FT8 decodes only every 30sec

DF1LON Lars

I have a very strange problem with FT8. FT8 does a great job as long as I run only one instance of WSJT-X. As soon as I add more WSJT-X decoders, the decoders are only decoding every 30 seconds.
 

And they do it all the same. If I add 3 WSJT-X instance, each one to a different SmartSDR audio stream, they all decode in the same way! They skip RX even and decode only on RX 1st. After some time they even switch the decode periods...
I have seen it with WSJT-X 1.8 and 1.9 and even with JTDX. Reinstalling the DAX drivers did not made any difference. Currently I am running SmartSDR 2.1.33.

Any idea what I am missing?

73 de Lars
DF1LON

Mark WS7M

This sounds to me like your clock is off.  These modes require almost exact syncing of the clock.  Are you running a time sync utility like  Meinberg?   The one built into windows is no where near good enough.

Dave - WB5NHL

Check your cpu utilization. WSJT-X is heavy user when decoding. If the cpu max out at 100% you won't get decode. I have had this problem and make sure I don't have any other heavy users particularly with multiple instances.

bahillen

I had problems with decode on FT8 caused by CPU 100%. It only takes a short time, like less than 1 second. The surprising culprits are chrome spiking, monitor driver if not an external video driver, other apps like outlook and apps open in background. To check go to task monitor with performance graph and set speed to fast so peak CPU is seen. JTAlert-X is doing a lot of lookups too at that time. On my laptop, I close things I am not using including chrome, minimize what I don’t need, sometimes SmartSRD too. Using my older laptop I5 processor I keep CPU below 50% I hate messing up the pace on FT8 with missed time slots. Confusion added and delay to QSO. From my shack desktop computer I run 4 WSJT-X instances and 4 JTAlert-X instances. I had to upgrade to a Dell 3668 I7 7700, NVDIA video card and 256SSD. I didn’t want to experiment with less speed because as new apps and incredible future need I didn’t want to upgrade next year. I have found times when everything is running on 4 bands I have seen 45% CPU on new computer. 73 Bill W9JJB

Steve-N5AC

I believe you need to create a separate directory for each instance of WSJT-X in the servings panel.  This is the directory where it stores audio for processing.

DF1LON Lars

Thanks for all the hints.

The clock is definitely not off. Otherise WSJT-X would not start decoding 15sec later. It just skips every second decode periode.

I already had a look at CPU load. It goes up to 100% for less than a second and then goes back to normal. But, if WSJT-X skips the decoding periode, the CPU load stays low. WSJT-X just does nothing. Even no UDP broadcast.

I just split WSJT-X and SmartSDR. So SmartSDR runs on a different PC which gives me a bit more CPU power on the PC where the multiple WSJT-X instances are running. Now I have less decoding drops. So it might be the CPU power... Just ordered a new CPU to replace my I3 processor. Let's see if that makes the difference!

73
Lars DF1LON

Dave - WB5NHL

"I already had a look at CPU load. It goes up to 100% for less than a second and then goes back to normal. "  that's all it takes to lose a round of decodes.  I don't have a problem with 2@WSJT-X instances and SMARTSDRas long as I don't get greedy with other stuff (logging, dxclusters etc).  The real bad culprits were the continuous backup software and MS Onedrive.

Dave   WB5NHL

bahillen

Don’t need separate directories just separate rig-names. This gives separate setting files. This has been documented in the past few months in this community. 73 Bill W9JJB

K1DBO

Right, but... each rig-name seems to create a separate directory to hold the separate settings file.  It would be helpful to know if the settings file associated with a rig name can live inside the default settings directory.  

DF1LON Lars

Just a quick feedback: I replaced my i3 CPU with an i7 CPU. All those drop outs are gone. CPU load is about 60% with 5xWSJT-X running during the decoding peak.

Case closed

73 de Lars
DF1LON