Manual clocking the 3900x

[revv]

I'm not much of an overclocker, but my system was running hot with stock settings (1.4v), so I started to do some research and found a detailed post on what to do to manually set clock speed and voltage.

Right now I'm at 1.275 volts and seeing stable 4.2 Ghz speeds with temperatures not exceeding 65C. Cinebench R20 is coming in at 7351.

I can only achieve this with Ryzen Master and attempting to set the BIOS to these setting make the system unstable, so I have to manually start Ryzen Master every time I start my computer and select the profile that I want to use.

Do these numbers look decent? Is there anything I can do to improve these speeds? Also, I'm assuming that I'm loosing boost speeds with these settings, correct? I was never able to hit 4.6 Ghz anyway and was hitting 4.4 Ghz for moments at a time.

Pictures:
(Ryzen Master pictures show settings and status while running Cinebench test)

My system: Aorus X570 Ultra, Ryzen 3900x, G Skill 32 GB @ 3600 Mhz, Aorus RTX 2070S, Corsair H100i v2

 

[Mchart]

Seems like a lot of effort when you could leave the settings default and the thing should be all core boosting to 4.2ghz anyways with far better single core boost speed as well.

 

[revv]

I guess the only advantage is to keep voltage under 1.3v at all times, instead of letting it ride at the 1.4v level, where it seems to be at every time I check.

Very new at this stuff, so I'm not sure if I'm thinking about this correctly.

 

[Mchart]

There is nothing wrong with it going to 1.4~ volts for lighter loads to hit those higher boost clocks. It's only an issue if it's doing that for an all-core load.

You need to trust in AMD's implementation and let the chip do what it needs to do. This has been covered to death, but manually overclocking this generation is basically a waste of time outside of certain edge-cases.

Basically, you're overthinking it. Just focus on the RAM settings and leave the CPU default.

 

[bal3wolf]

i been in same boat i can do a all core clock of 4300Mhz with 1.23-1.26 but i played with pbo bug some today got it to mostly hold 4275-4350 on all core and hit 4700 on limited cores and was able to get its max volts down to low 1.444 gonna let it ride this way for a few days see how i like it. Only thing i dont like is it gets hotter running a pbo overclock then a straight all core overclock.

 

[kamikazi]

MY 3900x would never boost above 4.0 all core stock in CB20. But, I did some extensive testing with manual clocking and voltage. Mainly, I was testing different voltages at 4.25 GHZ, which was the sweet spot for that chip. It could do 5 minutes of CB20 below 1.2v using a 240mm AIO. https://hardforum.com/threads/binning-my-2-3900x-chips.1989775/post-1044482237

I just quoted the last post in the thread. If you want to read the whole thread, I was binning two 3900x chips against each other.

 

[bal3wolf]

it seems like newer chips are boosting/clocking much better then ones 2-3months ago from differnt threads i have seen. I just took my water block off a few times to see if i could get even better temps might been a mistake lol i tried some as5 this time tho and it has a breakin period.

 

[revv]

There is nothing wrong with it going to 1.4~ volts for lighter loads to hit those higher boost clocks. It's only an issue if it's doing that for an all-core load.

You need to trust in AMD's implementation and let the chip do what it needs to do. This has been covered to death, but manually overclocking this generation is basically a waste of time outside of certain edge-cases.

Basically, you're overthinking it. Just focus on the RAM settings and leave the CPU default.

This is my first time overclocking anything. For good reason. Up until now, I don't think I've had a processor that has numbers that act so weird. I've never really had to overclock as I was happy with the stock speeds and general working of the chip.

However, while Zen 2 are incredible processors, they seem to act kind of weird.

After I read your post, I put everything back to stock and ran Cine R20. During the test, these were the values:

Temp: 71C
Speed: 4.1 Ghz
Power Draw: 110 W
Volts: 1.38
Bench Score: 7021

Going back to the profile I created, I had this during the test:

Temp: 65C
Speed: 4.2 Ghz
Power Draw: 96W
Volts: 1.27
Bench Score: 7340

Everything about these settings seem better than stock, although admittedly I'm not an expert.

Do these numbers not tell a better story in your opinion?

 

[thesmokingman]

This is my first time overclocking anything. For good reason. Up until now, I don't think I've had a processor that has numbers that act so weird. I've never really had to overclock as I was happy with the stock speeds and general working of the chip.

