Definitive guide to configuring the Ryzen 3900X/3950X and all other 3000 Series CPUs

[thesmokingman]

I have no disclaimers to spare AMD, but still fellow members put up a challenge when I have to quote him. That is a problem. No one believes it when it is in his written statement that precision boost overdrive is for threadripper, alone.


Yes, but when all is said and done, his reputation doesn't have a following.

I just ignored the OP's silly rant about marketing. Just shows how this is not anything definitive.

 

[TheoneandonlyMrK]

I have no disclaimers to spare AMD, but still fellow members put up a challenge when I have to quote him. That is a problem. No one believes it when it is in his written statement that precision boost overdrive is for threadripper, alone.

Your hard er to follow then even me.

Who are you on about and why.

I'm putting words in no mouths.

Quote who you want including me but at least get the context right.

Try and actually understand what you're quoting first too.

I quoted the Op, but relative to this whole thing ,not Robert hallock who is totally irrelevant here imho

 

[mtcn77]

I just ignored the OP's silly rant about marketing. Just shows how this is not anything definitive.

Well, you were scolding him(op)/,or me doesn't matter, for not referencing him(robert hallock). You don't get to turn up with a consecutive marketing dissuasion.

 

[thesmokingman]

Well, you were scolding him(op)/,or me doesn't matter, for not referencing him(robert hallock). You don't get to turn up with a consecutive marketing dissuasion.

Huh, what?

 

[Michael Nager]

So we should blindly follow you, over the hundreds of engineers and scientists, interesting.

What are you offering besides one perspective as proof you know better, not much as far as I can see.

Certainly not more than the qualified people who designed and built it IMHO.

You dodge valid questions, then bemoan AMD for being vague, hypocrisy.

How many of those "hundreds of engineers and scientists" at AMD have you personally spoken to?

Let me hazard a wild guess, that would be none, zip, zero, nada.

The only thing you have heard at best is from some talking heads, salescritters and marketdroids at AMD.

What I am offering is the result of eight and a half months of intensive trial and error experimentation - because there sure as hell wasn't any good advice coming from either the Tech Media or Tech YouTubers. These guys essentially get their shit for free and don't give a flying one at a rolling doughnut how they configure the system, as long as they get their benchmarks and when overclocking they do it in the laziest possible way, with no regard whatsoever for the longevity of the CPU.

Hardware Unboxed for instance managed to fry their 3900X - confidence inspiring performance there.

Let's just say I have been underwhelmed.

The likes of Buildzoid or Steve from Gamers Nexus have their focus on LN2 overclocking than on 24/7 stable performance (my systems run 24/7 whereby my previous main machine ran 24/7 for 1,900 days or 5.2 years exactly before I decommissioned it in favour of my current main machine, which has been running 24/7 with a mild overclock for just over 1,750 days or4.8 years).

That being said, Steve from Gamers Nexus has come closest to getting the results I have even though I have a higher ambient temperature (7-8 °C higher depending on which day I run the tests) than he was running his benchmarks at, and even then, he didn't quite manage to get a beat my scores with non LN2 overclocking - whereby I am not overclocking. Buildzoid in his 3950X overclocking videos didn't even come close.

Now admittedly, doing it my way, I take a bit of a hit on the single core score on CineBench as you can see from one of my CineBench R20 runs in the picture.

My focus from the start was not to get the best benchmark scores, but rather to see what performance I could tickle out of first the 3600X and then the 3950X for 24/7 usage. The fact that I have gotten the best benchmark scores outside of LN2 overclocking has been an added bonus.

There are no questions that I have dodged. I might have ignored stupid statements couched as questions, but I do answer sincerely meant inquiries with an answer commensurate with the best of my knowledge - and when I don't know something then I have no problem whatsoever in saying those three words, "I don't know".

 

[mtcn77]

Don't pay any notice, it is their misfortune that they don't make a personal assessment any more. It is all what he says, she says...

 

[TheoneandonlyMrK]

How many of those "hundreds of engineers and scientists" at AMD have you personally spoken to?

Let me hazard a wild guess, that would be none, zip, zero, nada.

The only thing you have heard at best is from some talking heads, salescritters and marketdroids at AMD.

