[SOLVED] What's best for checking 3900x stability?

Sohaib

Distinguished
Mar 6, 2007
253
1
18,815
Hello all, i have been playing in Ryzen Master and i managed a 4.15 on all cores at 1.2V . I ran Cinebench R15 & R20, Timespy test and handbrake encode for about 5 minutes and its stable so far. What should i use for extended testing and for how long? Please note my cooler is not high end, its CM Hyper 212 Led Turbo so i don't really want to push higher voltage as i was getting into 90s in cinebench R20. My motherboard is ASRock x570 Creator.

Also, CPU-Z is showing consistent voltage of 1.216V even though when my CPU is totally idle. Shouldn't voltage drop when cpu is in idle state? I am worried that it might degrade cpu over time running consistently at higher voltage as i tend to leave my pc running at idle for longer periods and that might reduce processor longevity.

EDIT: So i have a stable overclock at this point and good temps. My current concern is now to only manage CPU voltage/clock at idle. If you have nothing to add in that respect please don't bother replying
 
Last edited:
Solution
I think i didn't make my intentions clear when i created this thread. My purpose is NOT TO obtain best possible overclock for my chip.
My purpose is to get best multi clock speeds i can get from all cores at low voltages. I am totally not fine pushing my processor into 1.35x range and certainly not into 1.4 range.

I am a multi thread user, i do lots of video encoding which uses all threads so if i can get 100 more MHz at lower voltages i am fine. Single core performance don't worth <Mod Edit> to me or else i would have gotten myself a 9900K. I game at 4K on a 1080 Ti (skipping 2000 series) so i am always GPU bound in games. On auto max single core boost i got was 4300 MHz. So if i am getting 150 MHz less boost on single core...

Phaaze88

Titan
Ambassador
You're better off just getting a better cooler and leaving the cpu on auto, as it would most likely score even higher, but since you don't have the appropriate cooling anyways...

A static all core frequency doesn't benefit Ryzen 3000 in the same manner as an Intel cpu - you have to push a much higher all core frequency to get any real benefit over auto.
 

Sohaib

Distinguished
Mar 6, 2007
253
1
18,815
You're better off just getting a better cooler and leaving the cpu on auto, as it would most likely score even higher, but since you don't have the appropriate cooling anyways...

A static all core frequency doesn't benefit Ryzen 3000 in the same manner as an Intel cpu - you have to push a much higher all core frequency to get any real benefit over auto.
Auto sucks, I have seen voltage going into 1.4x stable and at load its always around 1.36x which easily push temps into high 80s and/or 90s and that's all at 4.05 GHz all core. Doing 4150 all core at 1.20 i have beaten multi-core score in cinebench R20 and almost reached single core (about 3% slower) while reducing temps by as much as 10-15C which is very significant. Now under stress i am not hitting above 75 C.

Also using auto i never saw single core going above 4.3 GHz. This is also my initial overclock. I have not pushed 4.175 or 4.2 GHz on 1.2V yet. I WILL if i am stable at 4.15.

Sorry but i am simply not going back to crazy 1.36x V. Just sitting at desktop i can see temps touching higher 50s or even 60s doing nothing.
 

Phaaze88

Titan
Ambassador
Auto is normal, and operating within AMD's specifications.
The reason it's not boosting as high is because your temps suck - insufficient cooler.
It's boost behavior is very similar to Nvidia's Gpu Boost, which is also temperature sensitive.

Ryzen 3000 is a very 'bursty' chip yielding higher idle temps, yet lower max load temps than the competition.
Idle is also never truly 'idle'. Windows is always doing stuff in the background.
Your current experience is expected due to the budget cooler you have mounted on it. Treating it like an Intel cpu - the static frequency - isn't ideal either.
 

Sohaib

Distinguished
Mar 6, 2007
253
1
18,815
Auto is normal, and operating within AMD's specifications.
The reason it's not boosting as high is because your temps suck - insufficient cooler.
It's boost behavior is very similar to Nvidia's Gpu Boost, which is also temperature sensitive.