However, while Zen 2 are incredible processors, they seem to act kind of weird.

After I read your post, I put everything back to stock and ran Cine R20. During the test, these were the values:

Temp: 71C
Speed: 4.1 Ghz
Power Draw: 110 W
Volts: 1.38
Bench Score: 7021

Going back to the profile I created, I had this during the test:

Temp: 65C
Speed: 4.2 Ghz
Power Draw: 96W
Volts: 1.27
Bench Score: 7340

Everything about these settings seem better than stock, although admittedly I'm not an expert.

Do these numbers not tell a better story in your opinion?

They look ok but the power draw vs perf seems off. You're looking at discrete clocks, not the real clocks or more accurately effective clocks. There is a small learning curve with Matisse chips and later releases. Thus I suggest you read up on the way Matisse, Zen 2 works. It is different from 99.9% of chips before it. I wrote a quick post on setting up monitoring these chips. The linked info will help you understand how the chips work. If you are on stock voltage, you can ignore voltage readings. As for overclocking, I don't really bother with it 24/7 that is. When I do overclock I use a ratio overclock, and skip allcores no point to that.

https://hardforum.com/threads/build...2-raid-render-monster.1990145/post-1044457505

 

[kamikazi]

This is my first time overclocking anything. For good reason. Up until now, I don't think I've had a processor that has numbers that act so weird. I've never really had to overclock as I was happy with the stock speeds and general working of the chip.

However, while Zen 2 are incredible processors, they seem to act kind of weird.

After I read your post, I put everything back to stock and ran Cine R20. During the test, these were the values:

Temp: 71C
Speed: 4.1 Ghz
Power Draw: 110 W
Volts: 1.38
Bench Score: 7021

Going back to the profile I created, I had this during the test:

Temp: 65C
Speed: 4.2 Ghz
Power Draw: 96W
Volts: 1.27
Bench Score: 7340

Everything about these settings seem better than stock, although admittedly I'm not an expert.

Do these numbers not tell a better story in your opinion?

Your settings are better than stock for heavily multi-threaded apps. Where they will fall down is with gaming. Your single core performance will be locked at 4.2 GHz.

Unfortunately, Ryzen throws more voltage than necessary at all core performance for the sake of being able to reach max clock speeds on individual cores. This creates more heat.

If someone could create a UEFI that adapted in real time to changing workloads between heavily threaded and lightly threaded in a better way, that would be nice, but I have a feeling the juice isn't worth the squeeze. The current implementation is the best compromise from AMD and it's pretty good.

You should look into overclocking by CCX or CCD to try and get the best of both worlds. Have one part of your chip locked to an all core overclock and let AMD do what they want with the rest.

 

[thesmokingman]

If someone could create a UEFI that adapted in real time to changing workloads between heavily threaded and lightly threaded in a better way, that would be nice, but I have a feeling the juice isn't worth the squeeze.

Dude, it's not actually running 1.5v. You need to use hwinfo to see what the actual voltage is under core voltage and not VID. The voltage that is freaking ppl out is a misnomer.

 

[Mchart]

This is my first time overclocking anything. For good reason. Up until now, I don't think I've had a processor that has numbers that act so weird. I've never really had to overclock as I was happy with the stock speeds and general working of the chip.

However, while Zen 2 are incredible processors, they seem to act kind of weird.

After I read your post, I put everything back to stock and ran Cine R20. During the test, these were the values:

Temp: 71C
Speed: 4.1 Ghz
Power Draw: 110 W
Volts: 1.38
Bench Score: 7021

Going back to the profile I created, I had this during the test:

Temp: 65C
Speed: 4.2 Ghz
Power Draw: 96W
Volts: 1.27
Bench Score: 7340

Everything about these settings seem better than stock, although admittedly I'm not an expert.

Do these numbers not tell a better story in your opinion?

That seems off to me, on my board I literally can't manually overclock any better for all-core unless I go way out of my way to do CCX/CCD specific stuff. Typically without even touching anything you should be seeing 4.2ghz all core, but I guess that really depends on the motherboard/cooling.

Make sure you don't have PBO enabled or any other weird motherboard vendor 'OC Mode' type of things enabled.

For reference with my 3900x I see 4.2~ghz all core, 4.5+~ghz single core with all default settings relating to the CPU. I have a similar style cooler and see around 70c heatsoaked full load all core.

