Question What Is The Best Method of Overclocking an AMD?

chaos_viper15

Distinguished
Feb 8, 2010
171
1
18,695
1
Let me start by stating what I have set up on my 3900x in bios.

D.O.C.P. 3200mhz FCLK 1600
PBO Manual
PPT/TDC: 240A
EDC: 10
Scalar: Auto
C-states: Enabled
Power Supply Idle Control: Low Current Idle
CPPC Preferred Cores: Enabled
CPPC: Enabled
Power Plan Balanced (set with processor 85% min 100% max)

Moving the mouse around on desktop with only HWINFO running is see jumps to 4.350Ghz - 4.375Ghz. I see them in cincebench, Aida64, or in a game like minecraft. Not sure whats happening or what I could do.

Main uses are: Streaming/recording and gaming and the same time. I don't do this vary often but occasionally, render videos and edit photos maybe more later on. Only if i can make animations on davnci.

So what is the best method (static, PBO or what else) for getting the best performance possible as safest for longevity?
 
Last edited:

Lutfij

Titan
Moderator
Please pass on the specs to your build in the body of your thread. If you're going to suggest people look at your sig space, please don't. The logic behind that is that sig space specs can and will change over time, making this thread and it's relevant suggestions moot to the person in the same boat as you're in right now.

You can follow this reddit;
https://www.reddit.com/r/overclocking/comments/lr8dx3 View: https://www.reddit.com/r/overclocking/comments/lr8dx3/guide_overclocking_ryzen_2_3900x/

though it'd be a good idea for the community and future readers to know what your motherboard is and your current BIOS version. This adds context to the thread and any other community readers.

You could also look into this video by Jayz2Cents;
View: https://www.youtube.com/watch?v=Ssuqhyqah2k
 

chaos_viper15

Distinguished
Feb 8, 2010
171
1
18,695
1
Please pass on the specs to your build in the body of your thread. If you're going to suggest people look at your sig space, please don't. The logic behind that is that sig space specs can and will change over time, making this thread and it's relevant suggestions moot to the person in the same boat as you're in right now.
Ah right my bad I'll edit above. with system specs. Very true thousands of different builds. I'll remove sig specs too. Sorry I havent been on here it quite some time.
Also the video really does say what to set for voltage and certain other settings (at that time) outta whack. I'll check out that reddit post as well. I looked for one on there didn't see this one i guess a year ago is why.
 
....

Main uses are: Streaming/recording and gaming and the same time. I don't do this vary often but occasionally, render videos and edit photos maybe more later on. Only if i can make animations on davnci.

So what is the best method (static, PBO or what else) for getting the best performance possible as safest for longevity?
Already give are some 'how to' suggestions but what's best is really what works for you. For me the best method is going to be PBO. That simply because it's arguably the safest since it leaves the boost algorithm in control which will throttle back and lower voltage as necessary to protect the CPU.

But also: static overclocking any 2nd or 3rd gen CPU rarely results in actually improved all-around performance. If it does it's only for heavier multithreaded workloads and lightly threaded gaming workloads will end up suffering. This is because the stable achievable clock with conventional cooling methods is usually going to be lower than the peak boost clocks the CPU can achieve on it's own with an optimized PBO on light threaded workloads (such as gaming). Most of the people I'm aware to have statically OC'd to max boost frequency were either using very high voltage (drastically reducing CPU life) or were giving up on heavy threaded stability, or both.

One thing to absolutely not forget about is cooling whichever OC method you settle on: the biggest/best CPU cooler you can fit and proper case ventilation. You may not really see it with lower temperatures but with better performance when using PBO because the boost algorithm is 'temp seeking'; it keeps boosting until it gets to a certain temperature for the workload it's processing.

What I see you're trying is the "edc = 10" bug trick. It works with my 3700X too: you might try varying it though. I seem to remember 12 core CPU's saw the best effect at about EDC = 15 when I was first experimenting with it. But then, things have changed and different motherboards/BIOS' may respond differently: I found the latest BIOS works best at EDC = 5 on my MSI B450M Mortar/3700X.

