Question I don't understand this.

[Deleted member 2720853]

Another cooling related issue I think?

Recently re-built my Ryzen build because my Gigabyte board died on me. This time I have an ASUS ROG Strix B550-F and Ryzen 9 3900X, with a NH-D15 as cooler, using NT-H2 thermal paste.

The way I applied the paste is by putting a dot in the middle (a little less when I put the motherboard in my PC after testbenching -- I put too much on my testbench and it oozed out of the cpu as I mounted the cooler). However even with a single dot, the paste DID spread over the entire IHS perfectly when I took off the cooler.

I tightened the cooler alternating between screws until both screws stopped and I couldn't tighten anymore. On the other build, I didn't tighten the cooler completely and I had reached 90C in AIDA64 and sitting over 75C in Cinebench. However on this build, I reached a max of 81C in AIDA, alternating between 79-81. In Cinebench it's around 71C max, but mostly sitting at 70C.

In gaming, it's sitting between 50-55, in the more CPU heavy games going to 65C.

What I don't get is that my boost clocks are the same. 4.02 GHz on stock.

What am I doing wrong? Why is the CPU not boosting as it should? I see people reaching over 4.2 GHz with PBO on this chip and I'm sitting here reaching 4.02 GHz.

When I apply my manual OC of 4.35 GHz on one CCD and 4.2 GHz on the other CCD, at 1.25V, the temps are better and the scores are higher.

But wht I don't get is why the stock CPU settings don't boost as high.

The only change I made to the CPU cooler fans is setting them to go 1100 RPM max because of the temp spikes on Ryzen making fans ramp up.

Did I mess up the cooler installation or paste installation again? What is the best way to apply NT-H2 to this CPU? I don't get it.

 

[siaan312]

Another cooling related issue I think?

Recently re-built my Ryzen build because my Gigabyte board died on me. This time I have an ASUS ROG Strix B550-F and Ryzen 9 3900X, with a NH-D15 as cooler, using NT-H2 thermal paste.

The way I applied the paste is by putting a dot in the middle (a little less when I put the motherboard in my PC after testbenching -- I put too much on my testbench and it oozed out of the cpu as I mounted the cooler). However even with a single dot, the paste DID spread over the entire IHS perfectly when I took off the cooler.

I tightened the cooler alternating between screws until both screws stopped and I couldn't tighten anymore. On the other build, I didn't tighten the cooler completely and I had reached 90C in AIDA64 and sitting over 75C in Cinebench. However on this build, I reached a max of 81C in AIDA, alternating between 79-81. In Cinebench it's around 71C max, but mostly sitting at 70C.

In gaming, it's sitting between 50-55, in the more CPU heavy games going to 65C.

What I don't get is that my boost clocks are the same. 4.02 GHz on stock.

What am I doing wrong? Why is the CPU not boosting as it should? I see people reaching over 4.2 GHz with PBO on this chip and I'm sitting here reaching 4.02 GHz.

When I apply my manual OC of 4.35 GHz on one CCD and 4.2 GHz on the other CCD, at 1.25V, the temps are better and the scores are higher.

But wht I don't get is why the stock CPU settings don't boost as high.

The only change I made to the CPU cooler fans is setting them to go 1100 RPM max because of the temp spikes on Ryzen making fans ramp up.

Did I mess up the cooler installation or paste installation again? What is the best way to apply NT-H2 to this CPU? I don't get it.

The temps are not the issue here.
The cpu just seems to.. not boost properly.
Are you using an 8 pin eps connector or just 4 pins? This seems like a power issue to me for some reason.

 

[wi5pa]

Your cooler seems to be working fine, and this is unrelated to your boost clocks.

Your boost clocks might be locked if you have Enabled XMP ??
Your boost clocks might be regulated by the motherboard,
The CPU might also be limited by the silicone lottery
A manual overclock might be the best way forward ?

 

[Deleted member 2720853]

I'm using the 8 pin connector and the PSU is fine. I don't know if I should repaste or not.

Why would XMP lock my boost clocks?

 

[wi5pa]

XMP will raise your base clock speed as the CPU has to work harder with the faster RAM

The base for a Ryzen 9 3900x is 3.8ghz. The max boost is 4.6ghz (single core)

Your running at all cores 4.02 ghz which is most likely the XMP enabled adjusted core clocks.

If you remove XMP your core will probably go back to base speed 3.8ghz and turbo to 4.6ghz
If you add XMP and then manual overclock you will get better core speed like you already said at 4.35ghz.

 

[Deleted member 2720853]

Okay I repasted with 5 dots, the spread was worse than with 1 dot.