 

[bal3wolf]

my 3900x/asus board took a ton of tweaking to get a decent pbo overclock playing farcry 5 i see 4375-4400mhz it uses 4-5 cores most of the time. And i had a wierd bug where if i did a all-core overclock my sata speeds slowed down so to get around that i did the ccx overclock in bios just used same vcore and that worked great.

 

[somebrains]

I hope you guys have a better clocking experience with 3900xt.

 

[kamikazi]

Dude, it's not actually running 1.5v. You need to use hwinfo to see what the actual voltage is under core voltage and not VID. The voltage that is freaking ppl out is a misnomer.

I wasn't implying it was running 1.5v, I was pointing out that the voltage selection baked into UEFI for multi-threaded workloads will never be as efficient as something you dial in yourself. It’s going to throw more voltage in to ensure stability across a broad range of chips.

 

[kamikazi]

That seems off to me, on my board I literally can't manually overclock any better for all-core unless I go way out of my way to do CCX/CCD specific stuff. Typically without even touching anything you should be seeing 4.2ghz all core, but I guess that really depends on the motherboard/cooling.

Make sure you don't have PBO enabled or any other weird motherboard vendor 'OC Mode' type of things enabled.

For reference with my 3900x I see 4.2~ghz all core, 4.5+~ghz single core with all default settings relating to the CPU. I have a similar style cooler and see around 70c heatsoaked full load all core.

In what workload? Certainly not something like CB20?

 

[kamikazi]

my 3900x/asus board took a ton of tweaking to get a decent pbo overclock playing farcry 5 i see 4375-4400mhz it uses 4-5 cores most of the time. And i had a wierd bug where if i did a all-core overclock my sata speeds slowed down so to get around that i did the ccx overclock in bios just used same vcore and that worked great.

SOC above 1.1 will slow down SATA speeds.

 

[thesmokingman]

I wasn't implying it was running 1.5v, I was pointing out that the voltage selection baked into UEFI for multi-threaded workloads will never be as efficient as something you dial in yourself. It’s going to throw more voltage in to ensure stability across a broad range of chips.

You're completely wrong. Allcore boost will only run the voltage it needs, and that situation is a high current load which is around 1.325v maximum. My 3900x for example runs up to 1.25v for a 3900mhz allcore boost on stock. How is that too much voltage??? Whenever you see 1.4v+ under vcore, using hwinfo, you can then see which CCD is heated up. That will tell you whether its single core or not. As BOTH CCD will never be heated up when vcore is showing more than 1.4v. Whenever the chip shows 1.4v+ it is in a low current load, ie. single thread/core.

Also this suggestion that it's one setting for all is bogus as heck. The algorithms that AMD has developed are not one size fits all. Every chip will have it's own unique behavior driven by the chips thousand sensors.

 

[Mchart]

You're completely wrong. Allcore boost will only run the voltage it needs, and that situation is a high current load which is around 1.325v maximum. My 3900x for example runs up to 1.25v for a 3900mhz allcore boost on stock. How is that too much voltage??? Whenever you see 1.4v+ under vcore, using hwinfo, you can then see which CCD is heated up. That will tell you whether its single core or not. As BOTH CCD will never be heated up when vcore is showing more than 1.4v. Whenever the chip shows 1.4v+ it is in a low current load, ie. single thread/core.

Also this suggestion that it's one setting for all is bogus as heck. The algorithms that AMD has developed are not one size fits all. Every chip will have it's own unique behavior driven by the chips thousand sensors.

So much wrong with this statement, but i'll leave it alone beyond this post.

 

[kamikazi]

I'll drop this here if anyone has an hour to kill...watch it on 1.25 speed.

Buildzoid Safe Static Voltage on Ryzen 3000 Series

thesmokingman Thanks for the links about monitoring with HWINFO and RTSS. I'm going to set that up tonight.

 

[bal3wolf]

Yea i watched that thats why i started using more vdroop and did the fit test to see how much my cpu wants which is 1.27-1.29 at stock under prime95 smallfft

 

[thesmokingman]

I'll drop this here if anyone has an hour to kill...watch it on 1.25 speed.

Buildzoid Safe Static Voltage on Ryzen 3000 Series

thesmokingman Thanks for the links about monitoring with HWINFO and RTSS. I'm going to set that up tonight.