Clocks are way too dynamic to use for checking if your PBO is working well for you. The best way to check results is with benchmark scores: CB20 is great for this. Check both MT and ST though since it's possible to peak one up and tank the other, very common with static OC's. And as well when trying out different EDC settings.
 
Last edited:

chaos_viper15

Distinguished
Feb 8, 2010
171
1
18,695
1
Already give are some 'how to' suggestions but what's best is really what works for you. For me the best method is going to be PBO. That simply because it's arguably the safest since it leaves the boost algorithm in control which will throttle back and lower voltage as necessary to protect the CPU.

But also: static overclocking any 2nd or 3rd gen CPU rarely results in actually improved all-around performance. If it does it's only for heavier multithreaded workloads and lightly threaded gaming workloads will end up suffering. This is because the stable achievable clock with conventional cooling methods is usually going to be lower than the peak boost clocks the CPU can achieve on it's own with an optimized PBO on light threaded workloads (such as gaming). Most of the people I'm aware to have statically OC'd to max boost frequency were either using very high voltage (drastically reducing CPU life) or were giving up on heavy threaded stability, or both.

One thing to absolutely not forget about is cooling whichever OC method you settle on: the biggest/best CPU cooler you can fit and proper case ventilation. You may not really see it with lower temperatures but with better performance when using PBO because the boost algorithm is 'temp seeking'; it keeps boosting until it gets to a certain temperature for the workload it's processing.

What I see you're trying is the "edc = 10" bug trick. It works with my 3700X too: you might try varying it though. I seem to remember 12 core CPU's saw the best effect at about EDC = 15 when I was first experimenting with it. But then, things have changed and different motherboards/BIOS' may respond differently: I found the latest BIOS works best at EDC = 5 on my MSI B450M Mortar/3700X.

Clocks are way too dynamic to use for checking if your PBO is working well for you. The best way to check results is with benchmark scores: CB20 is great for this. Check both MT and ST though since it's possible to peak one up and tank the other, very common with static OC's. And as well when trying out different EDC settings.
This is very helpful thank you for all this information. I will try with EDC I usually just go in increments of 5. Scalar wasn't doing anything for me so got rid of it. I'll probably go with PBO in the end. I remember seeing 4.3 boosts occasionally in CPU intense games. Recently been feeling slower than it should be. This is a fresh windows and motherboard just got RMA return. Maybe I'm missing something.
 
...I will try with EDC I usually just go in increments of 5...
Once you're at EDC setting of 10A you pretty much have to go in increments of 1. When I was working on my 3700X I found just going from 6 down to 5 made a significant change for the better. And by an EDC of 12 or so it wasn't working very well...by 15 not at all.

I also found I had to disable Advanced C-States or single thread performance went in the tank when using that low of EDC settings. I don't know why it would still work well for MT and only ST tank but I kept in mind this EDC=10 bug is anomalous behavior in the boost algorithm so it doesn't have to make sense. Just test performance and go with what works best.

And about voltage: I've rarely seen knowledgeable recommendations about what a "safe" voltage for 2nd and 3rd gen Ryzen CPU's might be for fixed OC's. Aside from admonitions to "keep it in AUTO" there's been very little from AMD on that too. What we do know is it's very tolerant of higher voltage, up to 1.5V, but ONLY when core temperature and current is low as when boosting single cores on bursty, light threaded workloads. People used to think that a maximum voltage of 1.25V or so would be safe under heavy loads (when current is high) so long as you can keep temperature low. But what's low enough; it's very hard to cool 7nm geometry because of the lack of surface area to remove heat across.

Only the boost algorithm can lower the voltage and clocks fast enough when core temperature gets too hot on a light threaded boost. It's working based on FIT data to keep voltage, frequency, temperature and current within safe limits. That's why I prefer PBO.
 
Last edited:
There is one method I know of and that I used to set up my OC: https://linustechtips.com/topic/1410940-ryzen-5-3600-max-safe-voltage/?do=findComment&comment=15255510
I'll quote the post too if the link stops working. This is true for both the 3000-series as well as the 5000-series.

"For daily, you want to stay near the FIT voltage used while CPU is under load and OC around that.

