Question 'Undervolt' Ryzen 3900X to reduce summer temperatures

[Fardringle]

I recently put together a Ryzen 9 3900X system with a Gigabyte Aorus X570 Elite Wifi motherboard, 32 GB of G.Skill RipJaws 3600Mhz CL16 RAM, and a Scythe Fuma 2 cooler. (plus my previous SSD and an EVGA GTX 1060 3GB)

Overall I'm very happy with the performance results. However, it's starting to get warmer here, and the already somewhat high temperatures in the system are just going to go higher as summer progresses, so I'm looking for the most efficient way to reduce heat without losing too much performance.

I already undervolted the GTX 1060 using this guide and dropped full load temps on the GPU by about 8C, which is a great start as it not only lowered the temps but actually boosted the GPU benchmarks a bit as well since the card was throttling itself a bit and not boosting all the way up to 1911MHz before the changes...

I'd really like to do something similar with the CPU if possible. Don't mind if I lose a small amount of performance if the temperature drop is good enough. Boosting performance and lowering temperature is probably too much to ask for on the CPU. I've found quite a few guides online for undervolting the CPU, but there are so many different options, and some of them seem to be excessively complicated and/or have pretty questionable results, so I'm hoping to get some good advice here.

It seems like the easiest option would be to simply lower the PPT to something below the stock 142W maximum, and with some trial and error I can probably find a suitable setting. But if there are better options, I'm definitely open to suggestions.

 

[pauldun170]

I have the 3900X paired with a Fuma 2 on an Aorus Elite non wifi

What temperatures are you seeing now? What do you mean by "high temperatures"?
What case are you using?
Are you using the Ryzen power plan?

 

[DrMrLordX]

Undervolting will not necessarily change how much power the CPU uses. It CAN reduce reported temperatures, and that may have more to do with how heat is distributed across the die as you lower operating voltages. But you have to keep in mind that if you pretty much have to undervolt via offsets or by using low LLC settings to induce vdroop in high-current workloads. Otherwise you're talking static voltage, which leads to static clocks. Or you can try p-state undervolting which is a recipe for instability. If you undervolt via offsets, the worst you're going to get is false-clocking (which hurts performance).

I run undervolted all the time. If I load up CBR20 or Blender Benchmark (for example), I still get sustained package power draw of 142W, which is the most the CPU can use during default operation. Actually let me record results here to give you an idea of what's going on:

Undervolted (LLC off ("Normal" in gigabyte UEFI), -.1v offset)

Blender Package Power, peak: 145.537W
SoC current, peak: 12.059a
Core current, peak: 61.929a
Tctl, peak: 60.8C

Completion time: 1 minute 52 seconds

No undervolt (LLC Auto, no voltage offsets)

Blender Package Power, peak: 145.550W
SoC current, peak: 11.765a
Core current, peak: 60.612a
Tctl, peak: 60.8C

Completion time: 1 minute 55 seconds

As you can see, in this instance, the reported temps don't change at all. Current shifts upwards a bit as I lower voltage to fill out the package power limits. Also, in my case, performance goes up. But temps stay about the same. If you really want to lower temps and lower power output, you need to lower PPT.

 

[Shmee]

Interesting, Ryzen boost factors sure can be weird. So does this mean that reducing applied voltage with an offset can dynamically increase clocks? How do you test for stability with that?

 

[DrMrLordX]

Interesting, Ryzen boost factors sure can be weird. So does this mean that reducing applied voltage with an offset can dynamically increase clocks? How do you test for stability with that?

Near as I can tell, if you set negative voltage offsets, the effects are largely dependent on cooling. It has no obvious way of detecting that voltages have been lowered, so it's going to just keep running volts and clocks up until package power is met, following the curve defined by temperature. The higher the temp, the less aggressive the voltage/clockspeed curve.