That guy rambles on too long and I'm not sure he got it either. If you want the safe voltage for your particular cpu, it was explained how to get there by the Stilt a ways back. Here's a reddit post that encapsulates the important parts.

So basically this means the maximum FIT voltage is around 1.325V (not 1.325V, around 1.325V) under high current scenarios but this maximum FIT voltage, depends on the silicon characteristics. Meaning chips with different silicon characteristics will have different voltage tolerance meaning one voltage can not be the maximum safe voltage for all chips.

^^see AMD's boost/voltage algorithms adjusting to silicon in effect.

So yeah, 1.325V is not safe. The maximum safe voltage for your personal zen 2 cpu is... I have no clue. If you want to be able to find this voltage turn on PBO and max PPT, TDC and EDC and run a worse-case workload as of The Stilts recommendation.

The above test will give your specific chips maximum safe FIT voltage. Also, be aware that Stilt's degraded his chip at 1.33v in a month with high temps.

 

[bal3wolf]

we dont even really know if pbo overclocking wont degrade the chips or how fast or slow it does it i tested my chip by loading defauts then loading it down with prime 95 smallfft and it still gave the chip 1.29 so thats my limit on how much i will ever send to it under heavy loads if im not pbo overclocking. And ryzen 4k could be a whole new animal to tame and figure out whats safe.

 

[thesmokingman]

we dont even really know if pbo overclocking wont degrade the chips or how fast or slow it does it i tested my chip by loading defauts then loading it down with prime 95 smallfft and it still gave the chip 1.29 so thats my limit on how much i will ever send to it under heavy loads if im not pbo overclocking. And ryzen 4k could be a whole new animal to tame and figure out whats safe.

That's true about PBO, concur. I was going to add a caveat about using PBO jacked up long term but figured no one is doing that. It shouldn't do any harm for as a test for the purpose of finding your silicon's max FIT voltage for high current loads though, crosses fingers.

 

[revv]

I let HWInfo run for a few hours last night while I was doing things like gaming, editing, etc.

Here is what I have:

As you can see, I have two cores that hit 4.5 Ghz, two that hit 4.4 Ghz, 5 that hit 4.3 Ghz. The rest are lower than that. I was doing a ton of stuff, from gaming to editing, to rendering audio and video, etc. Would it be correct to assume that if I have these numbers for several hours of use, they could be the maximums of these cores?

You'll also see that average voltage applied throughout this time was just under 1.4 volts which my brain still says is a lot.

I did even more testing today, going from stock to the profile that I mentioned here (4.2 all core @ 1.275v). Here are my findings:

		Temp	Peak Speed	Peak CPU Power %	Average Volts	TDC%	EDC%	Cine Score
Stock	Idle	44		15.5	1.41	22	18
	Cine	72	4.1	115 (not a mistake)	1.37	96	100	7108
RM Profile (4.2 @1.275v)	Idle	42		18	1.09	3	4
	Cine	66	4.2	95.4	1.275	17	23	7307

As you can see, I have small gains both in peak speed as well as Cine Score. However, these small gains are happening with lower temperatures when running Cinebench, likely due to the big difference in volts being sent to the CPU.

So, for everyday computing it seems like this profile would likely be a good balance between speed, power, and temperature, thus fan quietness.

My next step would be to look into overclocking by CCD or CCX, particularly to get better speeds for gaming. A few days ago, I was playing Red Dead Redemption 2 and streaming to Twitch via OBS and my fans were running at either full speed or very close to it, The case was very loud, so loud that I could hear it through my headphones while playing the game although I use open back phones.

I have to add that I have never seen the current speed at 4.5 or 4.4 Ghz. If these cores are hitting those speeds, they are for very brief moments and they do not do it frequently. Taking this into consideration, would it make that much of a real-world difference if I managed to clock some of these cores at 4.5 Ghz or 4.4 Ghz via CCD or CCX manual speed settings? We're only talking about 7% here...

 

[KarVi71]

You cant trust HWinfo 64 completely, as it cant detect cores that are put to sleep.

See this comparison of Ryzen Master vs HWinfo64 on my almost completely idle 3900X. HWinfo shows much higher clocks. HWinfo does not know the core has been powered down, so it'l show the last active clock.
The lower clocks on CCD0 are (AFAIK) cores that have been partially idle / sleeping during the sample period, that gives an average clock for the period.

Dont get me wrong, HWinfo is a very good program, and I use it a lot myself.