Ryzen 3000 is a very 'bursty' chip yielding higher idle temps, yet lower max load temps than the competition.
Idle is also never truly 'idle'. Windows is always doing stuff in the background.
Your current experience is expected due to the budget cooler you have mounted on it. Treating it like an Intel cpu - the static frequency - isn't ideal either.
I don't follow you. So what you are saying basically is that i should leave on auto which is 1.368V @ 4.05 GHz all core because its better then 1.2V @ 4.15 GHz?
Nothing short of a huge radiator under custom loop is going to give you less then 80s under stress when auto is pushing voltage to almost 1.4V while getting no speed out of it.
Sorry but i am not comfortable with that voltage and i am going back to my origginal question.
That was the best utility for long stress tests.
 

Phaaze88

Titan
Ambassador
The reason it's only boosting to 4.05ghz is because of your cooler.
The 3950X doesn't pull so much power that the only option is a radiator. The 9900K still pulls more power than it does and pretty much demands a radiator - 280mm or greater.
The NH-D15S, Dark Rock Pro 4, and a Freezer 34 esports Duo are easily viable air cooling solutions for your cpu; a Hyper 212 isn't going to cut it though.

cinebench R20
You were already running one of the best cpu only stress tests - not nearly long enough though.
You will need to do a full system test eventually. After all, some of the gpu's heat gets absorbed into the cpu heatsink.
If you thought the cpu was hot at 90C on just a few minutes of R20, Asus Realbench, AIDA 64, and some Folding@Home are really going to push the heat.

Just get a better cooler. A big ol' slab of radiator isn't necessary for that cpu.
 

zx128k

Reputable
Best way to overclock the 3900x without going mad and getting phase change is getting a chiller. With a chiller you should be able to hit high frequency with, "water would sit around +3 under load. CPU would spike under load to the teens/twenties." This video is a 4.75GHz all cores overclock @1.4volts approx.


Note that keeping the cpu at low temps is key for getting the best overclock but is expensive. Even water is unable to get you really low temps. With my 3800x with a 360 rad 6 fans in push/pull, an all core 4.4GHz will see me hit over 80c in prime95 8fft. The more cores the worse it becomes. As the temps rise the clocks drop.

For the 3900x decent cooling is a must. I would get the best AIO you can afford or better. The biggest radiator in the world like the Watercool MO-RA3 360 PRO Black (25020) is pointless because you can't get the coolant below the ambient air temps and the money is better spent on a chiller.

Low temps are the best way to get higher clock speeds, also a chiller can keep these low temps stable. Getting close to zero is the way to go. Between 0 and 25c you get most of the boost you can get from the cpu. At these temps you should get approx. 4.26GHz all cores overclock by doing nothing (stock boost).

amd-ryzen-3900x_cold-scale_all.png

source

All core 4.20GHz should be possible, if you win the silicon lottery and get a decent AIO. Best Liquid CPU Coolers 2020 The more you aim to rise the all core clock frequency under heavy loads, then the more you will spend on cooling.
 
Last edited:

Sohaib

Distinguished
Mar 6, 2007
253
1
18,815
I think i didn't make my intentions clear when i created this thread. My purpose is NOT TO obtain best possible overclock for my chip.
My purpose is to get best multi clock speeds i can get from all cores at low voltages. I am totally not fine pushing my processor into 1.35x range and certainly not into 1.4 range.

I am a multi thread user, i do lots of video encoding which uses all threads so if i can get 100 more MHz at lower voltages i am fine. Single core performance don't worth <Mod Edit> to me or else i would have gotten myself a 9900K. I game at 4K on a 1080 Ti (skipping 2000 series) so i am always GPU bound in games. On auto max single core boost i got was 4300 MHz. So if i am getting 150 MHz less boost on single core in games it won't make any difference again because i am gaming at 4K.

Also people always assume that everyone in the world is living either in US or EU who can easily get best parts at lowest rates. I never asked anyone's suggestion for which cooler i use, i know VERY WELL what are the best CPU coolers today. But since everyone needs to go out of topic and suggest whats best for me here take a look how much a Noctua NH-D15 will cost me if i order it at amazon:
ANwqRVH.png


I definitely know what's better for my workload at my budget then anyone else here. I certainly not going to buy NH-D15 which isn't available locally at $370 USD.
No one has bothered to answer my real question which i asked other then just giving their own opinions about what cooler i am using.