As for testing stability, I have yet to find any instance where negative offsets have lead to unstable operation. The boost algo will introduce false clocks instead - and sometimes fall short of package power limits in the process. So I guess I should amend my previous observations that if you attempt undervolting via offset beyond a certain point, you can (maybe) reduce CPU package power by making the CPU downclock itself while "false clocking" often enough to make it appear to be operating normally to monitoring software. Benchmark scores and real-world performance will both tank.

 

[Fardringle]

I should note that I run distributed computing projects, so 100% load almost all of the time, except when I'm using the PC to play a game and pause BOINC..

I have a Corsair Carbide Air 540 case, with three intake fans in the front, two exhaust on top and one exhaust in the back. I've tried different arrangements of the fans and while some varieties definitely increased temperatures, I didn't find any that lowered them.

Ambient temperature in the house is about 72-80F (22-27C) now that outside temperatures are getting a lot warmer. A few weeks ago, when inside temps stayed right around 70F and outside temps were still cooler, the CPU peaked at around 80-81C under 100% load, usually sitting closer to 78-79C. With the warmer outside and inside temperatures, I'm seeing peak load temps go as high as 88-89C, and the average sitting around 85-86C.

I haven't played with the voltage offset yet, but I did try adjusting the PPT from 142W to other levels.

130W resulted in max boost clocks of about 3.9-4.0MHz instead of the 4.1-4.2 that I got on stock settings. This actually only dropped benchmark scores by less than 5%, but didn't change the CPU temps by any noticeable amount.

120W made it so the CPU didn't boost at all and just stayed right at the base 3.8MHz. Temps did drop a bit, back to the "cold weather" range of 78-80C. Also dropped about 15% more on benchmark scores. I'm not sure why it was such a big drop, but I can also "feel" stuttering and delays in normal functions on the PC at this setting so I definitely don't think it's a good option even though temperatures are a bit better.

 

[Fardringle]

Is it possible my Fuma 2 is somehow defective, getting such high temperatures? I have checked to make sure it is installed properly, and the Noctua NT-H1 thermal paste is applied correctly as well. Both fans on the heatsink are working, and are spinning in opposite directions as specified by the installation instructions...

 

[.vodka]

The correct way to do this on 3rd gen Ryzen is to drop the power envelope from 105w to 65w (enabling eco mode)

You'll still get your juicy single core turbo and low threaded performance which fits fine in the 65w bracket, but multicore performance and power consumption will drop as expected.

 

[JoeRambo]

If You don't mind loosing some minor gaming performance, good old static overclock with fine tuned voltage is still awesome. It of course requires 24/7 load to fully benefit from it.
I'd not bother with running CCX at diffent clocks, but 1.1-1.15V on load after droop will give you both great 24/7 performance due to 3.9-4.1 clocks and improved thermals. At the loss of some minor gaming perf of course

 

[Fardringle]

I really don't care much about gaming performance. It's the 24/7 100% load temperatures that I'm hoping to control without losing too much performance in the distributed computing projects.

 

[VirtualLarry]

Honestly, what I found running 24/7 PrimeGrid load on my Ryzen R5 3600, and my attempts at a static voltage all-core OC, were less than spectacular, in the temps dept. These chips run HOT! That said, hotspot temps, are not the same as thermal output (BTUs, etc), and may not heat up the room as much as some might believe, solely based on what CPU temps are reporting.

Honestly, as much as it pains me to say it - unless it's throttling somehow or affecting performance... just let the hotspot temps hit 80-90C. (Maybe this is not the best advice, and feel free to get other opinions.) Just that it seems futile to attempt to control temps, as even on 240mm AIO WC, my Ryzen R5 3600 hits 85C under 100% load (~4Ghz all-core load).

 

[Rigg]

Go into the AMD overclocking page in BIOS. In the manual PBO menu you can set a temperature limit.The CPU will throttle itself accordingly to maintain that temp. You can alternatively manually set the TDP or use Eco mode. I used a Fuma 2 with my 3900X for a stint. it's a very nice cooler but wouldn't be my first choice to cool a 3900x in a high ambient temperature environment.

 

[DrMrLordX]

I really don't care much about gaming performance. It's the 24/7 100% load temperatures that I'm hoping to control without losing too much performance in the distributed computing projects.