You just cant trust it completely as a tool for this kind of analysis.

Otherwise I have no comment to your endeavour about manually clocking the 3900x, I'm certain you can strike a good balance.
I leave my 3900x on auto, and trust AMD to know what they are doing with the clocks and voltages.

 

[thesmokingman]

You cant trust HWinfo 64 completely, as it cant detect cores that are put to sleep.

See this comparison of Ryzen Master vs HWinfo64 on my almost completely idle 3900X. HWinfo shows much higher clocks. HWinfo does not know the core has been powered down, so it'l show the last active clock.
The lower clocks on CCD0 are (AFAIK) cores that have been partially idle / sleeping during the sample period, that gives an average clock for the period.

View attachment 248326

Dont get me wrong, HWinfo is a very good program, and I use it a lot myself.

You just cant trust it completely as a tool for this kind of analysis.

Otherwise I have no comment to your endeavour about manually clocking the 3900x, I'm certain you can strike a good balance.
I leave my 3900x on auto, and trust AMD to know what they are doing with the clocks and voltages.

FYI... You need to read the link I posted and then you'd learn that RM's algorithms are propriety and that AMD will not share how they calculate what they calculate, even though they worked with HWINFO.

On Zen2 (Matisse) systems this method can provide results closer to Ryzen Master (RM) per-core clock values, especially because it respects sleeping cores. It is assumed that such cores are marked as sleeping by RM when the effective (average) clock is below a certain threshold (somewhere around 50 MHz and below). Please note, that RM uses a different (proprietary) technique to measure clocks, so there might be some differences between the effective clock in HWiNFO and RM. While we work with AMD on the best way to access more accurate data to measure clock and voltage values, this remains the only method. Additionally, the current effective clock method is an architectural feature meaning that it doesn't depend on a certain CPU model, but is rather universal across a broad range of CPU families.

And regarding sleep states, you just ignored that hole part. Scroll down.

Another disadvantage is that cores in modern CPUs that have no workload are being suspended (lower C-States). In such case when software attempts to poll their status, it will wake them up briefly and thus the clock obtained doesn't respect the sleeping state.

Hence a new approach needs to be used called the Effective clock. This method relies on hardware's capability to sample the actual clock state (all its levels) across a certain interval, including sleeping (halted) states. The software then queries the counter over a specific polling period, which provides the average value of all clock states that occurred in the given interval. HWiNFO v6.13-3955 Beta introduces reporting of this clock.

 

[thesmokingman]

I let HWInfo run for a few hours last night while I was doing things like gaming, editing, etc.

Here is what I have:

As you can see, I have two cores that hit 4.5 Ghz, two that hit 4.4 Ghz, 5 that hit 4.3 Ghz. The rest are lower than that. I was doing a ton of stuff, from gaming to editing, to rendering audio and video, etc. Would it be correct to assume that if I have these numbers for several hours of use, they could be the maximums of these cores?

You'll also see that average voltage applied throughout this time was just under 1.4 volts which my brain still says is a lot.

I did even more testing today, going from stock to the profile that I mentioned here (4.2 all core @ 1.275v). Here are my findings:

		Temp	Peak Speed	Peak CPU Power %	Average Volts	TDC%	EDC%	Cine Score
Stock	Idle	44		15.5	1.41	22	18
	Cine	72	4.1	115 (not a mistake)	1.37	96	100	7108
RM Profile (4.2 @1.275v)	Idle	42		18	1.09	3	4
	Cine	66	4.2	95.4	1.275	17	23	7307

As you can see, I have small gains both in peak speed as well as Cine Score. However, these small gains are happening with lower temperatures when running Cinebench, likely due to the big difference in volts being sent to the CPU.

So, for everyday computing it seems like this profile would likely be a good balance between speed, power, and temperature, thus fan quietness.

My next step would be to look into overclocking by CCD or CCX, particularly to get better speeds for gaming. A few days ago, I was playing Red Dead Redemption 2 and streaming to Twitch via OBS and my fans were running at either full speed or very close to it, The case was very loud, so loud that I could hear it through my headphones while playing the game although I use open back phones.

I have to add that I have never seen the current speed at 4.5 or 4.4 Ghz. If these cores are hitting those speeds, they are for very brief moments and they do not do it frequently. Taking this into consideration, would it make that much of a real-world difference if I managed to clock some of these cores at 4.5 Ghz or 4.4 Ghz via CCD or CCX manual speed settings? We're only talking about 7% here...