One of my question was that changing voltage and core frequency in Ryzen Master locks it to that even when i am idle. Is there anyway to make frequency and voltage to go down when i am not doing anything significant and just idling? Like it does when things are on auto?

Finally i was looking at silicon lottery and it seems i have a very good chip which can sustain 4.15 GHz all core @ 1.2V. It would be silly to leave it to auto which pushes voltage into 1.4V range just to get a few MHz more on a single core.
 
Last edited by a moderator:
Consistent with your stated aim to get best multi clock speeds from all cores at low voltages, I completely agree with the others saying you should simply leave it in AUTO.

Check out this video where Buildzoid shows you how to (fairly easily) tweak PBO settings to provide maximum performance in all scenarios while staying in safe operating ranges. That means heavy loads or light bursty loads and still be temp controlled in extreme stress tests. All loads are all performing well. As it is, with that manual all-core overclock you've managed to kill performance for lightly threaded, bursty loads which all games present to a CPU and as well most productivity apps depend on.

It's also important to understand that voltage isn't the thing that degrades CPU's when under heavy loads; it's temperature accompanying current. When you leave settings in AUTO then the CPU's control mechanisms are left in place to lower frequency and voltage to stay within safe design parameters as temperature increases under load. Many people who've locked themselves into manual overclocks at voltages they thought 'safe' are finding their CPU's already degrading. I can imagine that's because the assumed 'safe' voltage is actually higher than the boost algorithm would allow, so it kept the CPU from protecting itself.

Ryzen's boosting follows very strict thermal limits, so it's definitely true that the more cooling you can provide one the better it will hold boosts. But going below ambient room temp, as with a chiller, has never proven practical except short term...as for chasing high BM scores in competitions. That's because of the condensation on electronic components that kills them early, no matter how much effort you put into preventing it. That's why it's largely faded into oblivion; if you're going to do it do it all the way with LN2.
 
Last edited:
  • Like
Reactions: zx128k and Phaaze88
To answer your question you can run Prime95 v.26.6 or latest with AVX turned off. Let it run for a few hours and see if you get a crash or any workers stopped running.

And when you overclock and set a static voltage it'll stay there all the time. You can search P-state overclocking and see if it's what you're looking for.
 

zx128k

Reputable
I think i didn't make my intentions clear when i created this thread. My purpose is NOT TO obtain best possible overclock for my chip.
My purpose is to get best multi clock speeds i can get from all cores at low voltages. I am totally not fine pushing my processor into 1.35x range and certainly not into 1.4 range.

I am a multi thread user, i do lots of video encoding which uses all threads so if i can get 100 more MHz at lower voltages i am fine. Single core performance don't worth <Mod Edit> to me or else i would have gotten myself a 9900K. I game at 4K on a 1080 Ti (skipping 2000 series) so i am always GPU bound in games. On auto max single core boost i got was 4300 MHz. So if i am getting 150 MHz less boost on single core in games it won't make any difference again because i am gaming at 4K.

Also people always assume that everyone in the world is living either in US or EU who can easily get best parts at lowest rates. I never asked anyone's suggestion for which cooler i use, i know VERY WELL what are the best CPU coolers today. But since everyone needs to go out of topic and suggest whats best for me here take a look how much a Noctua NH-D15 will cost me if i order it at amazon:
ANwqRVH.png


I definitely know what's better for my workload at my budget then anyone else here. I certainly not going to buy NH-D15 which isn't available locally at $370 USD.
No one has bothered to answer my real question which i asked other then just giving their own opinions about what cooler i am using.

One of my question was that changing voltage and core frequency in Ryzen Master locks it to that even when i am idle. Is there anyway to make frequency and voltage to go down when i am not doing anything significant and just idling? Like it does when things are on auto?

Finally i was looking at silicon lottery and it seems i have a very good chip which can sustain 4.15 GHz all core @ 1.2V. It would be silly to leave it to auto which pushes voltage into 1.4V range just to get a few MHz more on a single core.

