Starlink terminal revision 4: overview and tests

Revision 4 is the new version of the Starlink user terminal. As of February 2024, the new version is only available in the US.
I decided to test the new hardware and compare it to the well-known REV3 (a.k.a. Gen2/V2 or “Standard Actuated”).

Revision 4

If you don’t understand why this model is called REV4, I explain it in the unboxing video.
Here is my unboxing and initial review of the new Starlink user terminal:

Revision 3 and Revision 4 specifications

This is a comparison of basic specifications from the Starlink website.

Breaking into

The IP rating is higher because the new housing is not ultrasonically welded anymore. It has a complex construction and is glued with an automotive-class sealant.
Unfortunately, this housing is quite hard to disassemble without partial destruction.

At this moment, I have no idea how to seal it back. I’ll share more details on the teardown process in future posts.

Revision 4 PCB

The new Starlink terminal has a resigned PCB with all-new antenna electronics.

The next-gen Shiraz digital beamformers support more RF channels. There are only six beamformer ICs (rev3 has 16 chips). FEM elements are connected in a daisy chain and combined by groups. I didn’t do detailed reverse-engineering yet.

SoC, RAM, and eMMC are the same as in REV3.

The main difference is the PoE circuit. Now, it’s on a separate board. This PCB receives 48 PoE and steps it down to 12V and 3.3V. STM32 MCU controls PoE negotiation and board operation.

I will not focus on further reverse engineering in this article. It’s just an overview.

REV4 vs REV3: basic tests

Location: Kyiv, Ukraine
Test date: 23 January 2024
Time: 4 p.m.
Starlinks firmware version: a8a61034-4207-4484-b8ea-4c4c01f07cde
Standard user setup with routers. No bypass mode.
Conditions: clear sky

Please watch my initial and basic tests below. I’m comparing startup time, power consumption, and network speed. Snow melt mode was disabled.

During these synthetic tests, the new REV4 performed better. The total power consumption of the REV4 terminal and the new REV3 router was under 100W. This meets the stated specifications. But without the snow melt mode. Please note this.

I believe that most of the differences were caused by routers.

REV4 vs REV3: advanced tests

Location: Kyiv, Ukraine
Test date: 4 February 2024
Time: 3 p.m.
REV3 firmware version: a8a61034-4207-4484-b8ea-4c4c01f07cde.uterm.release
REV4 firmware version: c4718913-5c91-4cf2-b31a-45c187e4bca2.uterm_manifest.release
No routers were used.
Conditions: cloudy sky

During the second test, I used flent to run rrul to netperf-eu.bufferbloat.net and collected information about bandwidth and response fluctuations. Additionally, I ran a Cloudflare speed test. All tests were run separately for both Dishys, except one test where I ran the Cloudflare test simultaneously on both devices.

The video of the tests is currently editing.

During the tests, I used two GEN1 PoE bricks and two laptops with direct Ethernet connections.

Test command line: flent -p all_scaled -l 60 -H netperf-eu.bufferbloat.net -t "rrul_"$DISHY_NAME -o rrul_$DISHY_NAME.png

Results for REV3, single run:

Download the Flent file.

Results for REV4, single run:

Download the Flent file.

As you can see, it’s pretty good! The ping plot fluctuates around 100ms without any crazy jumps.

Additionally, REV3 2 TCP Streams upload and download vs. ping.

Test command for upload: flent tcp_2up -p totals -l 60 -H netperf-eu.bufferbloat.net -t "rrul_"$DISHY_NAME -o rrul_$DISHY_NAME.png
Test command for download: flent tcp_2down -p totals -l 60 -H netperf-eu.bufferbloat.net -t "rrul_"$DISHY_NAME -o rrul_$DISHY_NAME.png

Download the Flent file for 2 streams down.
Download the Flent file for 2 streams up.

REV4 2 TCP Streams upload and download vs. ping

Download the Flent file for 2 streams down.
Download the Flent file for 2 streams up.

I noticed a slight packet loss on both devices. That looks okay. For some reason, REV4 had a higher loss rate.

No more problems were noticed with either device.

REV4 power consumption

I ran the new terminal with the Snowmelt mode ON and found this device very power-hungry. The consumption jumped way higher than I expected – 186W!
Note this if you consider running this device with an autonomous power supply.

Additional tests for REV4

In a few days, I decided to run additional tests with the new terminal.
It was 2 a.m., curfew. I wasn’t able to visit my previous test site. So, I put the terminal outside my window and tested it with many obstructions.

It started snowing during testing, so I measured the power consumption again.

Location: Kyiv, Ukraine
Test date: 8-9 February 2024
Time: 2 a.m.
Firmware version: c4718913-5c91-4cf2-b31a-45c187e4bca2.uterm_manifest.release
Two modes were tested: the Gen3 router and the Gen3 router in bypass mode.
Conditions: cloudy sky, snow

The initial synthetic speed test showed some fantastic results despite not ideal conditions. This is the speed record so far!

Power consumption during the test remained around 85W. Activation of the Snowmelt mode caused a crazy jump to 200W. It wasn’t a fluctuation, but stable consumption during the test time.

This is even higher than the actual REV4 PSU advertised consumption (197W).

I ran rrul 5-minute tests in two modes: with the Gen3 router and the Gen3 router in bypass mode. There were interruptions during the tests.

Test command: flent rrul -p all_scaled --socket-stats -s 0.5 -l 300 -H netperf-eu.bufferbloat.net -t "rrul_"$DISHY_NAME -o rrul_$DISHY_NAME.png

Results for the first 5-minute run with the router in normal mode:

Download the flent file.

Results for the first 5-minute run with the router in bypass mode:

Download the flent file.

Update. February 18, 2024

The new firmware 5776ed0f-38b5-47bf-8ff4-d38f7d2c326b.uterm_manifest.release performs slightly better in terms of latency and stability.

Download the flent file, 300 seconds test.

Download the flent file, 60 seconds test.

I haven’t reverse-engineered the new router yet. It looks like here we are dealing with the same software switch (Linux bridge) with all known problems.

On the other side. I noticed that the new antenna better picks up the signal in poor (snow and physical obstructions) conditions. This obviously makes it possible to achieve higher modulation rates.
Please look at the obstruction map during the last test.

I also noticed that the obstacle map was built quite quickly. But perhaps this is simply the result of a software update.

There will be more materials on the new terminal, so stay tuned.
Thanks for reading.