PBO on and everything else auto. Run Prime95 128k FFT, in-place un-ticked and watch v-core (SVI2 TFN). That will be the fitness voltage for your chip. Some chips are under 1.3v some are over 1.3v. Every chip will differ because of the cooling, which has good impact on the boost algorithm."

I know it as the Buildzoid-method. Watched his video on it. Take the lowest voltage your CPU reaches. That is your FIT voltage. My 5600X was sitting at 1.285v or so during initial FIT testing. So that is what I aimed for during heavy loads. I had to play with LoadLine Calibration (LLC) to get it there. In my case, the static voltage is set to 1.3v but thanks to my LLC level, it drops to 1.285v during heavy loads. CPU is perfectly fine running at 1.3v. I mean, stock PBO runs the CPU at 1.4-1.45 volts or something during light loads. So I dropped a ton on the temp too. At idle in Windows I would sit at 50 C with PBO, with static OC I am sitting at 35 C.
I run a static OC for noise reasons. Fans would ramp up and down constantly when running PBO. 4.45 Ghz is what is stable for me.
Your clocks will vary. So will your FIT voltage. Every bit of silicon is different, doesn't matter if it's the same model or not.

Even if you don't go for the static OC, FIT voltage limits apply.

Why Prime95 and SmallFFTs? Because Cinebench isn't a heavy load, if you thought that. Don't use Cinebench please for setting your overclock. If you do something heavier than Cinebench, you will wonder why your system is unstable/crashes. To put it into context. My CPU topped at 70 C during Cinebench runs. During Prime95, 90+ Celsius. It's a huge difference.

There is a Reddit dedicated to Overclocking. https://www.reddit.com/r/overclocking/ I would search that subreddit for your CPU and read what turns up. You can also create your own thread/ask questions. Some knowledgeable people frequent the place.
 
Last edited:

chaos_viper15

Distinguished
Feb 8, 2010
171
1
18,695
1
Once you're at EDC setting of 10A you pretty much have to go in increments of 1. When I was working on my 3700X I found just going from 6 down to 5 made a significant change for the better. And by an EDC of 12 or so it wasn't working very well...by 15 not at all.

I also found I had to disable Advanced C-States or single thread performance went in the tank when using that low of EDC settings. I don't know why it would still work well for MT and only ST tank but I kept in mind this EDC=10 bug is anomalous behavior in the boost algorithm so it doesn't have to make sense. Just test performance and go with what works best.

And about voltage: I've rarely seen knowledgeable recommendations about what a "safe" voltage for 2nd and 3rd gen Ryzen CPU's might be for fixed OC's. Aside from admonitions to "keep it in AUTO" there's been very little from AMD on that too. What we do know is it's very tolerant of higher voltage, up to 1.5V, but ONLY when core temperature and current is low as when boosting single cores on bursty, light threaded workloads. People used to think that a maximum voltage of 1.25V or so would be safe under heavy loads (when current is high) so long as you can keep temperature low. But what's low enough; it's very hard to cool 7nm geometry because of the lack of surface area to remove heat across.

Only the boost algorithm can lower the voltage and clocks fast enough when core temperature gets too hot on a light threaded boost. It's working based on FIT data to keep voltage, frequency, temperature and current within safe limits. That's why I prefer PBO.
Right I had EDC on 10-12 for a while currently I have PBO on as well. My cooler is pretty good I get idle temps around 36-38c.

There is one method I know of and that I used to set up my OC: https://linustechtips.com/topic/1410940-ryzen-5-3600-max-safe-voltage/?do=findComment&comment=15255510
I'll quote the post too if the link stops working. This is true for both the 3000-series as well as the 5000-series.

"For daily, you want to stay near the FIT voltage used while CPU is under load and OC around that.

PBO on and everything else auto. Run Prime95 128k FFT, in-place un-ticked and watch v-core (SVI2 TFN). That will be the fitness voltage for your chip. Some chips are under 1.3v some are over 1.3v. Every chip will differ because of the cooling, which has good impact on the boost algorithm."