What I am offering is the result of eight and a half months of intensive trial and error experimentation - because there sure as hell wasn't any good advice coming from either the Tech Media or Tech YouTubers. These guys essentially get their shit for free and don't give a flying one at a rolling doughnut how they configure the system, as long as they get their benchmarks and when overclocking they do it in the laziest possible way, with no regard whatsoever for the longevity of the CPU.

Hardware Unboxed for instance managed to fry their 3900X - confidence inspiring performance there.

Let's just say I have been underwhelmed.

The likes of Buildzoid or Steve from Gamers Nexus have their focus on LN2 overclocking than on 24/7 stable performance (my systems run 24/7 whereby my previous main machine ran 24/7 for 1,900 days or 5.2 years exactly before I decommissioned it in favour of my current main machine, which has been running 24/7 with a mild overclock for just over 1,750 days or4.8 years).

That being said, Steve from Gamers Nexus has come closest to getting the results I have even though I have a higher ambient temperature (7-8 °C higher depending on which day I run the tests) than he was running his benchmarks at, and even then, he didn't quite manage to get a beat my scores with non LN2 overclocking - whereby I am not overclocking. Buildzoid in his 3950X overclocking videos didn't even come close.

Now admittedly, doing it my way, I take a bit of a hit on the single core score on CineBench as you can see from one of my CineBench R20 runs in the picture.

My focus from the start was not to get the best benchmark scores, but rather to see what performance I could tickle out of first the 3600X and then the 3950X for 24/7 usage. The fact that I have gotten the best benchmark scores outside of LN2 overclocking has been an added bonus.

There are no questions that I have dodged. I might have ignored stupid statements couched as questions, but I do answer sincerely meant inquiries with an answer commensurate with the best of my knowledge - and when I don't know something then I have no problem whatsoever in saying those three words, "I don't know".

Let's just say I too am underwhelmed.

 

[moproblems99]

It might be that the reason why it isn't clocking that little bit higher is because of the voltage droop under load.

I set my LLC from 3 to 5 to lower the the voltage at load and sustain higher boost clocks.

But I suppose that static electricity which has a high voltage but no current couldn't possibly hurt a CPU.

Static electricity isn't 1.4v

 

[eidairaman1]

I set my LLC from 3 to 5 to lower the the voltage at load and sustain higher boost clocks.



Static electricity isn't 1.4v

ESD is real, Ive zapped ICs before, (sucks).

Voltage is EMF/Potential Difference/Pressure that gets electrons flowing in a conductor electrons are already in a conductor but without voltage they remain static, voltage induces dynamic flow lol

I just know that what Buildzoid did with the voltage I wont be doing on any cpu since this arch is way different than AM3+ or older.

 

[Michael Nager]

I set my LLC from 3 to 5 to lower the the voltage at load and sustain higher boost clocks.



Static electricity isn't 1.4v

I know that static electricity isn't 1.4 Volts, it was a reducto ad absurdum on the premise that it is current that causes damage to CPUs and not voltage.

The term "Boost Clock" is a bit of a sick joke as far as the Ryzen 3000 series is concerned.

Sure you have the feelgood experience when you look at something like HWMonitor and you see the maximum clockspeed at 4.55 or 4.6, but those maximums are only achieved when the system is NOT under load, and just as it is true that RGB adds at least 35 FPS to gaming, the real time when you need the system boosting clockspeed is when it is doing absolutely naff-all.

When you then put the system under load, such as with a CineBench R20 all core run, you are lucky if you get the CPU clocking at over 4.0 GHz, with a little dustcloud on the horizon with your "Boost" right in front of it, that is how far removed "Boost" is from actually doing anything.

 

[ratirt]

I know that static electricity isn't 1.4 Volts, it was a reducto ad absurdum on the premise that it is current that causes damage to CPUs and not voltage.

The term "Boost Clock" is a bit of a sick joke as far as the Ryzen 3000 series is concerned.

Sure you have the feelgood experience when you look at something like HWMonitor and you see the maximum clockspeed at 4.55 or 4.6, but those maximums are only achieved when the system is NOT under load, and just as it is true that RGB adds at least 35 FPS to gaming, the real time when you need the system boosting clockspeed is when it is doing absolutely naff-all.

