Overclock.net banner

Latencymon showing high Interrupt to process latency on 14900K and 13900K systems.

4.4K views 12 replies 8 participants last post by  ajolly  
#1 ·
I'm wondering if there are others out there who might be experiencing this. I've been reading various comments, threads, and even seeing youtube videos of people complaining about high latency on their 12th/13th/14th gen systems. Now I've been using my 14900K system for a couple months now doing various things and haven't noticed any significant lag to the extent others were complaining about. The system has been overall pretty snappy with the odd hitching here or there and sometimes audio popping. This did prompt me to download latencymon and check out what it's reporting for my system and I noticed it was showing some high Interrupt to process latency.
Image


SPECS: 14900K (HT Disabled & undervolted)
Mobo: Asus ROG Z790 Apex Encore
RAM: Teamgroup 48GB(2x24) OC 8000C36
GPU: Asus ROG Strix RTX 3090
SSD: Crucial T500 2TB
PSU: Corsair RM850X

I've also got a secondary system I use to mess around with, that's got a 13900K on a MSI Z790 Carbon wifi and latencymon exhibits similar figures but again not noticing any detrimental lag or hitching with that system.
 
#3 ·
Seeing not only top 1 but the top 10 would help more since often times the software is the issue here and then it´s mostly the users fault.

But yes, your are generally right.
My 12600k and 13900ks ran tweaked (e-cores off, HT off) and haven´t been fit for live audioproduction this way.
Latencymon showed the same jitter-fest.

good news:
You can´t feel it.
You get the same benchmarking results.

Only runnning all e-cores with HT active truly fixed it for me.

So i sold my apex and ks and now i am running a
14600k overclocked with e-cores and HT active:

Image


Fixes remedies:
MSItool.exe to check and activate MSI mode for the gpu
NVcustomer.exe to not let the gpu throttle down that much

As you can see both are gpu related. Nobody said the problem is simple.

I gladly hear and see more information about this from other sources.

As far as i can tell anybody should be able to replicate the issue if some or all e-cores are off and HT has been deactivated.

It should not be possible to get under 200,00 for a minute in idle that way.

Anybody care to test this?
 
#4 ·
SPECS: 14900K (HT Disabled & undervolted)
Mobo: Asus ROG Z790 Apex Encore
RAM: Teamgroup 48GB(2x24) OC 8000C36
Are you sure your UV and mem oc is stable?

Here is mine sitting there for 3 min with an ancient OS. I've used 4 motherboards on this install.

 
  • Rep+
Reactions: Hfhjfg
#7 ·
I have stability tested my OCs 24h+ and found them to be stable and haven't noticed any problem while field testing (actually using my system). But just for a sanity check I reverted everything to stock in the bios, no xmp, intel limits enforced (not asus's default OC crap), and it is still happening.
 
#5 ·
LatencyMon could possibly be reacting to how you have C states configured in BIOS -- I see similar kernel spikes at idle with my 13700K, but nowhere near that high though, only 1000uS or so here.

The tabs to the right might offer a clue for you might be able to determine what specifically is causing them in your case. I would also be suspicious of a faulty component driver.
 
#11 ·
FYI: I had latency problems (and audio crackling) too in the past. Went through 3 different mainboards and weeks of testing... turned out it was the RAM! Simply selecting XMP in the BIOS is not guaranteed to be stable anymore. Programs were running fine, but as soon as I started to download something latency would go up like crazy. Try setting your RAM to 5600 and see if that fixes the problem.

Also disabling C-States and selecting the high performance power plan could help.
 
#12 ·
I figured it out. The issue was caused by how Windows 11 was parking cores.
I downloaded UnparkCPU which allowed me to unpark all cores and tested again. As you can see from the screenshot the spikes are nowhere near as bad and are infrequent compared to before
Image


I assuming this might have something to do with Intel's cstates because even on the ultimate performance plan, the spikes were still there.