This is the 1 dot spread:

Also, is a little bit of cooler sag on the NH-D15 expected? Not significant, but like 1mm or so? I tightened it all the way so I dont see how it's because of me.

 

[wi5pa]

Post number 2 and 3 both say that the temperatures are not the problem.

I dont understand why you are blaming the cooler

 

[Deleted member 2720853]

Because I don't understand Ryzen at this point. What on earth is making it boost so low? Even with PBO which raises the limits, and I'm on a board with decent VRMs for it.

 

[Darkbreeze]

What am I doing wrong? Why is the CPU not boosting as it should? I see people reaching over 4.2 GHz with PBO on this chip and I'm sitting here reaching 4.02 GHz.

There is nothing wrong. This is normal. FEW samples are able to do what you seem to think should be standard for everybody.


https://www.tomshardware.com/news/a...ing-advertised-boost-speeds-survey,40291.html

https://www.pcgamesn.com/amd/ryzen-9-3900x-overclock

 

[Deleted member 2720853]

There is nothing wrong. This is normal. FEW samples are able to do what you seem to think should be standard for everybody.

I see. Alright, that answers that question. Another question: Ambient room temp is 25C or so, is an idle temp of 35C-42C normal on a NH-D15? I went back to my original 1 dot method as the 5 dot method gave me complications and didn't spread as well. Temps seem to be better now so I probably did a better job at applying it and mounting the cooler.

Also, what about the very minuscule CPU cooler sag (like 1mm or maybe less)? It's tightened completely. I just want to make sure I didn't do something wrong.

 

[Deleted member 2720853]

https://www.tomshardware.com/news/a...ing-advertised-boost-speeds-survey,40291.html

https://www.pcgamesn.com/amd/ryzen-9-3900x-overclock

By the way I have looked at these and my CPU is able to reach 4.2 GHz all core on 1.25V, given that my current manual OC is 4.35 GHz on 6 cores and 4.2 GHz on the other 6 at 1.25V.

So why is the CPU not reaching these speeds on stock?

 

[Darkbreeze]

Because on a manual overclock you are overriding the built in profile. You may be using a lower voltage than what is being used by the PBO2 profile at any given clock frequency. That increased voltage, usually, results in a higher temp for the core, the package and the VRMs, and any one of those plus a few other sensor inputs that you may not even have access to monitoring might individually or collectively have a direct impact on how the boost is affected or applied.

For example. And these numbers are just loosely being thrown out there. They aren't meant to represent any actual configuration, at all, nor is this specific to only Ryzen platforms because it is almost a universal truth across the board when it comes to any CPU or motherboard manufacturer supplied profile where they are much more worried about erring to the side of stability than they are to the side of reducing heat or voltage.

Let's say you have a 4.4Ghz OC and that you believe (Because about 50% of the time somebody says they are "stable at X voltage with Y frequency" it turns out that once they run the test routines they SHOULD be running to establish that, they really aren't stable or thermally compliant at all.) it is stable at (For example) 1.3v. With this configuration you report that you max out at 79°C, which is within the allowed range by recommended spec. Great.

Now, we're saying that is a manual configuration, with a manual voltage and frequency setting. Now we use one of the overclocking automatic profiles or OC utilities to configure the same overclock, automatically. For our purposes we can assume that PBO/PBO2 will behave much the same as those profiles since they all manufacturer supplied and tweaked by the board manufacturer, or whatever. The fact is, that at the same clock frequency, so, 4.4Ghz, the system is going to almost always, 100%, use a higher voltage AND likely automatically apply a higher LLC as well, than what we would use when manually configuring the overclock, because they are more concerned with making sure that the system is stable than they are about a reduction of heat or other considerations.

So my assumption, based on that, and it may not be the right answer for your specific situation, I have no way to know without having the system in front of me to actually lay hands on and test and get first hand results from, would be one of three things.

Either you are not actually stable at that voltage with that 4.35/4.2Ghz frequency because what you are using to demonstrate stability isn't what you actually should be using.

Or, you are not actually thermally compliant at that voltage but are not using an acceptable utility to validate thermal compliance, both of which mean that if you WERE doing those things correctly (And I'm NOT saying that you're not, I'm just saying IF, because we see the IF turn into a reality quite often) you would not be able to say your configuration was both stable and thermally compliant with those settings, and would likely be closer to what you are seeing with PBO enabled.

Or, it's like I said, and the system is overvolting to maintain stability regardless that it's reducing the peak and duration of the boost.

It's also possible I suppose that it's some other factor. I'd wonder if you've tried disabling PBO and just using the standard XFR2 stock boost profile? I'd also wonder if you have the MOST recent motherboard BIOS version installed?

Also, for what it's worth, myself and a good many others do not believe that Aida or Cinebench are acceptable metrics for thermal compliance testing. They are fluctuating workloads, and fluctuating workloads can never be desirable when doing thermal testing. You NEED a steady state workload and there are a few ways to get a steady state workload but most utilities out there are either not steady state, or they are steady state but they don't bring the CPU to full TDP, which both Intel and AMD datasheets outline as a standardized methodology when you are trying to demonstrate thermal compliance.

For testing thermals, you want to download and run Prime95. Choose the "Small FFT" option. Not "Smallest FFT". Not "Large FFT". Not "Blend". ONLY run the "Small FFT" option when performing thermal compliance testing. Also, you need to disable all available AVX instruction options on the main preferences window when first configuring the test. Some options may be grayed out at first, but may become available after disabling other AVX options. We do not want any AVX options enabled for the purpose of baseline thermal testing. If you run games or applications that make heavy use of AVX instructions, you can of course later go back and do additional testing with AVX instructions enabled and an offset configured in the BIOS if you wish, or need to do so.

For our purposes we generally want those options fully disabled though.

If you can run Prime Small FFT for 15 minutes without exceeding 80°C and without ANY errors/workers dropping out, then that configuration is pretty much thermally compliant.

For stability, you can run a lot of utilities, like Aida, or Cinebench, but I'd recommend running those as secondary tests rather than as the primary determination. I personally think running Realbench first is a much better validation. Choose the stress test option. Select half of your installed memory capacity. Run it for one hour between changes and 8 hours before saying any given configuration is "good". Then, if you can pass an 8 hour run, feel free to run some secondary testing with any of these other utilities because it can't be denied that we've all seen various configurations over the years perform completely stable with one or two different utiltiies and then fail abysmally with another one.

And generally speaking, if this is to be a daily driver a lot of us who've been overclocking for a long time like to fine tune the final OC by dropping the frequency by 100mhz once we are satisfied we've reached the peak of what that cheap can do and remain both stable and thermally compliant, in order to reduce the probability of any increased risk of electromigration/degradation from riding that fine line where the tolerance for too much voltage gets fuzzy. If you are well within tolerance based on manufacturer spec, then obviously that is something that you probably don't need to worry about but considering that most OC's are already well beyond what the all core boost would be, and considering that the loss of 100mhz is generally a very small percentage of overall performance, it's not a bad insurance policy when it comes to making sure your CPU has a nice long run.

 

[Deleted member 2720853]

Because on a manual overclock you are overriding the built in profile. You may be using a lower voltage than what is being used by the PBO2 profile at any given clock frequency. That increased voltage, usually, results in a higher temp for the core, the package and the VRMs, and any one of those plus a few other sensor inputs that you may not even have access to monitoring might individually or collectively have a direct impact on how the boost is affected or applied.

For example. And these numbers are just loosely being thrown out there. They aren't meant to represent any actual configuration, at all, nor is this specific to only Ryzen platforms because it is almost a universal truth across the board when it comes to any CPU or motherboard manufacturer supplied profile where they are much more worried about erring to the side of stability than they are to the side of reducing heat or voltage.

Let's say you have a 4.4Ghz OC and that you believe (Because about 50% of the time somebody says they are "stable at X voltage with Y frequency" it turns out that once they run the test routines they SHOULD be running to establish that, they really aren't stable or thermally compliant at all.) it is stable at (For example) 1.3v. With this configuration you report that you max out at 79°C, which is within the allowed range by recommended spec. Great.

Now, we're saying that is a manual configuration, with a manual voltage and frequency setting. Now we use one of the overclocking automatic profiles or OC utilities to configure the same overclock, automatically. For our purposes we can assume that PBO/PBO2 will behave much the same as those profiles since they all manufacturer supplied and tweaked by the board manufacturer, or whatever. The fact is, that at the same clock frequency, so, 4.4Ghz, the system is going to almost always, 100%, use a higher voltage AND likely automatically apply a higher LLC as well, than what we would use when manually configuring the overclock, because they are more concerned with making sure that the system is stable than they are about a reduction of heat or other considerations.

So my assumption, based on that, and it may not be the right answer for your specific situation, I have no way to know without having the system in front of me to actually lay hands on and test and get first hand results from, would be one of three things.

Either you are not actually stable at that voltage with that 4.35/4.2Ghz frequency because what you are using to demonstrate stability isn't what you actually should be using.

Or, you are not actually thermally compliant at that voltage but are not using an acceptable utility to validate thermal compliance, both of which mean that if you WERE doing those things correctly (And I'm NOT saying that you're not, I'm just saying IF, because we see the IF turn into a reality quite often) you would not be able to say your configuration was both stable and thermally compliant with those settings, and would likely be closer to what you are seeing with PBO enabled.

Or, it's like I said, and the system is overvolting to maintain stability regardless that it's reducing the peak and duration of the boost.

It's also possible I suppose that it's some other factor. I'd wonder if you've tried disabling PBO and just using the standard XFR2 stock boost profile? I'd also wonder if you have the MOST recent motherboard BIOS version installed?

Also, for what it's worth, myself and a good many others do not believe that Aida or Cinebench are acceptable metrics for thermal compliance testing. They are fluctuating workloads, and fluctuating workloads can never be desirable when doing thermal testing. You NEED a steady state workload and there are a few ways to get a steady state workload but most utilities out there are either not steady state, or they are steady state but they don't bring the CPU to full TDP, which both Intel and AMD datasheets outline as a standardized methodology when you are trying to demonstrate thermal compliance.

For testing thermals, you want to download and run Prime95. Choose the "Small FFT" option. Not "Smallest FFT". Not "Large FFT". Not "Blend". ONLY run the "Small FFT" option when performing thermal compliance testing. Also, you need to disable all available AVX instruction options on the main preferences window when first configuring the test. Some options may be grayed out at first, but may become available after disabling other AVX options. We do not want any AVX options enabled for the purpose of baseline thermal testing. If you run games or applications that make heavy use of AVX instructions, you can of course later go back and do additional testing with AVX instructions enabled and an offset configured in the BIOS if you wish, or need to do so.

For our purposes we generally want those options fully disabled though.

If you can run Prime Small FFT for 15 minutes without exceeding 80°C and without ANY errors/workers dropping out, then that configuration is pretty much thermally compliant.

For stability, you can run a lot of utilities, like Aida, or Cinebench, but I'd recommend running those as secondary tests rather than as the primary determination. I personally think running Realbench first is a much better validation. Choose the stress test option. Select half of your installed memory capacity. Run it for one hour between changes and 8 hours before saying any given configuration is "good". Then, if you can pass an 8 hour run, feel free to run some secondary testing with any of these other utilities because it can't be denied that we've all seen various configurations over the years perform completely stable with one or two different utiltiies and then fail abysmally with another one.

And generally speaking, if this is to be a daily driver a lot of us who've been overclocking for a long time like to fine tune the final OC by dropping the frequency by 100mhz once we are satisfied we've reached the peak of what that cheap can do and remain both stable and thermally compliant, in order to reduce the probability of any increased risk of electromigration/degradation from riding that fine line where the tolerance for too much voltage gets fuzzy. If you are well within tolerance based on manufacturer spec, then obviously that is something that you probably don't need to worry about but considering that most OC's are already well beyond what the all core boost would be, and considering that the loss of 100mhz is generally a very small percentage of overall performance, it's not a bad insurance policy when it comes to making sure your CPU has a nice long run.

Thanks for the detailed answer.

What I learned from this is that I should probably leave the CPU on stock settings. I'm going to do that.

 

[Deleted member 2720853]

Okay well, new problem...

I put the CPU back on stock settings, Ryzen Balanced power plan, cores parked, and the CPU now boosts to 4.6 on a few cores. Yes I have the latest BIOS with AGESA 1.1.0.0.

I enabled PBO in the BIOS because apparently that's better for gaming on Zen 2, and I launched GTA 5 to see how it performs.

Temps are obviously higher, in the 60s, which is expected because PBO is pushing the chip, but then I looked at the voltage in Ryzen Master on my second monitor while playing and I saw 1.44. My CPU was well over 4.2 GHz and at around 60C when this happened. It remained there constantly.

Is this damaging my chip? 1.44 sounds too high. Is PBO unsafe?

Will a negative -.102mv offset affect the 4.6 boost?

 

[Darkbreeze]

I'd disable PBO, which is what I've done on most Ryzen builds and I know a lot of other guys have been doing this as well. Leave the PB/XFR2 stock boost profile enabled. See how that does. It may not boost as high, but from what I've seen it tends to STAY boosted LONGER at a slightly reduced clock speed. Try it and see. Can't hurt. You can always enable PBO again.

I think if you configure an offset, it will adjust max boost accordingly, and yes, probably reduce the peak boost. I'm not 100% sure on that though. I think ANY kind of manual adjustment like that has an effect on the boost profile automatically though. It might even automatically disable the PBO configuration.