If you game then the 3800x is better. You get higher clock speeds and you don't get the added latency of two CCD's. In games withthe 3800x you will get up to 4.45GHz boost with water, with the 1.0.0.3 ABBA BIOS. The issue with the 3900x is the cooling you will require. Games don't require 12 cores, 6-8 cores at most. To get the same clock speeds as the 3800x you will need a chiller.

With my 3800x overclocked to the maximum I get 11800 Time spy cpu with IF 1900 and RAM 3800. https://www.3dmark.com/spy/9573602 Basically I score the same as a 9900k in games, with the same gpu. A 9900ks on average scores 11730 points.

mUfTa4qL4JakU4iAMYcWk8-650-80.png


The 3900x is a bad processor for gaming as the clock speeds are lower and for games a 8 core 9700k is basically enough with a 5-5.1GHz or more overclock.

The 3800x is just as fast in games stock as the 3900x.

Assassins-Creed-Odyssey-min-1024x576.jpg

Battlefield-V-min-1024x576.jpg

The-Division-2-min-1024x576.jpg


jpg


In games the 3900x is just not that much faster than the 3700x/3800x in games.

Consistent with your stated aim to get best multi clock speeds from all cores at low voltages, I completely agree with the others saying you should simply leave it in AUTO.

Check out this video where Buildzoid shows you how to (fairly easily) tweak PBO settings to provide maximum performance in all scenarios. That means heavy loads or light bursty loads and still be temp controlled in extreme stress tests. All loads are all performing well. As it is, with that manual all-core overclock you've managed to kill performance for lightly threaded, bursty loads which all games present to a CPU.

Ryzen's boosting follows very strict thermal limits, so it's definitely true that the more cooling you can provide one the better it will hold boosts. But going below ambient room temp, as with a chiller, has never proven practical except short term...as for chasing high BM scores in competitions. That's because of the condensation on electronic components that kills them early, no matter how much effort you put into preventing it. That's why it's largely faded into oblivion; if you're going to do it do it all the way with LN2.

You can run phase change 24/7 and that makes running a chiller child's play. Example

 
Last edited by a moderator:
Solution
I ran OCCT heavy test for 2 hours and didn't get any crash.
Now i need to search how to manage my voltage at idle.
What are your CB20 scores... both multithreaded and single threaded.

If being able to keep voltages artificially low is your single purpse and you're not concerned about killing performance just disable SMT, disable cores, set a low TDP limit.
 
Last edited:

zx128k

Reputable
For games he should use pbo and auto OC. For productivity he should manually push the all core clock speeds as high as possible. The main part of the overclock will be the RAM. If lucky then an IF of 1900 will be stable. If possible then set the ram to 3800 and tighten the timings as much as possible.

The RAM overclocking really increases performance a hell of a lot, much more than the core clock speeds. For example my timings are cl15 (but as low as you can get with everything else) @ 1.44 volts on the RAM using samsung b-die.

With video encoding you use NVENC.

 
Last edited:
  • Like
Reactions: Phaaze88

Sohaib

Distinguished
Mar 6, 2007
253
1
18,815
If you game then the 3800x is better. You get higher clock speeds and you don't get the added latency of two CCD's. In games withthe 3800x you will get up to 4.45GHz boost with water, with the 1.0.0.3 ABBA BIOS. The issue with the 3900x is the cooling you will require. Games don't require 12 cores, 6-8 cores at most. To get the same clock speeds as the 3800x you will need a chiller.

With my 3800x overclocked to the maximum I get 11800 Time spy cpu with IF 1900 and RAM 3800. https://www.3dmark.com/spy/9573602 Basically I score the same as a 9900k in games, with the same gpu. A 9900ks on average scores 11730 points.

mUfTa4qL4JakU4iAMYcWk8-650-80.png


The 3900x is a bad processor for gaming as the clock speeds are lower and for games a 8 core 9700k is basically enough with a 5-5.1GHz or more overclock.

The 3800x is just as fast in games stock as the 3900x.