I know it as the Buildzoid-method. Watched his video on it. Take the lowest voltage your CPU reaches. That is your FIT voltage. My 5600X was sitting at 1.285v or so during initial FIT testing. So that is what I aimed for during heavy loads. I had to play with LoadLine Calibration (LLC) to get it there. In my case, the static voltage is set to 1.3v but thanks to my LLC level, it drops to 1.285v during heavy loads. CPU is perfectly fine running at 1.3v. I mean, stock PBO runs the CPU at 1.4-1.45 volts or something during light loads. So I dropped a ton on the temp too. At idle in Windows I would sit at 50 C with PBO, with static OC I am sitting at 35 C.
I run a static OC for noise reasons. Fans would ramp up and down constantly when running PBO. 4.45 Ghz is what is stable for me.
Your clocks will vary. So will your FIT voltage. Every bit of silicon is different, doesn't matter if it's the same model or not.

Even if you don't go for the static OC, FIT voltage limits apply.

Why Prime95 and SmallFFTs? Because Cinebench isn't a heavy load, if you thought that. Don't use Cinebench please for setting your overclock. If you do something heavier than Cinebench, you will wonder why your system is unstable/crashes. To put it into context. My CPU topped at 70 C during Cinebench runs. During Prime95, 90+ Celsius. It's a huge difference.

There is a Reddit dedicated to Overclocking. https://www.reddit.com/r/overclocking/ I would search that subreddit for your CPU and read what turns up. You can also create your own thread/ask questions. Some knowledgeable people frequent the place.
I can't seem to get over just under 4050 in prime 95 to matter what I do PBO and settings testing. I get nothing worth. Might have to go for a static OC. Thing is I used to get 4.3ghz. I don't know what happened to that. That I noticed while gaming anyways. That was basic PBO enabled. Nothing touched but D.O.C.P. on my ran an fclk of course. Now obviously not working as before. Cinbench barely hitting 4.1ghz and clocks on prime95 half and voltage are 1.280 in the "current" section for 20 minutes of test. Trying again.
C-states disabled.
offset may help as before but not sure.

Report: I'm still trying might be stuck at 4.075ghz. Highest I could get stable after an hour on prime95. Vcore 1.280 stabilized. I'm not sure what LLC is never used on my intel. Well I mean for a static voltage I guess. I guess just expect to much cause my intel only took me a day 5.0ghz with 1.33v. A nice lucky chip with that low voltage. I know I'll never hit 5 but wanted at least 4.4.
 
Last edited:
...
I can't seem to get over just under 4050 in prime 95 to matter what I do PBO
...
That could be about the best to expect from a 3900x in P95 with PBO, I'm not sure. Of course, a lot also depends on the FFT size being tested. The core currents and powers are highly variable with FFT's as they seem to utilize the CPU's internal circuitry differently.

And if testing P95 with a fixed OC always be sure to disable AVX. It's so highly unrealistic to come across such tightly looped and efficient AVX instructions in code...and potentially damaging to a CPU that's not able to lower clock and voltage to protect itself.

And another thing: P95 is a stress test. It's good for testing stability for extreme use cases but mainly for stressing the cooling systems of thermally constrained PC's...which almost all modern high performance desktop processors almost certainly are. That's especially important to keep in mind with Ryzen when running PBO: if it's clocks are trending lower it's quite possible it's because the boost algorithm is protecting the CPU by keeping it within it's optimal FIT range.

Personally, I don't consider P95 all that important except as a stress test. For performance testing I look for real-world processing work loads, especially ones that are typical of your use case. CB 20 or 23 is good if you do 3D rendering a lot, for instance. It's also useful to comparing to results others get since it's fairly repeatable. I have a benchmark video that I render out in Handbrake.
 
Last edited:
That could be about the best to expect from a 3900x in P95 with PBO, I'm not sure. Of course, a lot also depends on the FFT size being tested. The core currents and powers are highly variable with FFT's as they seem to utilize the CPU's internal circuitry differently.

And if testing P95 with a fixed OC always be sure to disable AVX. It's so highly unrealistic to come across such tightly looped and efficient AVX instructions in code...and potentially damaging to a CPU that's not able to lower clock and voltage to protect itself.