When you then put the system under load, such as with a CineBench R20 all core run, you are lucky if you get the CPU clocking at over 4.0 GHz, with a little dustcloud on the horizon with your "Boost" right in front of it, that is how far removed "Boost" is from actually doing anything.

I need to chip in a bit. You have referenced that HWmonitor shows 4.55 or 4.6 and that is to please somebody's eyes but when running Cinebench R20 full load on all cores (despite if it is 3900x or 3950x or even 3700x ) it doesn't boost that high. Well there is a simple explanation for this and it does't not matter how much volts you apply to the processor. The advertised boost clock for All RYZENS works with a single core only which means it does not apply for all cores. It is obvious that when you have 3950x and you utilize all cores and threads (like Cinebench R20) it will not go 4.6Ghz all cores. Even if you OC the CPU and add voltage, the all core frequency will not be the boost speed, EVER. You may get a bit more when you optimize configuration of your set up but it wont be the advertised boost clock speed. That is not how Ryzen works. Blaming AMD for this is not fair.

 

[Michael Nager]

I need to chip in a bit. You have referenced that HWmonitor shows 4.55 or 4.6 and that is to please somebody's eyes but when running Cinebench R20 full load on all cores (despite if it is 3900x or 3950x or even 3700x ) it doesn't boost that high. Well there is a simple explanation for this and it does't not matter how much volts you apply to the processor. The advertised boost clock for All RYZENS works with a single core only which means it does not apply for all cores. It is obvious that when you have 3950x and you utilize all cores and threads (like Cinebench R20) it will not go 4.6Ghz all cores. Even if you OC the CPU and add voltage, the all core frequency will not be the boost speed, EVER. You may get a bit more when you optimize configuration of your set up but it wont be the advertised boost clock speed. That is not how Ryzen works. Blaming AMD for this is not fair.

There are three columns in HWMonitor "Value", "Min" and "Max" where "Value" is the current state, "Min" and "Max" should be self explanatory.

When looking at the columns you will see entries for "Max" under "Clocks" that the CPU core has boosted to at some point, no matter how short the duration or how infrequently and of course at a point where it was under bugger-all load. That "Max" entry is what a number of people have spoken to me about and said, "My system runs at 4.6 GHz", in the case of a 3900X for instance.

Configuring my 3950X my way, when running with SMT On (16 Cores/32 Threads) the CPU will run at 4.3 GHz on all cores at a lower temperature than running the system with PBO On (or even Off) and the clockspeed being somewhere between 4.0 and 4.1 GHz. The fact of the matter is that I don't really see much of a change in performance with PBO On or Off, except that with PBO on the temps are higher for just about the same result.

So that would be well over 200 MHz less performance per Core/Thread at a higher temp, a higher voltage and a higher power draw. So the voltage is higher, the current is higher and the temperature is higher for far less performance.

What I am blaming a portion of AMD for and that portion would be the the cognitively redacted MarketDroids and SalesCritters who have misled people about the performance of 3rd Gen Ryzen.

I would point to this video from AMD featuring Robert Hallock, which I mentioned before as a prime example:

Damn straight that I imagine the Techs at AMD not only would be, but were, completely pissed at this bald faced liar and what he is saying. Unfortunately the Techs don't get to talk to the public - not if they want to keep their jobs that is.

However what people don't realise is that my 3950X running at an all-core of 4.3 GHz is the equivalent of about 4.86 GHz all core of an Intel CPU due to the 13% higher IPC of the Ryzen 3000 series compared to Intel CPUs.

I think you were misunderstanding what I meant when I mentioned "4.55 or 4.6" and I hope this clears it up.

 

[ratirt]

There are three columns in HWMonitor "Value", "Min" and "Max" where "Value" is the current state, "Min" and "Max" should be self explanatory.

When looking at the columns you will see entries for "Max" under "Clocks" that the CPU core has boosted to at some point, no matter how short the duration or how infrequently and of course at a point where it was under bugger-all load. That "Max" entry is what a number of people have spoken to me about and said, "My system runs at 4.6 GHz", in the case of a 3900X for instance.