Assassins-Creed-Odyssey-min-1024x576.jpg

Battlefield-V-min-1024x576.jpg

The-Division-2-min-1024x576.jpg


jpg


In games the 3900x is just not that much faster than the 3700x/3800x in games.



You can run phase change 24/7 and that makes running a chiller child's play. Example

Yeah and 3800x is also better then 3900x in video encoding? Yeah right, you obviously didn't read my post and what my main purpose for this processor is so i am not going to bother arguing.
 

Sohaib

Distinguished
Mar 6, 2007
253
1
18,815
I have been searching through my bios settings and found this:

m8bV8nm.jpg


According to the description Power saving features for idle cores remain active but HWinfo still reports static voltage and frequency so obviously its not working. I did find pstate settings but i would rather not get into it right now.
I have been using ryzen balanced power plan but switching to windows balance also doesn't downclock cores at idle.
 

zx128k

Reputable
I have been searching through my bios settings and found this:

m8bV8nm.jpg


According to the description Power saving features for idle cores remain active but HWinfo still reports static voltage and frequency so obviously its not working. I did find pstate settings but i would rather not get into it right now.
I have been using ryzen balanced power plan but switching to windows balance also doesn't downclock cores at idle.

You have set a static clock speed and a constant voltage.
 

Sohaib

Distinguished
Mar 6, 2007
253
1
18,815
Is Prime95 still reliable for stress testing? I ran a blend test on all cores and left, came back after an hour and saw temps are only 58C with CPU drawing only 38W power in Ryzen Master.
That surely can't be right? But CPU cores are all maxed out. Am i wasting time in prime95? I was thinking of running it all night as i pushed my all-cores to 4175 MHz today.

uNixq4X.png
 

zx128k

Reputable
Is Prime95 still reliable for stress testing? I ran a blend test on all cores and left, came back after an hour and saw temps are only 58C with CPU drawing only 38W power in Ryzen Master.
That surely can't be right? But CPU cores are all maxed out. Am i wasting time in prime95? I was thinking of running it all night as i pushed my all-cores to 4175 MHz today.

uNixq4X.png

Blend is not that high temperature wise, you should get 60c. I do with the 3800x. Try a HEVC load and watch what happens. Big increases in temps. Also aida64 will test more of the cpu but prime95 8k fft will really heat things up.
 
Is Prime95 still reliable for stress testing? I ran a blend test on all cores and left, came back after an hour and saw temps are only 58C with CPU drawing only 38W power in Ryzen Master.
...
What ver. Prime 95? Use Prime 95 ver. 29.4 or be sure to tick off use of AVX if a later version. AVX in P95 is quite simply a CPU destroyer with no value except to de-rate a system and not representative of how AVX instructions ever occur in real programming.

Use a custom setting, fix the FFT size to 128k min and 128k max and tick the box to run FFT's in-place.

On second thought, running AVX may best for you. Considering how successfully you're managing to throttle this CPU's performance, it appears you want to de-rate your system (for whatever purpose). Run P95 v 29.8b6 or later, run custom with FFT's fixed at 20K or smaller, run in-place and make sure all the AVX boxes ticked "on". This is the most extreme load possible, if you stay stable with temps well controlled in this you've successfully de-rated it for extreme environment operation. Such as: in a closed area under a desk or where someone is likely to forget it and let the dust accumulate.
 
Last edited:

zx128k

Reputable
What ver. Prime 95? Use Prime 95 ver. 29.4 or, if later, be sure to tick off use of AVX. AVX in P95 is quite simply a CPU destroyer with no value except to de-rate a system and not representative of how AVX instructions ever occur in real programming.

Use a custom setting, fix the FFT size to 128k min and 128k max and tick the box to run FFT's in-place.

On second thought, maybe you running AVX is best for you considering how successfully you're managing to throttle this CPU's performance. It appears you want to de-rate your system (for whatever purpose) and running P95 v 29.8b6 or later, run custom with FFT's fixed at 20K or smaller, run in-place and make sure all the AVX boxes ticked "on". This is the most extreme load possible, if you stay stable with temps well controlled in this you've successfully de-rated it for extreme operation. Such as: in a closed area under a desk or where someone is likely forget it and let it accumulate dust.