And another thing: P95 is a stress test. It's good for testing stability for extreme use cases but mainly for stressing the cooling systems of thermally constrained PC's...which almost all modern high performance desktop processors almost certainly are. That's especially important to keep in mind with Ryzen when running PBO: if it's clocks are trending lower it's quite possible it's because the boost algorithm is protecting the CPU by keeping it within it's optimal FIT range.

Personally, I don't consider P95 all that important except as a stress test. For performance testing I look for real-world processing work loads, especially ones that are typical of your use case. CB 20 or 23 is good if you do 3D rendering a lot, for instance. It's also useful to comparing to results others get since it's fairly repeatable. I have a benchmark video that I render out in Handbrake.
I for sure agree. Prime95 is a very heavy workload. Another thing I forgot to mention. Prime95 will not stop just because a Core crashed. It will just say "Worker thread terminated" or something. Meaning it is unstable. Keep an eye on all cores during Prime95. Another test I often use is OCCT. Like 5 minutes of it. I think it uses Linpack.
But yeah, that is just to check stability and that temps stay within operating limits.

For actual performance, I run benchmarks in whatever I do normally. A couple games, some CPU benchmark like Cinebench etc. If the performance is dropping, something is off. Have to figure out what. So, it should be the thing you changed last. Because you only changed 1 or 2 things, right? Otherwise it becomes impossible to tell.

I have Hwinfo64 up on separate monitor to check up on temps, clocks etc. Especially I pay attention to max temp and clocks. When I was doing PBO testing, I was checking which of the 3 PBO values is the limiter. And what the other 2 values reached. I remember going from 105 watts drawn to 140 watts on the CPU and very little extra performance gained. Wasn't worth it for me.
During game benchmarks I run MSI Afterburner so I can look at CPU, GPU stats in a condensed way. So I make sure I always have MSI AB and Hwinfo64 open, every run. Because those can affect FPS etc. So if I always have them running, it should be the same situation across the different runs. Hopefully comparing apples to apples. If I notice one run is acting weird, I will run the same benchmark again. Maybe Windows decided to check if theres updates or something, who knows.


I suspect it can be OPs cooling holding him back. 3900x is a beast. Maybe also check on the paste, is it caked and dry? You should change it every 2-3 years. Prime95 dropping clocks sounds like cooling can't keep up.
 
Last edited:

chaos_viper15

Distinguished
Feb 8, 2010
171
1
18,695
1
That could be about the best to expect from a 3900x in P95 with PBO, I'm not sure. Of course, a lot also depends on the FFT size being tested. The core currents and powers are highly variable with FFT's as they seem to utilize the CPU's internal circuitry differently.

And if testing P95 with a fixed OC always be sure to disable AVX. It's so highly unrealistic to come across such tightly looped and efficient AVX instructions in code...and potentially damaging to a CPU that's not able to lower clock and voltage to protect itself.

And another thing: P95 is a stress test. It's good for testing stability for extreme use cases but mainly for stressing the cooling systems of thermally constrained PC's...which almost all modern high performance desktop processors almost certainly are. That's especially important to keep in mind with Ryzen when running PBO: if it's clocks are trending lower it's quite possible it's because the boost algorithm is protecting the CPU by keeping it within it's optimal FIT range.

Personally, I don't consider P95 all that important except as a stress test. For performance testing I look for real-world processing work loads, especially ones that are typical of your use case. CB 20 or 23 is good if you do 3D rendering a lot, for instance. It's also useful to comparing to results others get since it's fairly repeatable. I have a benchmark video that I render out in Handbrake.
I for sure agree. Prime95 is a very heavy workload. Another thing I forgot to mention. Prime95 will not stop just because a Core crashed. I will just say "Worker thread terminated" or something. Meaning it is unstable. Keep an eye on all cores during Prime95. Another test I often use is OCCT. Like 5 minutes of it. I think it uses Linpack.
But yeah, that is just to check stability and that temps stay within operating limits.