Configuring my 3950X my way, when running with SMT On (16 Cores/32 Threads) the CPU will run at 4.3 GHz on all cores at a lower temperature than running the system with PBO On (or even Off) and the clockspeed being somewhere between 4.0 and 4.1 GHz. The fact of the matter is that I don't really see much of a change in performance with PBO On or Off, except that with PBO on the temps are higher for just about the same result.

So that would be well over 200 MHz less performance per Core/Thread at a higher temp, a higher voltage and a higher power draw. So the voltage is higher, the current is higher and the temperature is higher for far less performance.

What I am blaming a portion of AMD for and that portion would be the the cognitively redacted MarketDroids and SalesCritters who have misled people about the performance of 3rd Gen Ryzen.

I would point to this video from AMD featuring Robert Hallock, which I mentioned before as a prime example:

Damn straight that I imagine the Techs at AMD not only would be, but were, completely pissed at this bald faced liar and what he is saying. Unfortunately the Techs don't get to talk to the public - not if they want to keep their jobs that is.

However what people don't realise is that my 3950X running at an all-core of 4.3 GHz is the equivalent of about 4.86 GHz all core of an Intel CPU due to the 13% higher IPC of the Ryzen 3000 series compared to Intel CPUs.

I think you were misunderstanding what I meant when I mentioned "4.55 or 4.6" and I hope this clears it up.

Thanks for the lecture btw.
Well as a matter of fact you didn't get what i said regarding your previous post. I was referring about the "boost is a sick joke".
(that is wrong) I don't think it is. It works perfectly fine and as intended. What I get from your post is that you are claiming the boost is shit and AMD scammed people off with it because it is not boosting as it should. When you tweak your settings you can achieve more and I've mentioned it in my post but I guess you have missed it.
BTW I have two 3970x threadrippers and of course they are boosting to 4.5 when not heavy taxed (single core utilization). With Cinebench on the other hand the boost is going up to 3.8-3.9 but I didn't tweak anything and my cooling is an air cooler Noctua noctua nh-u9 tr4-sp3 crap. If I had liquid the situation would have been way different. Your tweaks, btw, are software and that is not as it should have been (unless I read your earlier posts incorrectly). Does this mean if you boost 200Mhz more, that means the boost is shit? Of course not. Nobody misled anything. Go to the AMD webpage https://www.amd.com/en/products/cpu/amd-ryzen-9-3950x and go to the boost section. It says single core boost 4.7Ghz. The default is rather
So if you claim that you have 4.3 instead of 4 on all cores that is tweaking. Getting better cooling instead of stock cooler. You don't have a stock cooler do you?

Sometimes I don't understand you people. You get some performance boost through OC/tweaks like you Mr. and you claim AMD boost is shit because you can get more out of the CPU
When AMD maxes out the CPU it is still shit because there is no OC headroom. The default settings are for all the processors released like 3950x. The silicon may vary one from the other ( you mentioned that too "silicon lottery) remember? The specs are for all 3950x's not individual. All processors released will work that way (safe default way not harming the hardware) if you get a chance to get more out of your CPU then be happy you can't blame something that ain't there.

That is about it At least that is how I get your issue with the AMD boost.

 

[moproblems99]

When you then put the system under load, such as with a CineBench R20 all core run, you are lucky if you get the CPU clocking at over 4.0 GHz, with a little dustcloud on the horizon with your "Boost" right in front of it, that is how far removed "Boost" is from actually doing anything.

I boost to about 4.15 all-core m

 

[eidairaman1]

Blame Jim Keller for having the initial design too sensitive

 

[bug]

@Michael Nager How much more performance did you squeeze out of the CPU? Because it seems like a lot of steps to go through and the need for Ryzen Master means you're out of luck if not running Windows...

 

[TheoneandonlyMrK]

Blame Jim Keller for having the initial design too sensitive

Why Jim, no one here actually spoke to ANY of the engineering team that built zen or zen2.

So by the Op's count they don't exist? and are irrelevant , time and again he's inflated his opinion because ,man does he like spilling it in pages, and he's here to talk so by default he's the man and right and because we can't speak to AMD staff they and their decisions are irrelevant and untrusted.

