5950x dropping multi under full load

[Waldorf]

running a bench when i saw lower than usual temps.
switching to occt/prime95, i see the cpu dropping multi to 27/28, while it should be 36, or am i missing something?
no changes since install/setup month ago, and since updating to latest fw and redoing settings, no further changes and initial test/gaming was fine until now.

just to see i switched to 1st bios with previous fw/settings, no difference, same multi drop to max ~2800 mhz for clocks.

 

[ShrimpBrime]

Ok, does the clocks droop straight away, or eventually?

What temps? Core, package, vrm, chipset???

Redo settings? Might have missed something I guess. I'd check here again before anything else.

 

[Waldorf]

Multi drops on all core load, mixed load gets up to 36x, only on idle desktop, i see spikes going higher, up to 50x.
cpu never past 57C any other sensor nevwr past 65C, while cpu reached 75-80C before, depending on room temp.

like i said, i switched to 1st bios chip running older fw with same settings i used before (and working properly), same results, with multi on auto for both versions.

had to eat, will do some other tests to see if it gets within (expected) perf levels.

 

[ShrimpBrime]

Multi drops on all core load, mixed load gets up to 36x, only on idle desktop, i see spikes going higher, up to 50x.
cpu never past 57C any other sensor nevwr past 65C, while cpu reached 75-80C before, depending on room temp.

like i said, i switched to 1st bios chip running older fw with same settings i used before (and working properly), same results, with multi on auto for both versions.

had to eat, will do some other tests to see if it gets within (expected) perf levels.

No, you did not say "switched to 1st bios switch"

You said "since updating to the latest fw and redo sertings"

I'm not ready to get into a debocal about what you said, but you need accuracy and detail in the technical field. Based on what I've read, I still don't know exactly what you are saying.

You flashed firmware.
Then

You flipped bios switch via jumper? Correct?
This one older fw though, same settings.

The bios switch is for recovery. You should run it defaults always no OC. Update the secondary bios and run that one with newer firmware.

Re-evaluate user settings again, or try defaults all first. That means clear cmos try again.

 

[natr0n]

When you update a fw/bios these days you lose performance.

 

[Waldorf]

@ShrimpBrime
my fault, probable that you read 1st post, while i edited.

yeah, because you cant carry over profiles from previous fw, so all settings have to be re-done, but i did same values as before, and initial ram bench/gaming didnt show anything out of the ordinary,
so i never looked at clocks/multi, only when i wanted to get a temp reading at max load, to see how much difference remove some air in the rad had made.

recovery it done by usb flashback with dedicated button on I/O panel, the switch is on the bottom of the pcb, to select chip 1 or 2.
when i saw low multi/clocks, i wanted to exclude settings/or fw update, so i switched to the other bios chip still containing previous fw with its settings, but no change.

there is no oc besides running ram @3600/1.35v (b-die), IF at 1:1, soc at 1.1v (auto), pbo/c-state disabled.
quick test with CB23 gets me identical numbers, while multi is doing 35-36x during bench, so not sure why full load from other stuff drops multi so low.


@natr0n
lol, not in form of multi going from 35-50x, down to 27x.

 

[A Computer Guy]

running a bench when i saw lower than usual temps.
switching to occt/prime95, i see the cpu dropping multi to 27/28, while it should be 36, or am i missing something?
no changes since install/setup month ago, and since updating to latest fw and redoing settings, no further changes and initial test/gaming was fine until now.

just to see i switched to 1st bios with previous fw/settings, no difference, same multi drop to max ~2800 mhz for clocks.

What OS are you running? Did you recently get a Windows update? There some hubbub about updates coming down the Windows 11 pipeline because of performance issues affecting 9550x (and below AMD CPU's) that were supposed to get fixed sometime soon.

 

[Waldorf]

nah, still running 10, and probably stay on it, until hw/games require 11.

 

[A Computer Guy]

nah, still running 10, and probably stay on it, until hw/games require 11.

The only time I saw clocks drop like that on 5950x is when I was testing NH-L9a-AM4 and it was throttling from heat but you said your temps were fine so I have no idea unless you set a temp limit by mistake in PBO or your set to power saver power plan.

 

[Waldorf]

yeah, definitely not temp related, its cooler than ever (flushed the loop while back, and finally removed the air trapped a fw weeks ago).
nope, pbo and the like is off, and while i use power saving for low load use, i always check before running stress/bench/games, that im on balanced (99/100% for min/max).

retested with R20 to warm up the loop, then did R24, both results within margin, vs testing on first install of 5950x, on even older fw.
just dont see why the drop below 35x on prime/occt, but not on things like CineBench, and they do all-core load as well.

will do some more messing around and see if it affects me perf wise, but so far everything i used, showed (most of the time) slightly better results, if im not looking at multi ratio.

 

[eidairaman1]

Any time you decide to do a bios update, save your profile to a usb drive, take pictures of your settings, write them down also a pic of factory defaults helps to determine if anything changed with the new bios...

 

[Thimblewad]

Did you maybe have some custom PBO limits on the currents? Amps on TDC etc.?

 

[freeagent]

That is normal behaviour. It will drop clocks when it reaches its power limit. The bios version, AGESA version, none of that will matter unless you manually tune PPT/TDC/EDC in conjunction with your curve offset.

Here is Linpack, you can see stock performance vs PBO with a tune.

No PBO:

Tuned PBO:

Outside of the Linpack results, take note of the minimum global frequency limit, as well as power figures.

