 1.4GHz.
On Xavier, max CPU clock is ~2.0GHz.

We are able to launch USB3 camera in 4Kp30 on Jetson Nano, TX1 and Xavier, so USB3 bandwidth should be sufficient. We don’t try the 5G dongle, but from the result of running with ethernet dongle, maybe we have to take CPU capability into consideration.

One thing you may try is to specify the server, please run

$ speedtest-cli --list

All servers will be listed. You can run

$ speedtest-cli --server 18445

The server is

18445) Chunghwa Mobile (Taipei, Taiwan) [1.62 km]

Hi,
We run Xavier + the ethernet dongle in two different mode and get the result:

// enable 6 cores at ~2GHz
$ sudo nvpmodel -m 0
$ sudo jetson_clocks

// enable 4 cores at ~1.5GHz
$ sudo nvpmodel -m 3
$ sudo jetson_clocks


Can observe the value of max CPU clock impacts the test result.

UPDATE We use Xavier 8GB in the test. Mode 0 is

< POWER_MODEL ID=0 NAME=MAXN >
CPU_ONLINE CORE_0 1
CPU_ONLINE CORE_1 1
CPU_ONLINE CORE_2 1
CPU_ONLINE CORE_3 1
CPU_ONLINE CORE_4 1
CPU_ONLINE CORE_5 1
TPC_POWER_GATING TPC_PG_MASK 0
GPU_POWER_CONTROL_ENABLE GPU_PWR_CNTL_EN on
CPU_DENVER_0 MIN_FREQ 1200000
CPU_DENVER_0 MAX_FREQ -1
CPU_DENVER_1 MIN_FREQ 1200000
CPU_DENVER_1 MAX_FREQ -1
CPU_DENVER_2 MIN_FREQ 1200000
CPU_DENVER_2 MAX_FREQ -1
GPU MIN_FREQ 318750000
GPU MAX_FREQ -1
GPU_POWER_CONTROL_DISABLE GPU_PWR_CNTL_DIS auto
EMC MAX_FREQ 0
DLA_CORE MAX_FREQ -1
DLA_FALCON MAX_FREQ -1
PVA_VPS MAX_FREQ -1
PVA_CORE MAX_FREQ -1
CVNAS MAX_FREQ -1

Mode 3 is

< POWER_MODEL ID=3 NAME=MODE_20W_4CORE >
CPU_ONLINE CORE_0 1
CPU_ONLINE CORE_1 1
CPU_ONLINE CORE_2 1
CPU_ONLINE CORE_3 1
CPU_ONLINE CORE_4 0
CPU_ONLINE CORE_5 0
TPC_POWER_GATING TPC_PG_MASK 1
GPU_POWER_CONTROL_ENABLE GPU_PWR_CNTL_EN on
CPU_DENVER_0 MIN_FREQ 1200000
CPU_DENVER_0 MAX_FREQ 1497600
CPU_DENVER_1 MIN_FREQ 1200000
CPU_DENVER_1 MAX_FREQ 1497600
GPU MIN_FREQ 318750000
GPU MAX_FREQ 905250000
GPU_POWER_CONTROL_DISABLE GPU_PWR_CNTL_DIS auto
EMC MAX_FREQ 1331200000
DLA_CORE MAX_FREQ 998000000
DLA_FALCON MAX_FREQ 600000000
PVA_VPS MAX_FREQ 768000000
PVA_CORE MAX_FREQ 537600000
CVNAS MAX_FREQ 960000000

Hi DaneLLL,

Thank you for providing the test data.
As we tested TX1 and the 5G Module, we did enable jetson_clocks.

May I check again, does Tx1 R28.2 support nvpmodel selection?
We could not find a way to select nvpmodel on Tx1 R28.2.

Thank you,

No, R28.2 only supports TX2/TX2i nvpmodel but no TX1.

Hi WayneWWW,

Thank you for the information.