For actual performance, I run benchmarks in whatever I do normally. A couple games, some CPU benchmark like Cinebench etc. If the performance is dropping, something is off. Have to figure out what. So, it should be the thing you changed last. Because you only changed 1 or 2 things, right? Otherwise it becomes impossible to tell.

I have Hwinfo64 up on separate monitor to check up on temps, clocks etc. Especially I pay attention to max temp and clocks. When I was doing PBO testing, I was checking which of the 3 PBO values is the limiter. And what the other 2 values reached. I remember going from 105 watts drawn to 140 watts on the CPU and very little extra performance gained. Wasn't worth it.
During game benchmarks I run MSI Afterburner so I can look at CPU, GPU stats in a condensed way. So I make sure I always have MSI AB and Hwinfo64 open, every run. Because those can affect FPS etc. So if I always have them running, it should be the same situation across the different runs. Hopefully comparing apples to apples. If I notice one run is acting weird, I will run the same benchmark again. Maybe Windows decided to check if theres updates or something, who knows.


I suspect it can be OPs cooling holding him back. 3900x is a beast. Maybe also check on the paste, is it caked and dry? You should change it every 2-3 years. Prime95 dropping clocks sounds like cooling can't keep up.
Yeah I ran cinebench r23 on loop 10 times. over 18000 score and clock around 4.2ghz temp never went above 76c on that loop. I also use OCCT same results. So changed a few settings and PBO+AutoOC and a offset of 0.0625 left PPT/TDC the same and EDC on auto. I'm not sure if Fmax should be on or not but I didn't touch it either. After testing a lot of different offset or turning off completely. Results achieved!

Ran prime95, OCCT, cinebench loop and Aida64 for an 30 minutes each last night was up til 3am. Testing different things and I have got 4.275-4.3ghz on all tests. Except prime of course cause the heat was almost thermal throttling my CPU at 87c at 20 minutes in. So I think I have finally reached the max I could achieve. I have 360mm AIO cpu cooler I was running it a little low at around I think 1120 and 1600 pump speed. turn in to 1500 and pump and 1900. Huge difference and still low noise. So true must have been cooler default settings making me not achieve what I wanted.

THANK YOU BOTH! So much. I've learned so much about Ryzen OC now for the future.

Like I was saying my first ever build was in 2017 with a 7700K put it together and started playing overclocking it after reading a little got 5ghz at 1.33v with 5 hours of testing that as well. Which I think spoiled me with 5ghz. lol. So you both solved this for me. Thanks again. Appreciate you replying and helping me out.
 
Last edited:
Yeah I ran cinebench r23 on loop 10 times. over 18000 score and clock around 4.2ghz temp never went above 76c on that loop. I also use OCCT same results. So changed a few settings and PBO+AutoOC and a offset of 0.0625 left PPT/TDC the same and EDC on auto. I'm not sure if Fmax should be on or not but I didn't touch it either. After testing a lot of different offset or turning off completely. Results achieved!

Ran prime95, OCCT, cinebench loop and Aida64 for an 30 minutes each last night was up til 3am. Testing different things and I have got 4.275-4.3ghz on all tests. Except prime of course cause the heat was almost thermal throttling my CPU at 87c at 20 minutes in. So I think I have finally reached the max I could achieve. I have 360mm AIO cpu cooler I was running it a little low at around I think 1120 and 1600 pump speed. turn in to 1500 and pump and 1900. Huge difference and still low noise. So true must have been cooler default settings making me not achieve what I wanted.

THANK YOU BOTH! So much. I've learned so much about Ryzen OC now for the future.

Like I was saying my first ever build was in 2017 with a 7700K put it together and started playing overclocking it after reading a little got 5ghz at 1.33v with 5 hours of testing that as well. Which I think spoiled me with 5ghz. lol. So you both solved this for me. Thanks again. Appreciate you replying and helping me out.
Glad you got it sorted. I have read about an EDC bug. If you set it to 1. Supposed to be available on 3000-series and possibly with older BIOS-versions. Look into that if interested.
For the pump speed, I always run mine at 100%. I can't hear it. My fans max out at 2300 rpm but they idle around 1400-1500. Not that noisy. Besides, when I play games, the fans on Nvidia 2080 are louder than radiator fans anyway. And I have headphones on.
But your use-case might be different. It's your PC, you can do whatever you want. That's the beauty of it.
When it comes to overclocking, you can look at guides and other peoples settings. But I do it just to get a set of ranges I should test. Maybe PPT between 80 and 200. EDC at 1-150. Voltage at 1.25-1.325v perhaps. These are just numbers I made up.