Video encoding uses avx and temps will hit as shown above over 90c. 3900x has a primary use of video encoding.

Have you tried a HEVC load?

AyoKeito

AMD 3900X / ASUS x570 Pro1 point · 2 months ago · edited 2 months ago

For CPU-Z stress test, i'm sitting at 80-85C.
During Adobe Media Encoder 4K x264-x265 very slow with Voukoder plugin (ffmpeg essentially) it's 85-93C. But i have to say that i use manual 4100-4000 1.3V profile for those encoding jobs. PBO would most likely be hotter. FHD encoding jobs, are, on the other hand, sitting at ~70C.
I'm using water temperature to adjust fans RPM and avoid fans spinning up on desktop and light workloads (see this thread). As for temperatures, those are pretty much what i had on Deepcool 360 AIO.
EK 360 slim radiator with 3 pull Vardar fans, EK Velocity water block.

source

360 AIO stock 81.5c full load. 4070MHz all cores. https://www.amazon.com/Enermax-LIQT...-LTTO360-TBP-W/dp/B07N8XDKQ9?tag=hardfocom-20
the temp difference betweet the aio and the custom loop is about 1-3C in most tests.
I have a 3900x cooled with an Enermax LiqTech II 360 and when running the AIDA64 stress test for an hour I was right around 83c and all cores at 4ghz.
Mine at wide open is hitting 90c in some benches even with an XSPC waterblock and custom loop. https://hardforum.com/threads/3900x-is-a-toaster-oven-mine-solved.1983849/
O6pffZL.png
 
Last edited by a moderator:
Video encoding uses avx and temps will hit as shown above over 90c. 3900x has a primary use of video encoding.
...
YES I am well aware video DOES use AVX instructions for encoding...but not nearly in the same tightly nested loops, incredibly well optimized to always stay within the closest local processor cache as Prime95 has done it. P95 is just unreal about how it does that, but it's a good way to de-rate a system for extreme use
 

zx128k

Reputable
YES I am well aware video DOES use AVX instructions for encoding...but not nearly in the same tightly nested loops, incredibly well optimized to always stay within the closest local processor cache as Prime95 has done it. P95 is just unreal about how it does that, but it's a good way to de-rate a system for extreme use

Prime 95 is not the heaviest load, some video encoding loads can hit the cpu hard. It's just good for testing cooling and power draw. I just find it funny when people get lower temps than I get with 4 more cores and the same clock speed (cooling and voltage).

People who think they are okay 4.4GHz 12 cores, when other people get 83c and all cores at 4ghz. Just be mindful of what a real heavy load can do to cpu temps. x264, x265 and VTM video encoders can use AVX2 or AVX-512 to speed up encoding. Blender uses AVX2 in the render engine cycles. source

Adobe Media Encoder 4K x264-x265 very slow with Voukoder plugin (ffmpeg essentially) it's 85-93C. EK 360 slim radiator with 3 pull Vardar fans, EK Velocity water block.



power2.png


temps2.png


Using a Corsair H100X 240mm AIO
A 280 AIO can't keep a 4.25GHz all cores overclock under control in aida64. Temperatures even hit 80c in Cinebench. source
 
Last edited:
Prime 95 is not the heaviest load, some video encoding loads can hit the cpu hard. It's just good for testing cooling and power draw. I just find it funny when people get lower temps than I get with 4 more cores and the same clock speed (cooling and voltage).
...

I'm sure some encoding loads can hit it hard... but try that P95 custom setup with AVX boxes ticked and see if it's not hitting your CPU hard. And the thing is, it's consistently and unrelentingly hard. And it never slacks off as scenes change or the end of an encoding comes up, allowing whole processor to cool down and the cooling stack time to shed heat as it loops (assuming you set it up with a script to do that). P95 loops a thread at a time as an FFT finishes, with the rest of the processor still dumping a lot of heat into the cooling stack. It's just unrealistic which why I feel it's good only for derating.

The other one is probably better for heat output though; with larger FFT's and no AVX I think it's consistently putting more of the CPU in use and that ticks up the heat output somewhat.
 
Last edited: