I’ve been running a Ryzen 5800X3D for a while now, and I’ve noticed that it tends to run pretty hot, especially at idle. I’m seeing spikes from 50-75°C even when just browsing or watching videos. It’s been a bit frustrating, especially with a decent cooler in place (Arctic Freezer 34 eSports with dual fans).
I’ve been looking into undervolting as a way to reduce the temps. I know the BIOS has some options like Curve Optimizer and even Kombo Strike profiles, but I’m still not sure what the best approach is for this CPU.
Has anyone here had success with undervolting their 5800X3D? What settings worked best for you?
Also, I found this article while searching for solutions on Google yesterday. It covers some interesting points about idle temps of 5800X3D and undervolting the 5800X3D, which got me thinking more about trying different tweaks. Thought I’d share it in case anyone else is having similar issues.
Looking forward to hearing your thoughts and tips!
I recently bought one so I’m still figuring the thing out. I was having the same experience at stock because I’m running it in an SFF case with a Noctua NH-L9x65.
I’m currently running just a -28 offset in Curve Optimizer. That helped out a lot the spikes and temperature under load. I tried -30 and it worked kinda sorta but I could see the CPU stressed out and got a crash while idling because voltage was too low. Now it seems to run more comfortably and I’ve yet to have a low load crash. We’ll see.
I don’t suggest you to jump to a negative offset that high, start with -20 and see how the system behaves. You’ll see differences especially under low loads so keep an eye out for those situations. Under load it will probably be fine.
Curve Optimizer won’t limit the power draw, only reduce the voltage at which a certain frequency is met.
If you want to reduce temps more drastically you can reduce power limits by capping PTT, TDC & EDC. You can reduce them manually or by enabling ECO mode in your BIOS’ CPU settings but ECO mode also limits performance as the power limits are harsher than manually tuned.
For reference a good video explaining for a 5800X, but same goes for the X3D version.
Reducing the voltage results in reduced power.
In my tests using a negative offset of 28 reduces PPT and TDC to the same values as a 65W part, which are 88W and 60A. The only value that stays high is EDC that’s around 115A (more or less).
So what gives results is curve optimizer with a negative offset and reducing EDC. The other values can stay the same.
You’re right, I’m aware of ohms law. I could’ve articulated my post better.
My train of thought was that if you’re temperature limited, your CPU automatically throttles down by design.
So by nudging CO a bit, your CPU won’t draw as much power to hit certain frequencies, but now when it’s cooler it can boost for longer and/or higher, which brings the temps right back up again.
Edit: Oh and sidenote: it’s totally possible to set a pretty aggressive offset in CO and have your PC stable enough to pass corecycling tests, benchmarks etc.
But if you now have microstutters in games then I wouldn’t really call that stable.
From my experience that’s a little high but not atypical of AM4 with vcache, 142 W PPT, and what I’d consider somewhat underspeced air cooling. The 5800X3D’s rated for continuous operation at its 90 °C throttle and, if it’s just bumping into a temperature limit, what I’ve noticed is Zen 3 moves the power to the next core in the performance order. So there’s not really a CPU-side functional concern apparent here.
It’s hard to show conclusively but the measurement data I have suggests spikes are mainly about base-IHS contact quality and heatpipe uptake as (at least for most definitions of them) spikes sit inside the fan control loop’s response time. The Freezers are ok but, besides power limiting, undervolting, and TF8/TC-5888/PTM7950, a maybe more effective option for taking off a few degrees is changing to a cooler well matched to the socket. In air that would be Phantom Spirit or Peerless Assassin.
Spikes are often something running on one or maybe two cores, so unless you pull PPT, TDC, and EDC down so far as to restrict single core boost, changing socket limits is likely to have limited effect on spikes. Eco mode may offer somewhat different tradeoffs but dropping 105 W TDP parts to 65 W is usually dramatic for all core throughput (and probably there’ll be no 95 W option). You could check to see if the motherboard’s boosting single core high (~1.4 V) and take a look at pulling it down (~1.3 V is more common in my experience of Zen 3), but the MSI AM4 boards I’ve built were reasonable on this by default.
Sometimes. In my use of negative core offsets I’ve often gotten modestly higher clocks with negligible change in power. I’ve also not usually seen EDC become limiting at Zen 3’s 105 W TDP defaults until seven or eight cores are doing AVX. At four to six core AVX (or more cores scalar, SSE, or lightly IO bound) PPT or temperature’s more frequently the limit encountered. Below that it’s typically how much boost the first few cores in the performance order can take.
I’ve also got parts where no core’s stable to -28 and know of cases where cores had to be put slightly positive not to fail CoreCycler. Nice to have the option of going that low but sometimes the silicon’s just nope.
This is the most common result I get from putting on negative curve offsets.
True, higher frequencies bring heat back up but lowering voltage makes up for it way more because, unless you’re throttling hard, with less temperatures you’re gonna get 100MHz boost on top. Not much more.
I didn’t have micro stutter even at -30. But I crashed while idling so I dialed it back down to -28. I think I got a good chip.
Also there’s no clock stretching to speak of.
Sure, that all makes sense. But these X3D chips seem to be tuned differently. They run with 1.32V~ peak and don’t run into PPT or TDC limits even at stock.
These are the results of running a Cinebench test at -27 CO with EDC limited to 115A. The only peak was in EDC, meanwhile PPT and TDC are way under the stock limits (140W and 95A respectively).
I’m not the one that’s having issues, but I can tell you that these parts can’t run over 1.32V so that’s what at most the CPU asks and will be given.
I think it also runs hot because the cores chiplet has a cache “hat” on top which worsen the thermals in bursty workloads.
Didn’t check stability with core cycler, but using OCCT, Cinebench and different games doesn’t show any sign of instability. No WHEA error to speak of aswell. The system is snappy and runs cool for what it is (an even smaller cooler than what OP is using, NT-H1 paste).
