Ryzen 3900X undervolting and limiting clock speed

[uyaem]

People on this forum and in the world-wide gamer communities have reported high temperatures with this CPU.
It's designed to run hot, and allegedly survive, but I'm a cautious person and don't like fans at maximum speed.

I've upgraded to a Ryzen 3900X myself yesterday. Using the stock cooler, FAH easily got it to 92°C, clock speeds capping out at around 4MHz, fan(s) maximized.
I've not done any overclocking since a Pentium 200MMX was the bee's knees, but with the recent UI-based tools that make the process much simpler, I thought I'd start tinkering with the settings a bit

Alas, I thought I'd share my findings, which are limited to my FAH experience only.

1. Because of the constant 100% load across most cores, the temperature hits the ceiling of 95°C so quickly, that the maximum boost clock speed of 4.6 GHz is only reached for split seconds. It then stabilizes at 4GHz for me.
2. According to many reports on the web, as well as my own experience, the board chooses a voltage by default that is way too high. In idle, mine sat at around 1.35 volts. (Yes, it'll regulate down a bit when it runs hot, but I've not seen it go under 1.17 volts.)
3. Because of the high temperatures, the fans spin up to maximum, but it's still not enough to prevent CPU throttling

I have therefore started using the Ryzen Master tool to reduce the Vcore,max. To increase stability and keep a stable temperature, I've also set the clock speed cap to the base frequency (3.8GHz).
I've tested the stability of each setting. First with Ryzen Master, then running FAH for 15 minutes each. And before anybody's wondering: All finished work units were uploaded with a result of WORK_ACK.

With the specific use case of FAH in mind, I also looked at the projected PPD, to estimate the performance hit. With surprising results.

Code: Select all

stock settings:              92°C, ~103W, 110k + 70k = 180k PPD
3800 MHz, 1.175 V:      78°C, ~78W,  104k + 68k = 172k PPD
3800 MHz, 1.150 V:      74°C, ~73W,  106k + 64k = 170k PPD
3800 MHz, 1.075 V:      68°C, ~61W,  106k + 66k = 172k PPD  (= Fan not maxed!)

Of course, there will always be a variance in points, so undervolting does not seem to have an effect on the computational power, unlike suggested on some sites. Again, this may be down to their use-case.
As a result, I've decided that I'll stick with the 1.075 V setting, as it offers a significant power saving (40W * 24h = approx 1 kWh = 30 Eurocents/day) and keeps the computer quiet.

Sources:

• https://www.notebookcheck.net/Undervolt ... 867.0.html

I'll add more sources to this posting once I found them in my browser history... I've looked at a lot of sites.

[foldy]

Another option would be to buy a good CPU cooler and case fans. Your CPU should be silent at full load even at more than 4GHz

[BobWilliams757]

Another option would be to buy a good CPU cooler and case fans. Your CPU should be silent at full load even at more than 4GHz

But on the flip side, why would you want higher power consumption regardless? It appears in this case productivity stayed up, while power consumption and noise stayed down..... but without any new hardware investment. I'd call that a win/win situation for the OP.

The same applies to those splitting cores up into multiple slots, but usually with the intention of increasing PPD without more heat, noise or power consumption. Both methods are wins if you ask me.

[NoMoreQuarantine]

That processor supports AMD Eco-Mode. I would try that first before manually underclocking. https://www.techarp.com/computer/ryzen-eco-mode-guide/

[uyaem]

But on the flip side, why would you want higher power consumption regardless? It appears in this case productivity stayed up, while power consumption and noise stayed down..... but without any new hardware investment. I'd call that a win/win situation for the OP.

Yes exactly, that was my point.

I am quite surprised how little you seem to lose considering 40% less power is consumed.
Also, we're heading into summer and this 10m² room gets superheated in July even with the PC off.