Every chip is different, so is yours.
Do not copy other peoples settings. Those work on their chip only. And my question is always: How well did they test the system? Are they going to come back in 6 months and say they have experienced instability and getting rid of the overclock solved it? I was just reading about exactly that yesterday on Reddit and a 3900X. Older post but anyway. These things happen.
It's like when Ryzen 5000-series launched and just about every Youtube hardware-reviewer was running their chips at 1.4-1,425v all-core. Some of those chips died within the week...
Be cautious, verify things with other sources.
 
Glad you got it sorted. I have read about an EDC bug. If you set it to 1. Supposed to be available on 3000-series and possibly with older BIOS-versions.
...
It was originally EDC=0; completely disabling it. But setting it that low (or =1) never worked for my 3700X: it would never boost or simply run at a base clock of 3600Mhz. It's tricky to get dialed in right since even setting it a bit high makes it appear to work for all-core workloads but not for single threaded workloads. It's behaviour did change with later BIOS's/AGESA/SMU releases: it would originally only work well at EDC=10 for me, now the sweet spot is at EDC=5.

You also have to disable Advanced C States so it will never drop into C6 deep sleep. While the cores do idle with a tad more average power consumption (maybe 2 or 3W according to HWInfo) I don't really think it a problem. But for AMD, the entire world-wide installed base of Zen 3 CPU's "wasting" 2 watts can be a lot when it doesn't have to. Ryzen runs so efficiently all the time even without C6.

In the end: I agree it's not for everyone. But for those of us who enjoy tweaking with our computers it's a novelty and source of endless hours of enjoyment. Or wasted life. Whatever.
 

chaos_viper15

Distinguished
Feb 8, 2010
171
1
18,695
1
Glad you got it sorted. I have read about an EDC bug. If you set it to 1. Supposed to be available on 3000-series and possibly with older BIOS-versions. Look into that if interested.
For the pump speed, I always run mine at 100%. I can't hear it. My fans max out at 2300 rpm but they idle around 1400-1500. Not that noisy. Besides, when I play games, the fans on Nvidia 2080 are louder than radiator fans anyway. And I have headphones on.
But your use-case might be different. It's your PC, you can do whatever you want. That's the beauty of it.
When it comes to overclocking, you can look at guides and other peoples settings. But I do it just to get a set of ranges I should test. Maybe PPT between 80 and 200. EDC at 1-150. Voltage at 1.25-1.325v perhaps. These are just numbers I made up.

Every chip is different, so is yours.
Do not copy other peoples settings. Those work on their chip only. And my question is always: How well did they test the system? Are they going to come back in 6 months and say they have experienced instability and getting rid of the overclock solved it? I was just reading about exactly that yesterday on Reddit and a 3900X. Older post but anyway. These things happen.
It's like when Ryzen 5000-series launched and just about every Youtube hardware-reviewer was running their chips at 1.4-1,425v all-core. Some of those chips died within the week...
Be cautious, verify things with other sources.
Oh yeah, I read so much on reddit, linustechtips, watching videos, listening to videos, load of just googling "best" settings for my chip. So I know what you mean. Some people got a late 2020 chip getting that 4.4Ghz all core. but I'm happy with the settings I finally landed on. I was gonna drop voltage manually but was unsure and was reading AMD roberts article post about Zen2 when it launched.
It was originally EDC=0; completely disabling it. But setting it that low (or =1) never worked for my 3700X: it would never boost or simply run at a base clock of 3600Mhz. It's tricky to get dialed in right since even setting it a bit high makes it appear to work for all-core workloads but not for single threaded workloads. It's behaviour did change with later BIOS's/AGESA/SMU releases: it would originally only work well at EDC=10 for me, now the sweet spot is at EDC=5.