Zens the product of a unicorn and a Smirth mating ,it wasn't designed ,fools.

 

[eidairaman1]

Why Jim, no one here actually spoke to ANY of the engineering team that built zen or zen2.

So by the Op's count they don't exist? and are irrelevant , time and again he's inflated his opinion because ,man does he like spilling it in pages, and he's here to talk so by default he's the man and right and because we can't speak to AMD staff they and their decisions are irrelevant and untrusted.

Zens the product of a unicorn and a Smirth mating ,it wasn't designed ,fools.

(It's sarcasm I posted lol)

 

[Waldorf]

@mtcn77
1. The article stating that PBO is for TR only, IS FROM 2018,
BEFORE ryzen 3xxx series was even out.
2. Have you read the small print on the bottom?
He clearly states that everything is his own OPINION, and NOT Amds.

Find an OFFICIAL info/statement from amd, where AMD says, pbo is TR only.

Personal view: ONE guy on the whole planet, and you believe him over everyone else/amd info? Ok.

 

[mtcn77]

Personal view: ONE guy on the whole planet, and you believe him over everyone else/amd info? Ok.

I don't feel this is getting enough due credit, but we aren't the ones playing to the tune of this overhyped redteam plus member. You want to feel responsibly accurate in your attributions? Let me show you who brought him on this message board and see for yourself who felt the same way clicking the message directory;

He has no clue what's going on. This has been explained by Hallock, crap I dunno how many times now and ppl still cannot get it thru their heads.

Stop mispresenting the issue and pestering Michael Nager and myself please if you have nothing other than derailing the thread into mindless banter.
Clearly, this is between people who know better vs. who don't.

PS: before you make any more oversimplifications, putting this dude at a pedestal and looking down on Michael Nager and myself, if you would actually consider this - I actually wanted to apply for the same position at that public query when I noticed he got elected, so there is that...

PS2: eventhough we come forward in saying enthusiasts aren't backing up the claims of this well-groomed mr. nice guy(no use hiding i'm jelly), we are being picked apart. Not nice gentlemen, not nice at all...

 

[Michael Nager]

Thanks for the lecture btw.
Well as a matter of fact you didn't get what i said regarding your previous post. I was referring about the "boost is a sick joke".
(that is wrong) I don't think it is. It works perfectly fine and as intended. What I get from your post is that you are claiming the boost is shit and AMD scammed people off with it because it is not boosting as it should. When you tweak your settings you can achieve more and I've mentioned it in my post but I guess you have missed it.
BTW I have two 3970x threadrippers and of course they are boosting to 4.5 when not heavy taxed (single core utilization). With Cinebench on the other hand the boost is going up to 3.8-3.9 but I didn't tweak anything and my cooling is an air cooler Noctua noctua nh-u9 tr4-sp3 crap. If I had liquid the situation would have been way different. Your tweaks, btw, are software and that is not as it should have been (unless I read your earlier posts incorrectly). Does this mean if you boost 200Mhz more, that means the boost is shit? Of course not. Nobody misled anything. Go to the AMD webpage https://www.amd.com/en/products/cpu/amd-ryzen-9-3950x and go to the boost section. It says single core boost 4.7Ghz. The default is rather
So if you claim that you have 4.3 instead of 4 on all cores that is tweaking. Getting better cooling instead of stock cooler. You don't have a stock cooler do you?

Sometimes I don't understand you people. You get some performance boost through OC/tweaks like you Mr. and you claim AMD boost is shit because you can get more out of the CPU
When AMD maxes out the CPU it is still shit because there is no OC headroom. The default settings are for all the processors released like 3950x. The silicon may vary one from the other ( you mentioned that too "silicon lottery) remember? The specs are for all 3950x's not individual. All processors released will work that way (safe default way not harming the hardware) if you get a chance to get more out of your CPU then be happy you can't blame something that ain't there.

That is about it At least that is how I get your issue with the AMD boost.

You point me in the direction of the 3950X stock cooler and I will tell you if I have it or not.

I will save you the trouble, there's no stock cooler that comes with the 3950X.

OK so now we have that out of the way I can continue with the "Lecture".