[PantherX]

...I have therefore started using the Ryzen Master tool to reduce the Vcore,max. To increase stability and keep a stable temperature, I've also set the clock speed cap to the base frequency (3.8GHz).
I've tested the stability of each setting. First with Ryzen Master, then running FAH for 15 minutes each. And before anybody's wondering: All finished work units were uploaded with a result of WORK_ACK. ...

While I am glad that you found the settings that worked for you, please note that the stability you reach is only for that CPU project. If in future, there are bigger CPU Projects that stresses it differently, then you may discover that your stable setting might not be stable after all. This did happen previously and systems that folded for months 24/7 which were overclocked we suddenly discovered to be unstable. From what I have skimmed read about Ryzen Eco-mode, it might be a safer option since it is being promoted by AMD who manufactured it so there's like likelihood of unstable CPU.

[Nathan_P]

PBO and stock coolers do not go together well. While the AMD cooler is reasonable its really only designed to run everything at stock speed. PBO will boost the cpu as far as the cooling will take it which for the AMD coolers means a fan running at very high speed. Long term I would suggest saving some money and investing in something like a dark rock pro 4 or Noctua HN12/NH14, that way you will be able to enjoy a pretty quiet system and still get the benefits of overclocking.

[uyaem]

While I am glad that you found the settings that worked for you, please note that the stability you reach is only for that CPU project. If in future, there are bigger CPU Projects that stresses it differently, then you may discover that your stable setting might not be stable after all. This did happen previously and systems that folded for months 24/7 which were overclocked we suddenly discovered to be unstable. From what I have skimmed read about Ryzen Eco-mode, it might be a safer option since it is being promoted by AMD who manufactured it so there's like likelihood of unstable CPU.

Thanks! I'll keep watching the log file, just wanted to add: It's not overclocked, it's actually locked at base speed, so that should help with stability.
I'll give Eco a try after Easter when I'm at the PC more.

[UofM.MartinK]

I am very interested in your results with Eco-mode!

I am about to build to my first Zen2-based systems (for FPGA synthesis, circuit simulations and folding if "idle") to replace i7-8086K Intel systems (since I really need ECC which the really nice single-thread-performance Intel CPUs never have). Secondary goal is optimizing CPU-perf/buck and CPU-perf/W, which your setup seems to do.

Which MB did you use? I am looking at the ASUS TUF B450M-PRO GAMING for $100 (2xM.2, 6xSATA, serious power regulation in case overclocking is desired after all, and MicroATX with back IO connectors physically enclose - nice for operating without case).

Happy Easter!

[gordonbb]

An NH-D15 did wonders for my 3900x:

Code: Select all

root@fold7:~# cat /proc/cpuinfo | grep MHz
cpu MHz         : 4055.808
cpu MHz         : 4055.850
cpu MHz         : 4052.056
cpu MHz         : 4056.664
cpu MHz         : 4056.656
cpu MHz         : 4056.638
cpu MHz         : 4016.169
cpu MHz         : 4055.469
cpu MHz         : 4025.955
cpu MHz         : 4056.668
cpu MHz         : 4050.432
cpu MHz         : 4056.614
cpu MHz         : 4055.651
cpu MHz         : 4055.688
cpu MHz         : 4055.574
cpu MHz         : 4056.507
cpu MHz         : 4044.759
cpu MHz         : 4056.484
cpu MHz         : 4014.576
cpu MHz         : 4055.275
cpu MHz         : 4024.631
cpu MHz         : 4056.519
cpu MHz         : 4056.431
cpu MHz         : 4056.465
fold7:~# sensors
k10temp-pci-00c3
Adapter: PCI adapter
CPU die:      +75.8°C  (high = +70.0°C)
CPU ctl:      +75.8°C
...
CPU Vcore:      +1.28 V  (min =  +0.00 V, max =  +3.06 V)
Vcore SoC:      +1.01 V  (min =  +0.00 V, max =  +3.06 V)
CPU Vddp:       +0.91 V  (min =  +0.00 V, max =  +3.06 V)
...
CPU fan:       1433 RPM  (min =    0 RPM)
CPU_Opt_Fan:   1436 RPM  (min =    0 RPM)
Ambient:        +23.0°C  (low  = +127.0°C, high = +127.0°C)  sensor = thermistor
CPU Temp:       +75.0°C  (low  = +127.0°C, high = +127.0°C)  sensor = AMD AMDSI
VRM_MOS Temp:   +50.0°C  (low  =  +0.0°C, high = -125.0°C)  sensor = thermistor

I'll likely take some time to under-volt it.

