prime95 am i using it wrong

[RodClutcher]

hi when i run prime i see workers stopped. from what i read that's unstable overclock

i have my R 5 3600 set to allcore 4.2 @ 1.35v pretty much everything else on auto and when gaming and everyday use im seeing no problem but if i run prime95 i get as shown in the photos

 

[newtekie1]

You have to expand the windows for the workers to see why they are stopped. But usually when a worker stops automatically it means the CPU is unstable.

The CPU can seem stable when playing games, because games don't tend to fully load the CPU, but unstable under full load like Prime95.

 

[oobymach]

1.35v is pushing it a bit high for these cpu's, try setting to 1.25v for 4.2ghz all cores. When running p95 do blend test with the window fullscreen so you can see workers. Your failures are probably ram related, use 1usmus ryzen ram calculator to figure that out, get it here.

 

[EarthDog]

One way to see what is stable is by running the right test for the job.

Small FFT tends to isolate CPU stability issues while Blend tests the CPU AND Memory. So to eliminate memory, see if small FFT runs.

What are your peak temperatures at 1.35V?

 

[tabascosauz]

1. Setting 1.35V in BIOS doesn't mean you're getting 1.35V to your CPU. There is such a thing as Vdroop. Download HWInfo and monitor the CPU Core Voltage SVI2 TFN sensor during Prime95 Small FFT for a most accurate die sense approximation of what's your actual Vcore under load.

2. Not all 3600s are guaranteed to be great overclockers capable of 4.4GHz @ 1.25V (as an (increasingly common) example). Silicon quality variance on these CPUs is significant. For example, if your chip was made in summer 2019, the probabilities are stacked in favour of it being a dud. Even if your chip is from late 2020, you can very well receive something that can't do all core 4.2.

3. Okay, so it appears you've just picked 128K which lies in the Small FFT test range. Can't remember if that's the most demanding FFT in that test, but you may as well just run Small.

4. By default, unless you check the boxes below disabling AVX, Small and Smallest run AVX workloads. Under stock settings, the CPU naturally responds by running a lower (~3.9-4.0GHz all-core) clockspeed and Vcore, because AVX is quite the burden. Now, you're forcing the CPU to run at full speed without the [pseudo-]"AVX offset" it enjoys at stock. Of course it's not going to be stable.

5. If you want, you can run Small with AVX disabled, but in that case no one can guarantee the accuracy of Prime95 testing results when you happen to run into another workload that uses AVX (e.g. Cinebench R20/R23).

6. You're already past the usual FIT safe voltage limit, because while stock all-core leverages up to 1.35V, stock clocks are smart enough to drop to 1.28-1.33V under AVX workloads. If the CPU can't be made stable all-core @ 1.35V, it may be time to consider going back to stock or trying some PBO instead of all-core.

 

[RodClutcher]

PBO shows 1.35v under full load and the normal 1.4...v under idl conditions so hints to why i stopped at 1.35v. also because anything lower will give an arror when testing with R20 but not with R15 but i guess thats because of the different world loads of the avx as mentioned

also temps get 80plus when using pbo personally i dont like pbos function i see lower temps with the setting i have atm but im happy to accept its not working correctly

after posting i have continued to change a few setting as shown in photos. changed pbo to manual settings and llc set to the lowers dropping from 1.35-1.28v ish under load. is that too much Vdroop

im not setting xmp yet because i thought removing that could of been the issue

blend test im guessing i have the wrong setting but as seen in photos my result from a quick 10 min run gave one stopped worker

 

[thesmokingman]

i have my R 5 3600 set to allcore 4.2 @ 1.35v pretty much everything else on auto

Yea, you cannot just randomly pick a number for voltage. That's a good way to degrade the chip and it clearly doesn't like it. All core overclocks are really an archaic method of oc today. I would use pbo, raise up the scalar and boost limit then you get the best of both worlds, single thread and multi thread improvements.

Can't remember if that's the most demanding FFT in that test, but you may as well just run Small.