Some "sources" say voltages up to 1.45v are within safe limits on Ryzen. Many disagree with that. I won't argue either way because I simply don't have enough technical knowledge about Ryzen, in this area, to fight on either side of that river.

You might find some useful information here:

https://community.amd.com/thread/249957

And here:

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

In the months that have passed since I started experimenting with the Ryzen 5 3600X on my X470 motherboard (GigaByte X470 AURUS Gaming 7 WiFi Rev. 1.1 and then helping my friend configure his Ryzen 3900X on his motherboard (ASUS X570 ROG Crosshair VIII Hero (WiFi). I bought the X570 motherboard...

www.techpowerup.com

 

[Phaaze88]

Didn't you try 1Usmus' Clock Tuner for Ryzen(CTR)?

 

[Deleted member 2720853]

Didn't you try 1Usmus' Clock Tuner for Ryzen(CTR)?

Yeah I did.

 

[Phaaze88]

So all that did was screw things up?
I was just curious.

 

[Darkbreeze]

I've seen a couple of people who've said the 1Usmus clock tuner AND the 1Usmus power plan had shown some promise for them, but most of the users I've talked with who've used it have said it didn't do jack.

 

[Deleted member 2720853]

It only helped in synthetic benchmarks, I don't do any heavily threaded work. But I don't see how that stock single thread boost helps me either especially considering it's barely happening and during gaming my cores never go past 4250 MHz.

 

[Darkbreeze]

I don't either really. Potentially, if you're playing an older game or a newish game that for some reason was not developed to be very multicore friendly then any increase in single core should be at least marginally helpful, possibly very helpful if it's a significant difference in speed, but otherwise it's useless because the second another thread kicks in, game over. Totally different profile and boost rules.

 

[Deleted member 2720853]

I don't either really. Potentially, if you're playing an older game or a newish game that for some reason was not developed to be very multicore friendly then any increase in single core should be at least marginally helpful, possibly very helpful if it's a significant difference in speed, but otherwise it's useless because the second another thread kicks in, game over. Totally different profile and boost rules.

I noticed it sometimes boosting to 4.5 on Age of Empires Definitive Edition which iirc still uses the same single core engine from ages ago.

 

[Darkbreeze]

Yes, that's a good example. In that case, you might see some benefit. For MOST games from the last couple of years and certainly almost anything out there going forward that is from a major developer, single core will likely rarely be a big factor except as the single core performance affect ALL cores, cumulatively.

 

[Deleted member 2720853]

Okay so in order to counter the idle shutdowns I disabled C-states. I also disabled PBO in the BIOS. I enabled CPPC and CPPC preferred cores. Ryzen Balanced power plan, all cores unparked by the plan, latest BIOS with AGESA 1.1.0.0 Patch B and latest chipset drivers. Windows 10 20H2 with latest updates.

I see the CPU still boosting to 4.6. In games it does boost higher on the cores the game uses, but not for long, in quick bursts. At idle, some cores go to sleep (is this supposed to happen? I thought no cores slept with C-states on).

Can someone tell me if XFR/PB2 is working as it should right now? This seems like the best mitigation for me so far -- countering idle shutdowns while keeping the boost behavior of the CPU.

 

[Darkbreeze]

C-states are specifically FOR low power operation and reduction. That's what it does. Different levels of C-state equal different levels of low power/sleep/inactive/hibernate etc.

My preference for Ryzen is this:

Cool N Quiet - Enabled (If this setting is not available in your BIOS, just worry about the rest)

Core CPPC - Enabled

CPPC preferred cores - Enabled

Advanced/Global C-states - Enabled

Precision boost overdrive (PBO/PBO2) - Disabled (Unless you have high end cooling installed and aren't having problems. Also, standard boost profiles like Precision boost (Non-overdrive) and XFR2 should be left enabled.)

Also, it is probably a good idea to disable hibernation/hybrid sleep, because it tends to cause a lot of problems, ever since Windows introduced it many years ago and regardless of which OS version you are using.

Hibernation

Disabling Hibernation is very easy.

You do not want to do this if you have a laptop as Hibernation is essential when a laptop's battery loses charge and the system needs to safely save it's state. If you have a laptop skip disabling Hibernation and instead disable Fast Startup and Hybrid Sleep if you are having issues.
To disable Hibernation:

1. The first step is to run the command prompt as administrator. In Windows 10, you can do this by right clicking on the start menu and clicking "Command Prompt (Admin)"

2. Type in "powercfg.exe /h off" without the quotes and press enter. If you typed it in correctly, the cursor will simply start at a new line asking for new input

3. Now just exit out of command prompt