My 9900k was a melting beast even at stock. I got that running stable at 4.5GHz all-core 1.12V on the mprime torture test but that took a 280mm Liquid Cooler. Folding it now runs 56-62C.

[gordonbb]

I am very interested in your results with Eco-mode!

I am about to build to my first Zen2-based systems (for FPGA synthesis, circuit simulations and folding if "idle") to replace i7-8086K Intel systems (since I really need ECC which the really nice single-thread-performance Intel CPUs never have). Secondary goal is optimizing CPU-perf/buck and CPU-perf/W, which your setup seems to do.

Which MB did you use? I am looking at the ASUS TUF B450M-PRO GAMING for $100 (2xM.2, 6xSATA, serious power regulation in case overclocking is desired after all, and MicroATX with back IO connectors physically enclose - nice for operating without case).

Happy Easter!

Be careful choosing your motherboard. While the board may support ECC it may not report single bit errors properly. Level One Techs has some good resources for this.

[jrweiss]

Hmmm...

I have a Ryzen 3700X with a Noctua NH-D14 cooler, on a MSI X570 MoBo, on stock settings. When Folding on 12 cores, Vcore runs around 1.440V with occasional peaks at 1.456, clocks in the 4.3-4.4GHz range, temps in the mid 70s with occasional peaks in the mid 80s. Vcore at idle is 0.912V. Though it's supposedly a 65W TDP, HWMonitor reports core power consumption peaks at 82W.

FWIW, I have an i7-4770K rig that runs much hotter. When I found it running in the 90s while Folding 6 cores, I tried disabling Turbo mode. That alone reduced temps to around 60, with occasional peaks in the mid 70s. Since I don't do any active work on that machine, the reduced performance doesn't affect anything except the Folding.

I haven't tried playing with voltage limits, and haven't tried overriding automatic voltage regulation since my old Opteron 246 days. Has anyone tried playing with the 3700X? The idea of reducing power consumption while retaining performance is tempting...

[HaloJones]

I just built a Ryzen 3600 on a B450M board and for stability and efficiency I set the clock to 3600 preventing it from boosting. As a result the voltage dropped from 1.4 in the BIOS to 1.1 and the temps are much much lower without really hurting the performance.

These boards try to overclock themselves and they do it by having a high voltage all the time from what I can discern.

[Kebast]

I just built a Ryzen 3600 on a B450M board and for stability and efficiency I set the clock to 3600 preventing it from boosting. As a result the voltage dropped from 1.4 in the BIOS to 1.1 and the temps are much much lower without really hurting the performance.

These boards try to overclock themselves and they do it by having a high voltage all the time from what I can discern.

I might have to play with my 3800x at some point. I left mine on auto thinking it would save power when cores were idle. Now I’m folding pretty much 100% of the time, so might try some manual settings to reduce power and heat. I have a scythe cooler on mine, and the chip stays at 83c all the time.

[puuteknikko]

My 3900X was boosted by PBO to about 4.05GHz. Temps over 80C and voltage mostly around 1.35-1.4V. As my computer most often uses all the cores, I would not get any benefit from those occasional higher boost speeds for single cores, so I set up the multiplier at fixed 40X and core voltage at 1.2V. Stable and keeps much cooler now.

I got the basic idea from here: https://www.tomshardware.com/news/amd-r ... 40157.html