When I replied to you I gave you the benefit of the doubt that I was not clear enough in my post and that a misunderstanding had ensued. Now I realise from this reply that you are just being obtuse.

I have to say that when you said about the boost that, "It works perfectly fine and as intended", I had a LOL moment.

What you are basically saying is that boost is great because it is there when you need it least.

You do not get my "issues" with AMD boost at all.

When speaking of "Boost" with regard to AMD CPUs we are talking about two completely different things:

1) PB2 or "Precision Boost" 2
2) PBO or "Precision Boost Overdrive"

PB2 kinda sorta works as long as you are doing next to nothing with your system (as you say light loads or single threaded loads), or as I like to say, "It does naff-all quicker".

PBO aside from punting in noticeably more voltage and current ends up not being of much use when under load and basically does naff-all even quicker in conjunction with PB2.

With regard to cooling I have tested a single chiplet Ryzen 5 3600X, a double chiplet Ryzen 9 3900X which I have on loan from a friend and my own Ryzen 9 3950X with both air cooling and water cooling (for the air cooling I have a Noctua NH-U12A and for the water cooling I have an AlphaCool Eisbaer 360 AIO with three 120mm Noctua NF-A12x25 fans).

On the single chiplet CPU, the 3600X, the air cooler was noticeably better than the AIO, which is not surprising when you consider that less than 25% of the business part of the AIO cooler, the part of the waterblock which has the heat exchanging fins, covers the hotspot; whereas with the NH-U12A half of the seven heatpipes cover and transfer heat away from that hotspot.

When it comes to the dual chiplet CPUs I have tested (3900X and 3950X) the water cooler is better at cooling the CPU than the air cooler. There's not as much in it as there would be if the CPU was located in the centre of the PCB as opposed to being off-centre, because again the AIO cannot bring its full cooling potential to bear on the CPU hotspots.

There is another component which is temperature sensitive which plays a bit role in performance and that would be Infinity Fabric. With the air cooler on the dual chiplet CPUs the temp, when running CineBench R20 all core went beyond 85 °C and had a tendency to hard crash the system when the IF was cranked up to 1867 MHz (3733 RAM speed) but would be stable at 1800 MHz (3600 RAM speed) whereas the AIO keeps the temp below that threshold and causes no problems.

When you speak of my tweaks being "Software" you do realise that the BIOS is also a piece of software don't you?

The fact of the matter is that because AMD supplies the actual BIOS as a binary (AGESA), what one would normally consider to be "The BIOS" is in fact just a configuration utility for the AGESA. For this reason, if you look at the guide I wrote, I stick to configuring the motherboard parameters (LLC for instance) to the motherboard BIOS and went with Ryzen Master to configure the CPU and RAM because from my extensive testing I have found that Ryzen Master works a lot better in conjunction with AGESA than the BIOS of my mother board.

I have no idea why you thought that linking to the AMD 3950X page when you obviously didn't read it yourself - otherwise you would have known that the 3950X does not come with a cooler i.e. the part where it states, "Thermal Solution (PIB): Cooler Not Included, Liquid Cooling Recommended".

I am however gratified to note that you recognise that what I have done, and what I suggest in the guide is, in your words a "Tweaking" of my system and not an "OverClock". I prefer the term "Configuration".

I notice you make no mention of what voltage your Threadrippers are running on "stock". It would also be nice to know what the specs are other than the cooler.

 

[moproblems99]

I know that static electricity isn't 1.4 Volts, it was a reducto ad absurdum on the premise that it is current that causes damage to CPUs and not voltage.

Well, running a marathon in 100 degree heat and 100% humidity will give you heatstroke at best and kill you at worst. Does that mean that sitting at the beach in 100 degree heat and 100% humidity will do the same? No. Your comparison wasn't a good one.

 

[TheoneandonlyMrK]

You point me in the direction of the 3950X stock cooler and I will tell you if I have it or not.

I will save you the trouble, there's no stock cooler that comes with the 3950X.

OK so now we have that out of the way I can continue with the "Lecture".

When I replied to you I gave you the benefit of the doubt that I was not clear enough in my post and that a misunderstanding had ensued. Now I realise from this reply that you are just being obtuse.