Since you are running DC workloads, you probably sustain a fairly constant clockspeed, correct? What clockspeed does the CPU hold? We can tune around that with a static overclock (props to @JoeRambo for mentioning that).

 

[moinmoin]

If changing ambient temperature is the worry, wouldn't it make more sense to tweak the TDC value to something that should be fine even in summer heat? Since ambient temperature is going to change imo it's better to let the CPU handle adapting to it instead making it run at a static frequency.

 

[JoeRambo]

If changing ambient temperature is the worry, wouldn't it make more sense to tweak the TDC value to something that should be fine even in summer heat? Since ambient temperature is going to change imo it's better to let the CPU handle adapting to it instead making it run at a static frequency.

Static freq can have that proper undervolt and since he is running 24/7 load, moving CPU lower on V/F curve means gains in efficiency.

 

[moinmoin]

Static freq can have that proper undervolt and since he is running 24/7 load, moving CPU lower on V/F curve means gains in efficiency.

Eco mode would be the quick way to more optimal efficiency.

Letting the CPU handle it dynamically is still a better idea than static frequency since with Zen a static frequency doesn't make any difference between heavy and light instructions, meaning depending on what kind of calculations lead to the 24/7 100% load the temperature will vastly differ while the same frequency is kept. And @Fardringle wants to keep the temperature in check, which static frequency doesn't really do.

 

[Rigg]

Eco mode would be the quick way to more optimal efficiency.

Letting the CPU handle it dynamically is still better idea than static frequency since with Zen a static frequency doesn't make any difference between heavy and light instructions, meaning depending on what kind of calculations lead to the 24/7 100% load the temperature will vastly differ while the same frequency is kept.

I still think if CPU temperature is the concern than setting a temp limit is the simplest and easiest solution to the OP's problem.

 

[coercitiv]

1. Lower power limit until temperatures are adequate.
2. Undervolt gradually until performance stops increasing. (would start with lowering LLC first)
3. Enjoy.

 

[Markfw]

Without messing with my setup, its 85F in the house and 85F outside (the AC can't keep up) and the temps is still just 80c on the computer.

 

[Fardringle]

I think I like the idea of setting a max temp that I'm comfortable with, and just let the system do whatever it wants as long as it doesn't go over that number. I'll give it a try, and if it doesn't get the results I want, I'll try some of the power limit/undervolt suggestions.

 

[Fardringle]

Running at 100% load for a bit over an hour now, with CPU temp limit set to 85C in the BIOS. HWInfo shows an average CPU temp is 82.8C with a max of 84.7C. Still rather toasty and warmer than I'd prefer to have it, but doesn't make me as nervous as the nearly 90C spikes I was seeing. According to HWInfo, average clock speed on the CPU cores is 4.014MHz with a max of 4.142. That is a bit lower than the 4.1 or so average and 4.2 max it was getting before, but not by much, and I haven't seen any weird stuttering or hesitation like I did when I was playing with the PPT settings. I'll let it go like this fir a day or two and see if the numbers stay in that range.

 

[UsandThem]

Running at 100% load for a bit over an hour now, with CPU temp limit set to 85C in the BIOS. HWInfo shows an average CPU temp is 82.8C with a max of 84.7C. Still rather toasty and warmer than I'd prefer to have it, but doesn't make me as nervous as the nearly 90C spikes I was seeing. According to HWInfo, average clock speed on the CPU cores is 4.014MHz with a max of 4.142. That is a bit lower than the 4.1 or so average and 4.2 max it was getting before, but not by much, and I haven't seen any weird stuttering or hesitation like I did when I was playing with the PPT settings. I'll let it go like this fir a day or two and see if the numbers stay in that range.

