|
Post by ki4dra on Dec 15, 2022 16:59:57 GMT
I'm not sure this is the right forum thread, but move it if it needs moving.
Anyhow, I have a Ryzen CPU that's fairly beefy meaning it has 24 threads available and runs at 4Ghz natively without me having to try and over-clock it. The issue is, even with all of this processing power and relatively fast RAM, my optimal settings are "Decoding Cycles" = 2 and SWL Decoding = 1. I keep threads at 18, because I noticed adding more does nothing to help latency. Of course latency also has to do with the number of stations being decoded, so the settings above do well to keep latency low even when decoding 50 or more stations.
However, having said that, I noticed JTDX total CPU utilization hardly ever goes above 30 total_cpu%. What happens is if I turn up more Cycles and SWL Decoding, the latency just gets worse, and no more CPU used. This makes me wonder if JTDX works better on Intel? I'm asking because my Elmer has an Intel system with 4 threads, that appears to be just a bit slower in decoding than my killer Ryzen system. The other thing is that on Steve's Intel CPU, JTDX uses 100% of the CPU for decodes. Now his station is just dedicated to JTDX so using 100% is a non issue.
Is this an AMD Optimization thing? Would I gain anything by trying to compile this on my Ryzen System?
I have also tried changing JTDX setting for CPU contention I.E. Real-Time and High instead of the default CPU setting. I've also tried setting Affinity, so I could chose the cores, but that made zero difference.
Thoughts?
Thanks,
David
|
|
|
Post by Arvo ES1JA on Dec 15, 2022 18:44:12 GMT
What is Your problem? I believe increasing thread count more than 8 does not have real performance gain, at least not for decoding bandwidth less than 3.5 kHz. You can run multiple instances of JTDX to have more load for cpu.
|
|
|
Post by Wolfgang OE1MWW on Dec 15, 2022 19:00:31 GMT
...
Is this an AMD Optimization thing? Would I gain anything by trying to compile this on my Ryzen System?
... David
The answer to your question is: 'not really'
"If you bring home a Pizza on a restricted speed road, it does not matter if you drive an old Ford or a new Ferrari." ;-)
There are several charts provided by OM Hartmut (DB6LL), how to optimize JTDX.
The minimum setting:
and the optimum settings:
|
|
|
Post by ki4dra on Dec 15, 2022 19:58:22 GMT
What is Your problem? I believe increasing thread count more than 8 does not have real performance gain, at least not for decoding bandwidth less than 3.5 kHz. You can run multiple instances of JTDX to have more load for cpu.
I'm kind of a Spinal-Tap sort of guy.... You know, crank it to 11? I have the computer hardware, and think I should be able to turn all the nobs up to maximum settings Otherwise what did I spend the money on? I was gaming for a while, but switched back to Ham, so it would be nice to be able to get the most out of it.
It sounds like there are no easy answers, like changing any configuration files etc, and that's okay. I'm just asking for options.
|
|
|
Post by Arvo ES1JA on Dec 16, 2022 6:40:23 GMT
Yes Your computer hardware is from top end. Additionally to Wolfgang-s recommendation I believe for You are best suitable Decoder settings: Decoding cycles = 2 Decoder sensitivity = use subpass
Then monitor that Lag value in full band, if it constantly goes to + side then set Decoding cycles to 1, if it constantly less than -0.5 then You can try Decoding cycles = 3. now You have best working conditions for Your hardware.
|
|