I have to say that when you said about the boost that, "It works perfectly fine and as intended", I had a LOL moment.

What you are basically saying is that boost is great because it is there when you need it least.

You do not get my "issues" with AMD boost at all.

When speaking of "Boost" with regard to AMD CPUs we are talking about two completely different things:

1) PB2 or "Precision Boost" 2
2) PBO or "Precision Boost Overdrive"

PB2 kinda sorta works as long as you are doing next to nothing with your system (as you say light loads or single threaded loads), or as I like to say, "It does naff-all quicker".

PBO aside from punting in noticeably more voltage and current ends up not being of much use when under load and basically does naff-all even quicker in conjunction with PB2.

With regard to cooling I have tested a single chiplet Ryzen 5 3600X, a double chiplet Ryzen 9 3900X which I have on loan from a friend and my own Ryzen 9 3950X with both air cooling and water cooling (for the air cooling I have a Noctua NH-U12A and for the water cooling I have an AlphaCool Eisbaer 360 AIO with three 120mm Noctua NF-A12x25 fans).

On the single chiplet CPU, the 3600X, the air cooler was noticeably better than the AIO, which is not surprising when you consider that less than 25% of the business part of the AIO cooler, the part of the waterblock which has the heat exchanging fins, covers the hotspot; whereas with the NH-U12A half of the seven heatpipes cover and transfer heat away from that hotspot.

When it comes to the dual chiplet CPUs I have tested (3900X and 3950X) the water cooler is better at cooling the CPU than the air cooler. There's not as much in it as there would be if the CPU was located in the centre of the PCB as opposed to being off-centre, because again the AIO cannot bring its full cooling potential to bear on the CPU hotspots.

There is another component which is temperature sensitive which plays a bit role in performance and that would be Infinity Fabric. With the air cooler on the dual chiplet CPUs the temp, when running CineBench R20 all core went beyond 85 °C and had a tendency to hard crash the system when the IF was cranked up to 1867 MHz (3733 RAM speed) but would be stable at 1800 MHz (3600 RAM speed) whereas the AIO keeps the temp below that threshold and causes no problems.

When you speak of my tweaks being "Software" you do realise that the BIOS is also a piece of software don't you?

The fact of the matter is that because AMD supplies the actual BIOS as a binary (AGESA), what one would normally consider to be "The BIOS" is in fact just a configuration utility for the AGESA. For this reason, if you look at the guide I wrote, I stick to configuring the motherboard parameters (LLC for instance) to the motherboard BIOS and went with Ryzen Master to configure the CPU and RAM because from my extensive testing I have found that Ryzen Master works a lot better in conjunction with AGESA than the BIOS of my mother board.

I have no idea why you thought that linking to the AMD 3950X page when you obviously didn't read it yourself - otherwise you would have known that the 3950X does not come with a cooler i.e. the part where it states, "Thermal Solution (PIB): Cooler Not Included, Liquid Cooling Recommended".

I am however gratified to note that you recognise that what I have done, and what I suggest in the guide is, in your words a "Tweaking" of my system and not an "OverClock". I prefer the term "Configuration".

I notice you make no mention of what voltage your Threadrippers are running on "stock". It would also be nice to know what the specs are other than the cooler.

In all your testing it should have become clear to you that at default the bios doesn't control the CPU, using bios settings you can prime the processors environment so as to assist it's boost behaviour, the ageesa may set control values but the board, and ageesa bios hand off control of the CPU to the CPU.
Software clocking overides that and puts AMDS software stack in control.
An avenue for any instability to corrupt if any software crashes not specifically Os or system level.

Software clocking is not the same I assure you of one thing, I wouldn't use it again long term.

For bench runs it's very good and useful as you see.

 

[Michael Nager]

I just found something interesting that addresses the problem I spoke of above with regard to the hotspot(s) of the Ryzen 3000 series of CPUs no longer being in the centre of the PCB.

You can check it out here:

 

[Zach_01]

I would like OP's or anyone else's commends on the following screenshots from my 3600 during different workloads
FYI, PBO settings are not just auto or enabled

1. Idle/browsing/watching videos


2. CB-R20 single thread


3. CB-R20 multi thread


4. FarCry NewDawn gameplay