As you can see, I have two cores that hit 4.5 Ghz, two that hit 4.4 Ghz, 5 that hit 4.3 Ghz. The rest are lower than that. I was doing a ton of stuff, from gaming to editing, to rendering audio and video, etc. Would it be correct to assume that if I have these numbers for several hours of use, they could be the maximums of these cores?

No, this is stock behavior, there is no point to watching the voltage threshold for this. When you are ascertaining the maximum FIT voltage for high current load, it is a specific process of maxing PBO and letting it do its thing. We are also not using Cinebench, as Cionebench is NOT hard to run, it does not even pass as a stability test. I run Prime which loads the hell out of the cpu. Then watch your voltage.

You'll also see that average voltage applied throughout this time was just under 1.4 volts which my brain still says is a lot.

Sorry, you are looking at it wrong. When using hwinfo the average volts is not a good indicator of volts at all. When the cpu raises volts to the 1.45v-1.5v region, it happens for milliseconds in everyday use. This will skew average volts. You also cannot compare an allcore setting with stock which is what you are doing. The 4.2ghz allcore will be faster than the base clock.

As you can see, I have small gains both in peak speed as well as Cine Score. However, these small gains are happening with lower temperatures when running Cinebench, likely due to the big difference in volts being sent to the CPU.

So, for everyday computing it seems like this profile would likely be a good balance between speed, power, and temperature, thus fan quietness.

Wrong again, as you are running an overclock with RM with a lower set voltage that will keep your chip from hitting the much needed single core boost. People, you all need to understand the voltage reporting is limited because it cannot show who where the voltage is going. If you are lucky enough to have a second CCD you can then see if the voltage being shown is actually going to the whole chip or the CCD that is currently running a low current load, ie. single core/single thread. This is why the average volts is useless to us except in cases where you reset the counter to get the average within a specific time, like when you are benching. For everyday computing you are losing single core/thread boosting to 4.6ghz so that to me is not ideal. Allcore overclocks are ideal for everyday use if everyday you spend all your time in Cinebench, otherwise stock is an obvious good compromise. If you want to maximize perf period, then a ratio overclock will net you the performance of an allcore while keeping one CCX higher clocked to still get that precious single core/thread boosting. For ex... note the multi vs single scoring.

You guys should consider running a ratio overclock. Generally speaking the first chiplet can run 4.5ghz at 1.325v or there abouts. Thus running an allcore at 4.3 means you lose out on that precious single core boost. Setting a ratio of 4.5 for the first CCX and 4.3 for the other CCX, gets you that allcore performance you want and a good compromise for single thread performance.

For example 4.3 all core

View attachment 247820

ratio overclock 4.45 to 4.3

View attachment 247821

For example last Dec I built a serious workstation with the 3970x. I could have run it at 4.1ghz to 4.2ghz but then we'd lose the precious single core boosting. That is seriously important even in a heavy crunching machine like that because single core boost is massively crucial in the prep stages of an app like Adobe before you get to the rendering stage, ie. many parts of AfterEffects for example are single threaded before you get to rendering.

 

[JustinC]

I would just leave the settings default and work with 4.2 Ghz of speed.
No need to control the chip to give you a speed of 4.4 Ghz when at that speed the life of the chip degrades.

 

[revv]

I would just leave the settings default and work with 4.2 Ghz of speed.
No need to control the chip to give you a speed of 4.4 Ghz when at that speed the life of the chip degrades.

So, at default, it is sending 1.4v to the chip, especially when idling. That's why I'm looking into this mess. If I tune it and go down to 1.275, I can get all cores up to 4.2 Ghz, which reduces all temps as well.

thesmokingman Would you say that Ryzen Master voltage readings suffer from the same problem as HWinfo? The reason why I ask is because I see current/average at 1.4v, even when idle.

Realistically, I'm not hitting 4.6 Ghz at all, so I could say that my max is 4.5 Ghz when boosting and this is only on two cores and for moments at a time. I cannot get the cores to sustain these speeds with anything: Cinebench, Prime95, Red Dead Redemption 2, Premiere, you name it.