Small is not that hard, more of a cooling test. Smallest would be the hardest as far as heat and power. But the thing with prime is that each test serves a different function, so even using smallest it doesn't amount to much on its own. Run blend for a few hours and then the chip will be put thru its paces covering heat/power, ram, imc.

 

[R-T-B]

Smallest would be the hardest as far as heat and power.

No, it's small. Small utilizes cache sized FFTs, and thus works the cache hard. Cache is hot. Thus more heat / power draw.

 

[RodClutcher]

Yea, you cannot just randomly pick a number for voltage. That's a good way to degrade the chip and it clearly doesn't like it. All core overclocks are really an archaic method of oc today. I would use pbo, raise up the scalar and boost limit then you get the best of both worlds, single thread and multi thread improvements.



Small is not that hard, more of a cooling test. Smallest would be the hardest as far as heat and power. But the thing with prime is that each test serves a different function, so even using smallest it doesn't amount to much on its own. Run blend for a few hours and then the chip will be put thru its paces covering heat/power, ram, imc.

all good i guess you didn't read what i added but idk who claiming picking random numbers all good though. i know what pbo does and by the looks i will end up reverting to it but i see allot lower scored using pbo vs attempting to allcore. i accept its unstable and unhappy so i should NOT keep running it this way BUT...

therapeutically whats the negative affects i will see from having a unstable oc in prime but seeing no issues in programs or gaming ect.

am i interdicting errors to the os or something catastrophic or just the fact it may crash when under heavy loads

 

[thesmokingman]

all good i guess you didn't read what i added but idk who claiming picking random numbers all good though. i know what pbo does and by the looks i will end up reverting to it but i see allot lower scored using pbo vs attempting to allcore. i accept its unstable and unhappy so i should NOT keep running it this way BUT...

therapeutically whats the negative affects i will see from having a unstable oc in prime but seeing no issues in programs or gaming ect.

am i interdicting errors to the os or something catastrophic or just the fact it may crash when under heavy loads

You'll eventually degrade the cpu. Another thing is that when you see posts of braggerts and their high all core overclock, it's kind of a joke really once you know what it all entails. No one in their right mind would run a high all core oc using high voltage daily. In most cases ppl use stupid high voltage to spit out a R20 bench and then go back to PBO.

 

[RodClutcher]

You'll eventually degrade the cpu. Another thing is that when you see posts of braggerts and their high all core overclock, it's kind of a joke really once you know what it all entails. No one in their right mind would run a high all core oc using high voltage daily. In most cases ppl use stupid high voltage to spit out a R20 bench and then go back to PBO.

i get that high v degrades chips considering i payed f all for the chip and its probably been abused already im not fully fussed but on saying that im not wanting to kill it jut play around to see what it can and cant do vs other peoples.

by the end of today i will end up resetting to stock as enough have informed me to BUT i would like to know why when i see pbo pushing 1.35v ish under full load whats the difference to setting that value in bio as a static voltage and using c state control + pbo auto.

im almost sure its dropping to 1.100v idl but the parked core thing showing the last know v is hard to tell and if it is dropping its not for long

ill reset to stock and test again but im pretty sure im seeing full load with pbo 1.35v ish. i guess ill do that now

ok i guess its not a full 1.35v underload with stock settings

it jump around down to 1.28ish for a small time but goes back to 1.3 ish so i dont get why 1.35v is too much

my 1600af oc run almost on par to the preference i see with the 3600 stock. maybe i should just let the thing fry then go back to the 1600

 

[oobymach]

1.35v is the maximum safe limit for the 3000 series cpu's, and on auto with pbo they can go past this and also run hot. If it's dropping under load that's good, it means the chip isn't trying to fry itself. I can run my 3600x @ 4.2ghz @ 1.25v but I don't run p95 for hours on end and I have a decent cooler.

P95 is an extreme torture test, even with cinebench you'll notice your minimum clocks drop to double digits with it on auto, if you're using an aftermarket cooler with 4 heatpipes or more you can manually set clocks and voltages safely, if not auto without pbo is your best option.