You could always throw a more powerful fan on the cooler since the included fans are more lower noise focused.
https://www.tweaktown.com/reviews/9223/scythe-fuma-2-cpu-cooler-review/index.html

Cooling the tower are a pair of Kaze Flex fans, but various thicknesses. The leading fan is 120mm in size, but only 15mm in thickness. Its speed ranges from 300 RPM on up to 1200 RPM, delivering 33.86 CFM of airflow and 0.9 mmH2O of static pressure, and is powered with a 4-pin PWM connection. The second fan, the one that installs between the towers is a 120mm fan, but this time it is 25mm thick. The speed range is the same as the thinner ones, but the CFM has increased to 51.17, static pressure is increased slightly to 1.05 mmH2O, and again, it is powered with a 4-pin PWM connector. As for the noise levels, the 15mm thick fan is shown to deliver 23.9 dB(A), while the 25mm thick fan is shown to top out at 24.9 dB(A).

 

[Thunder 57]

Without messing with my setup, its 85F in the house and 85F outside (the AC can't keep up) and the temps is still just 80c on the computer.

Damn, I would not take 85F inside very well.

You could always throw a more powerful fan on the cooler since the included fans are more lower noise focused.
https://www.tweaktown.com/reviews/9223/scythe-fuma-2-cpu-cooler-review/index.html

Isn't the point though to reduce heat production? Sure, you could cool the CPU better, but the heat still escapes into the room.

On my motherboard I have the following:

In the BIOS setup select the Advanced tab.
Choose AMD CBS.
Choose NBIO Common Options.
Choose System Configuration.
A menu should appear with the following options: Auto, 35W, 45W, 65W.

Mine is a bit different where it's set yo Auto or Manual, and under Manual control it would let me set any power limit I want. I even tested a few different ones and posted clock results. Just did a quick search and found it.

 

[Rigg]

Damn, I would not take 85F inside very well.



Isn't the point though to reduce heat production? Sure, you could cool the CPU better, but the heat still escapes into the room.

On my motherboard I have the following:

In the BIOS setup select the Advanced tab.
Choose AMD CBS.
Choose NBIO Common Options.
Choose System Configuration.
A menu should appear with the following options: Auto, 35W, 45W, 65W.

Mine is a bit different where it's set yo Auto or Manual, and under Manual control it would let me set any power limit I want. I even tested a few different ones and posted clock results. Just did a quick search and found it.

Setting a temp limit achieves the same goal with less screwing around. It accomplishes the exact same thing only using temperature limit rather than power limit to induce throttling. Since the goal here is lower temps why not use temp limit rather than power? He can set the temp limit to whatever he wants.

Running at 100% load for a bit over an hour now, with CPU temp limit set to 85C in the BIOS. HWInfo shows an average CPU temp is 82.8C with a max of 84.7C. Still rather toasty and warmer than I'd prefer to have it, but doesn't make me as nervous as the nearly 90C spikes I was seeing. According to HWInfo, average clock speed on the CPU cores is 4.014MHz with a max of 4.142. That is a bit lower than the 4.1 or so average and 4.2 max it was getting before, but not by much, and I haven't seen any weird stuttering or hesitation like I did when I was playing with the PPT settings. I'll let it go like this fir a day or two and see if the numbers stay in that range.

Using a slight negative offset voltage with low LLC as @coercitiv suggested earlier should also work with the temp limit. It's a bit weird to dial in though. With clocks set on auto the CPU will clock stretch rather than crash when power starved. The clocks will appear normal or improved even though performance is reduced. You must test performance when under-volting with clocks set on auto. Typically any more than a -50-100mv offset will reduce performance. This is also extremely hard to properly test since the behavior can vary depending on workload. Since your only running a single workload this shouldn't be an issue. If clocks are set on auto you can't trust the clock speed the monitoring software is showing you when under-volting.

That's the long explanation for

2. Undervolt gradually until performance stops increasing. (would start with lowering LLC first)

 

[Fardringle]

You could always throw a more powerful fan on the cooler since the included fans are more lower noise focused.
https://www.tweaktown.com/reviews/9223/scythe-fuma-2-cpu-cooler-review/index.html

Maybe, but if I'm going to throw more money at it, I might as well return the Fuma and get something else...

Sure would like to see the temps they reported in that review, though!