 

[Gamerdad]

Lots of good info from an AMD SME in this reddit post. It helped me when I first got my Ryzen 3700X. Here's a quote: "Please note that it is totally normal for your Ryzen to use voltages in a range of 0.200V - 1.500V -- this is the factory operating range of the CPU. It is also totally normal for the temperature to cycle through 10°C swings as boost comes on and off. You will always see these characteristics, as they're intended, so do not be surprised to see such values. "

Edit: forgot the link

 

[thesmokingman]

So, at default, it is sending 1.4v to the chip, especially when idling. That's why I'm looking into this mess. If I tune it and go down to 1.275, I can get all cores up to 4.2 Ghz, which reduces all temps as well.

thesmokingman Would you say that Ryzen Master voltage readings suffer from the same problem as HWinfo? The reason why I ask is because I see current/average at 1.4v, even when idle.

Realistically, I'm not hitting 4.6 Ghz at all, so I could say that my max is 4.5 Ghz when boosting and this is only on two cores and for moments at a time. I cannot get the cores to sustain these speeds with anything: Cinebench, Prime95, Red Dead Redemption 2, Premiere, you name it.

First, read the link from post 9 so you understand how Matisse works. Just by saying you see 1.4v average means you are misunderstanding the average voltage. Average voltage is USELESS to us EXCEPT when we are measuring within a specific timeframe, like for ex. the average voltage thru a benchmark run. And to make that happen you hit the reset button on hwinfo, run bench, check voltage, etc. Otherwise, average voltage in HWINFO is again useless. To understand this ya have to watch the actual core voltage or die average, package, core vid effective etc etc. The voltage readings of ALL monitors are inherently limited when it comes to Matisse because they cannot differentiate between 1.4v+ going to a single core vs the whole die. When a single core is boosting, it will call for 1.4v+ however that voltage is actually only going to the specific core that called for that voltage. The problem as I wrote is that monitoring apps see that voltage as going to the whole chip instead of which half of the chip it is really going to. Hence the voltage yall are seeing is not exactly accurate and why average voltage is USELESS.

If you are going to watch the voltage with a keen eye, you have to put it in context with the CCDs in the case of a 3900x or 3950x. When you see die average voltage at 1.4v+, you then have to take a look at the CCD temperature. CCD idle temperatures will never be mistake for loaded temperatures right? This is the important part, you will never see 1.4v+ being shown at the same time that both CCDs are heated up. This is because the chip only allows a single core to use the maximum voltage. To put it plainly you have two CCD. when one is heated up and it shows 1.4v+ that means it is only running a single core. That's how you verify what the real voltage is. Owners of single CCD chips don't have this luxury so I suspect they will always be freaking out if they don't understand how the voltage works.

As for your overclock, you are overclocking and limiting volts to 1.275v. Yea its going to run cooler because you are not allowing the chip to boost up and if you are running 4.2ghz you are running 300mhz above base clock. Thus ofc it is faster, BUT you are losing 400mhz on single core boost. Again read the links I posted, use hwinfo and monitor the right temp/voltage sensors. Read the link by the hwinfo developer and "effective clocks" vs "discrete clocks" and basically ignore core VID and core clock. Core clock is the most misleading stat as is core VID. The core VID is what is available to the chip, not what it is actually using. You have to scroll allllll the way down to "CPU core VID (effective)" to get the accurate voltage reading. And when doing so put it in context of which CCD is using said voltage. When both CCDs are loaded voltage will be below the maximum FIT voltage for high current loads. High current loads = a real sitaution where all cores are loaded.

As for Ryzen Master, I do not bother installing it. I've no use for it especially since it cannot overlay, so it's useless to me. Afterburner is a powerful too, even if you are not monitoring with it. You can use AB to check/confirm which apps are using HW acceleration. And then use that info copy/pasta into RTSS to create profiles to disallow X app from using HW acceleration, etc etc. Afterburner, RTSS, and HWINFO go hand in hand.

Lots of good info from an AMD SME in this reddit post. It helped me when I first got my Ryzen 3700X. Here's a quote: "Please note that it is totally normal for your Ryzen to use voltages in a range of 0.200V - 1.500V -- this is the factory operating range of the CPU. It is also totally normal for the temperature to cycle through 10°C swings as boost comes on and off. You will always see these characteristics, as they're intended, so do not be surprised to see such values. "

Edit: forgot the link

They only touch on the general details and not the nitty gritty unfortunately. AMD really should have written some whitepapers on how to effectively monitor and read the stats from the cpu. And they should freaking release their algorithm for clock speed calculations, geeze!