 

[Waldorf]

@eidairaman1
any proper board will prevent you from reusing a profile from previous fw, mine included.

while i did write down tweaks for my rig, i now had worked and owned so many ryzens, i can do (ryzen) optimized settings in my sleep, (my) ram timings as well (only 5 besides main have changed),
as i spend days to tune them.

@Thimblewad
no, pbo is manually disabled, not needed, with gsync/vrr (capped@58fps) most games will run fine on power saving profile (with clocks reduced to 50%).

@freeagent
i get that, but i was comparing absolute identical settings (outside fw itself), and no pbo.

@all
having now run multiple benches in 2k/4k, all returning slightly improved numbers as cooling is better with the air out of the rad, meaning even if there is a multiplier drop,
i dont see any negative impact perf wise.
so i guess i just have to apply some herbs and try induce some short term memory loss , and just forget about it..

thnx all for replies.

 

[eidairaman1]

@eidairaman1
any proper board will prevent you from reusing a profile from previous fw, mine included.

while i did write down tweaks for my rig, i now had worked and owned so many ryzens, i can do (ryzen) optimized settings in my sleep, (my) ram timings as well (only 5 besides main have changed),
as i spend days to tune them.

@Thimblewad
no, pbo is manually disabled, not needed, with gsync/vrr (capped@58fps) most games will run fine on power saving profile (with clocks reduced to 50%).

@freeagent
i get that, but i was comparing absolute identical settings (outside fw itself), and no pbo.

@all
having now run multiple benches in 2k/4k, all returning slightly improved numbers as cooling is better with the air out of the rad, meaning even if there is a multiplier drop,
i dont see any negative impact perf wise.
so i guess i just have to apply some herbs and try induce some short term memory loss , and just forget about it..

thnx all for replies.

I believe I recall that in some instances when the firmware is flashed it does remove the profiles, it is why I suggested a backup of the profile, or the other options.

 

[Waldorf]

not saying for all boards out there, but all i have used for couple customers and mine, all dont take any profile from prev fw, incl if saved on usb.
one major reason i try to get the 2nd amp Profil overwritten with my tuning, but with mussels mia, no luck yet..

 

[Zach_01]

running a bench when i saw lower than usual temps.
switching to occt/prime95, i see the cpu dropping multi to 27/28, while it should be 36, or am i missing something?
no changes since install/setup month ago, and since updating to latest fw and redoing settings, no further changes and initial test/gaming was fine until now.

just to see i switched to 1st bios with previous fw/settings, no difference, same multi drop to max ~2800 mhz for clocks.

What settings are you using for OCCT? Those can affect massively the frequency of the CPU.

Any CPU should be able to run under any load to at least at its base clock, no matter how severe the load is.
Anything below base clock might indicate something going on.

I'd like to see a full HWiNFO64 sensors under OCCT like this
Key points are marked

 

[Waldorf]

like i stated before, not a single test/bench showed less than expected perf, no matter if stock or with pbo, so it might just be that sensor refresh hits the "micro second" where the multi drops for 1st core, made the mistake of only looking at global/overall, not showing all cores/threads etc.

 

[Redwoodz]

I would think you would need to use the high performance Windows power plan instead of balanced, which may have changed values in an update.

 

[Waldorf]

nahh, only prevents down clock when picking its nose, and why those ryzen plans came out.

as long as your setting min cpu to 99, max to 100, so that the chips own algorithms will handle everthing (vs OS), and turn off any power savings for pci/usb (UPS will give "laptop" options), balanced is fine.

a lot of times it starts with the feeling something isnt right, e.g. i can tell when my tv brightness is off by <5%.

when i started putting load on the system to warm up the loop, i saw clocks graph wasnt in the "red" range (aida), so i started checking why, and thought i was still on power saving plan.
then used occt to look at power draw etc, all looking ok, but saw low mutli/clocks, then tried cpuz, same results.

but since it seems to be working fine, i even saw the slight gains (cpu stock) i expected from optimizing timing dependencies.
so while it still annoys me
i can give it a rest, since at least so far, works like it should with used parts/settings.

 

[Redwoodz]

nahh, only prevents down clock when picking its nose, and why those ryzen plans came out.

as long as your setting min cpu to 99, max to 100, so that the chips own algorithms will handle everthing (vs OS), and turn off any power savings for pci/usb (UPS will give "laptop" options), balanced is fine.

a lot of times it starts with the feeling something isnt right, e.g. i can tell when my tv brightness is off by <5%.

when i started putting load on the system to warm up the loop, i saw clocks graph wasnt in the "red" range (aida), so i started checking why, and thought i was still on power saving plan.
then used occt to look at power draw etc, all looking ok, but saw low mutli/clocks, then tried cpuz, same results.

but since it seems to be working fine, i even saw the slight gains (cpu stock) i expected from optimizing timing dependencies.
so while it still annoys me
can give it a rest, since at least so far, works like it should with used parts/settings.

I was suggesting for diagnostic purposes, to determine what's actually happening.

 

[freeagent]

I just use balanced.

 

[Waldorf]

@Redwoodz
i did check/change plans for testing..

but as long as actual perf (outside multi/clock numbers reported) isnt lower than before (or less than high perf plan), it cant be the cause/nor needed.

 

[R0H1T]

Do you have any screenshots from hwinfo? Everything in there, ideally with load & at idle.