RTK Torch FLEX NTrip/RTCM or Flex MQTT (Deprecated) - Can you Trust AI?

GPS/GNSS
1

I continue to learn how to use my RTK Torch. I am now on Firmware 2.2

There are three choices for the PointPerfect corrections

Menu: PointPerfect Service Selection
Select from the following PointPerfect services compatible with this device:

  1. Disabled
  2. Flex NTRIP/RTCM
  3. Flex MQTT (Deprecated)
    x) Exit

I have been using Flex MQTT which I believe to be working fine…. but AI (ChatGPT 5) says to use Flex NTRIP/RTCM

:white_check_mark: Use: FLEX NTRIP/RTCM

This is the correct and current method for receiving PointPerfect corrections via internet (Wi-Fi) on the SparkFun RTK Torch.

:cross_mark: Do NOT Use: FLEX MQTT (Deprecated)

  • That was an earlier method of receiving corrections from PointPerfect using MQTT over IP.

  • It’s deprecated and no longer officially supported by u-blox or by SparkFun’s firmware (v2.2 reflects this).

  • It often does not receive valid corrections due to changes in u-blox’s backend services.

:pushpin: Summary of Options:

Option Use Case Status
FLEX NTRIP/RTCM Best choice for internet delivery over Wi-Fi :white_check_mark: Recommended
FLEX MQTT Old/legacy delivery method :cross_mark: Deprecated
L-Band If you have an active L-Band antenna and subscription :white_check_mark: Alternative, but hardware-dependent

Here’s the key difference:

  • The RTK Torch has Wi-Fi and Bluetooth but no L-Band antenna.

  • PointPerfect can be delivered only over IP on the Torch, which means your only working option is the FLEX NTRIP/RTCM method.

MQTT worked on early Torch firmware builds when SparkFun first tested PointPerfect, but u-blox officially deprecated the IP-MQTT delivery mode for PointPerfect in 2024, and the Torch firmware v2.2 reflects that — it will appear as an option, but it won’t reliably receive corrections.

So yes — for the Torch, FLEX NTRIP/RTCM is the correct current setting for PointPerfect corrections over Wi-Fi.

Chris - who is clearly not AI – When I try that setting I get a Point Perfect error that I am not White Listed for that service

This device is not whitelisted. Please contact support@sparkfun.com or goto https://w ww.sparkfun.com/rtk_torch_registration to get the subscription activated. Please reference device ID: AC1518927CC203

And I no longer get RTK FIX, but with FLEX MQTT I do, and I can see the message scroll by I am using PointPerfect for corrections.

From reading PointPerfects Support document, seems like FLEX MQTT is still supported and available. Is FLEX MQTT going away or is ChatGPT having AI Hallucinations?

2

Hi Chris (@ChrisO ),

Your Torch is currently set up with a MQTT (SPARTN) plan. And, yes, MQTT is still supported and available. But if you’d like to move over to the new NTRIP/RTCM service, we can do that.

It’s either-or. Unfortunately you can’t swap service just by changing the Service Selection in the firmware. Sorry if the menu or the docs give you that impression. There’s some work we need to do behind the scenes to move you over to the other plan.

Let us know,
Paul

3

Thanks @PaulZC

Is there a cost difference?

is the older Flex MQTT going away eventually like AI suggested?

is Flex NTRIP / RTCM better performance? faster to converge, or any benefits that help me in my deep valleys and deep woods that are a big challenge?

AI says it’s NTRIP/RTCM is better and cheaper…. but we cannot trust AI….

4

ChatGPT AI Says….

Here’s the quick, practical breakdown for PointPerfect Flex NTRIP/RTCM vs Flex MQTT—with what matters for your RTK Torch in the East TN woods.

What’s the actual difference?

Aspect Flex NTRIP / RTCM Flex MQTT (SPARTN over IP)
Protocol & format TCP NTRIP streams industry-standard RTCM3 corrections direct to the receiver. Works with almost any RTK rover. u-blox Uses MQTT to deliver compact SPARTN SSR messages; a client/library on the device turns SPARTN into RTCM for receivers like ZED-F9P. developer.thingstream.io
Time-to-fix & stability Generally faster and more stable on SparkFun gear because it’s native RTCM end-to-end. Users report better, more consistent fixes vs SPARTN→RTCM path. docs.sparkfun.com Extra decode/convert step can add ~60 s and introduce variability; “OK accuracy (not as good as RTCM service)” per SparkFun docs. docs.sparkfun.com
Bandwidth A bit heavier; still small (kB/min), but continuous TCP session. trac.gateworks.com Most bandwidth-efficient (SPARTN is compact) and resilient to brief drops thanks to message design. Still needs IP data. developer.thingstream.io
Device fit (RTK Torch) Recommended / supported now on newer SparkFun firmware (Torch, EVK, etc.). docs.sparkfun.com Historically available, but SparkFun guidance and user reports favor NTRIP/RTCM on current firmware due to reliability. docs.sparkfun.comSparkFun Community
Coverage when cell dies Dies with cell data. Also dies with cell data. (SPARTN over L-band satellite is the way to ride through cell gaps—but Torch lacks L-band hardware/antenna, so that’s not an option on the Torch.) developer.thingstream.io
Accuracy (service) Flex is PPP-RTK: ~3–6 cm with seconds-level convergence in supported regions. (Same core service.) u-blox Same core service & accuracy target; delivery path is the difference. support.thingstream.io

Cost differences

For PointPerfect Flex the pricing is by connected-hours, independent of whether you pull via NTRIP/RTCM or MQTT over IP. Public references show entry plans around $5.90/month for 100 hours, with higher tiers lowering the per-hour rate. (Exact numbers can change, but protocol choice does not change the price.) portal.thingstream.iou-bloxArduSimple+1

What I’d use on your RTK Torch (Sweetwater, TN forests)

  • Use Flex NTRIP/RTCM as your PointPerfect delivery on the Torch. It’s the path SparkFun currently documents as the better performer on their firmware, with fewer “gotchas.” docs.sparkfun.com+1

  • MQTT/SPARTN over IP brings no cost advantage on Flex for the Torch and tends to be less consistent on that device family. docs.sparkfun.com

5

@PaulZC - is the legacy service going away eventually? what is the cost for the new service?

AI ChatGPT says the new service is more reliable, is it? Has anyone else been using it ?

6

I have not seen a specific End of Life statement for the service from U-blox (not sure if they do those), but IMO, based on how they have revised their service offerings, I doubt they keep it long term.

As far as ChatGPT, it seems to find a fact or statement and then make assumptions. I have had it sometimes give me good info, but then I have had it straight out lie to me. When using it instead of search for a quick find on library method/function calls (or for WordPress hooks)e, it has many times seemingly “made up the answer.” I confront it, ask it why it lied, and it just tells me “Good catch Doug, you’re right on. Here’s the way …” and then it continues to lie again. When it comes to code, I write my own and search the docs. It can’t be trusted. For general topics? Caveat emptor. P.S. - Claude AI does seem to be a bit more honest and say when it doesn’t know an answer.

1 Like