You also have to disable Advanced C States so it will never drop into C6 deep sleep. While the cores do idle with a tad more average power consumption (maybe 2 or 3W according to HWInfo) I don't really think it a problem. But for AMD, the entire world-wide installed base of Zen 3 CPU's "wasting" 2 watts can be a lot when it doesn't have to. Ryzen runs so efficiently all the time even without C6.

In the end: I agree it's not for everyone. But for those of us who enjoy tweaking with our computers it's a novelty and source of endless hours of enjoyment. Or wasted life. Whatever.
Absolutely love tweaking my PC, been messing with bios & teardowns of laptops and desktops since I was 14 (28 now) just always interested me. Sure it runs but would if it could run better? HAH. So enjoyment for me. Already wasted to much life in other aspects. Ya=eah I did disable c-statess & as for EDC seems like 15 is my sweet spot for my 3900x. I just set PPT/EDC to auto still I think it best with a set voltage offset or manual but I still have decent clocks. So I'm done till AM5 2nd gen or I REALLY need an upgrade.
 
Oh yeah, I read so much on reddit, linustechtips, watching videos, listening to videos, load of just googling "best" settings for my chip. So I know what you mean. Some people got a late 2020 chip getting that 4.4Ghz all core. but I'm happy with the settings I finally landed on. I was gonna drop voltage manually but was unsure and was reading AMD roberts article post about Zen2 when it launched.

Absolutely love tweaking my PC, been messing with bios & teardowns of laptops and desktops since I was 14 (28 now) just always interested me. Sure it runs but would if it could run better? HAH. So enjoyment for me. Already wasted to much life in other aspects. Ya=eah I did disable c-statess & as for EDC seems like 15 is my sweet spot for my 3900x. I just set PPT/EDC to auto still I think it best with a set voltage offset or manual but I still have decent clocks. So I'm done till AM5 2nd gen or I REALLY need an upgrade.
Oh yeah, I read so much on reddit, linustechtips, watching videos, listening to videos, load of just googling "best" settings for my chip. So I know what you mean. Some people got a late 2020 chip getting that 4.4Ghz all core. but I'm happy with the settings I finally landed on. I was gonna drop voltage manually but was unsure and was reading AMD roberts article post about Zen2 when it launched.

Absolutely love tweaking my PC, been messing with bios & teardowns of laptops and desktops since I was 14 (28 now) just always interested me. Sure it runs but would if it could run better? HAH. So enjoyment for me. Already wasted to much life in other aspects. Ya=eah I did disable c-statess & as for EDC seems like 15 is my sweet spot for my 3900x. I just set PPT/EDC to auto still I think it best with a set voltage offset or manual but I still have decent clocks. So I'm done till AM5 2nd gen or I REALLY need an upgrade.
I've been overclocking CPU since 2000. GPU the same. Just never tried to OC RAM much. These days, CPU and GPU OC is so fast and easy, it takes me a day or two. But of course around a week to know if it is stable. RAM is where I found my fun. Overclocked all 3 kits of my DDR4 RAM, all different chips. I spent over a year, possibly 2 years. Lot to learn, a LOT of variables to test. Some of them motherboard-specific.Extreme overclockers are a font of knowledge. I still learn something about DDR4 just by reading the OC subreddit.
I've had 2 AM4 motherboards too, the old B350 mobo died on me because I flashed it so much. 1st gen Ryzens RAM support wasn't great initially, then it peaked and a year after that, old RAM OCs would no longer be stable for me. So I must have flashed BIOS at least 20 times. Eventually it bricked.
I have a 5600X now, on X470. I could do the per-core overclock to get every last Mhz out of the CPU but I don't find that fun, just tedious, for possibly very little gain. Corecycler, OCCT, testing each core. Hoping CPU wont crash when it idles. I played a lot with Fclock, trying to find a stable setting over 1900 Mhz. No such luck. Just wont budge, no matter the voltages or settings, spits out WHEA errors. Only difference was, how many per time unit. I've tried maybe 100-1000 combinations, looking at what others have run successfully, on same mobo, same or similar RAM chips, Micron.
 

ASK THE COMMUNITY