I'm running Unraid on a Ryzen 3900X. A while back I was having all kinds of issues with the OS locking up randomly and requiring me to fully unplug the machine to reboot it. After some research, I found threads like this and this that suggested it was due to bugs in the linux kernel with Ryzen and power-management and C-states. I followed some guides and disabled all of that stuff in my BIOS, and it worked! No more lockups!

 

However, now my CPU is running at 80W idle which is... not good. Power is super expensive where I live and I want to get this down. All of my sleep state troubles were a while back and since then Unraid (and the Linux Kernel) has had multiple updates. So my question is: Is the original issue with Ryzen sleep states still a thing? Have others had success reverting back to more efficient power management in newer versions of Unraid?

It can still be an issue but you don't need to completely disable C-states, just setting the correct power supply idle control should be enough.

Okay good to know. Would this issue be better with a newer generation Ryzen CPU or an Intel one? I'm thinking of swapping out this CPU to use for something else anyway so if it would be better on a different CPU then I could do that.

I worry with this one because I know I've had issues in the past that caused major downtime, and I'm a bit uncomfortable with playing with setting and risking it (particularly since the issue is intermittent, so I won't know if I've broken something until I'm potentially away from home or something and can't be there to rip the plug)

For better general stability I recommend Intel for Unraid/Linux.

The folks who own all of those datacenters running EPYC servers would disagree with you, Jorge.  😜

27 minutes ago, ConnerVT said:

The folks who own all of those datacenters running EPYC servers would disagree with you, Jorge. 

I don't have a problem recommending EPYC, I have one myself, and never had a crash or other stability issues:

 

 

I was referring to consumer CPUs, like Ryzen and Threadripper, those can have some stability issues with Unraid and Linux, though in most cases they can be fixed by using the recommended settings, still for the typical user, I think Intel is likely a better option.

 

 

 

There is really little difference between EPYC and Threadripper/Ryzen.  They all use the same ZEN processor dies, with only the I/O chiplet that glues the processors together and handles connecting PCIe lanes and other connections to the physical package.

 

First Gen Ryzen does have an acknowledged bug in Linux.  My thoughts are that AMD was so focused on having Windows benchmarks that kicked Intel's butt that they didn't do enough to insure there weren't going to be problems in Linux.  Understandable - AMD had nearly no data center business at the time, and Linux isn't as sexy as gaming and Cinebench benchmarks for Dr. Su to flaunt in product release Powerpoint presentations.

 

Each generation has updated how the CPU power management is done on the die.  By ZEN2 (your EPYC) they have addressed the Linux bug.  If they hadn't, the AMD server market share would not have grown to its current 20%+ from a lowly single digit.

 

I have been running a 1st Gen Ryzen 1500 with Unraid for several years.  Was my first server, now is my backup doing NAS only service.  I did see lockups until I set the Idle Current correctly in BIOS, then never a problem.  My main server's 5700G (ZEN3) has been perfect since Day 1.

 

The 3900X (ZEN2) the original poster is using should be sure to follow the info for Ryzen in the Unraid FAQ - Setting the Typical Idle Current in BIOS and not exceeding the RAM speeds from the charts.

20 hours ago, ConnerVT said:

There is really little difference between EPYC and Threadripper/Ryzen.

That's true, but EPYC boards usually don't have the power supply idle setting, at least AFAIK, maybe because idle power consumption is not as important, also I don't remember any EPYC users having issues, even with 1st gen CPUs, and I see many 2nd and 3rd gen Ryzen users still needing to fiddle with the power supply idle setting to get it stable.

 

Multiple Ryzen 7XXX users also report stability issues, but those look unrelated to the power issue, possible just because the platform is too new for the kernel.

 

 

18 minutes ago, JorgeB said:

That's true, but EPYC boards usually don't have the power supply idle setting, at least AFAIK, maybe because idle power consumption is not as important,

 

I would think that would be correct.  Another piece of the puzzle is the AMD motherboard chip set, and the BIOS (and the AGESA it is built with).  The 3xx series had a limited amount of NVRAM.  With the AM4 socket having such a long life, the MB manufactures started to run out of space for their code, as new CPUs were released.  The AGESA started to grow in size, and the pretty things (fancy screens, overclocking, fan control etc.) the MB makers wanted in the BIOS, sometimes settings and such would disappear from the latest version of their BIOS.  I found using the last version of the BIOS released before the next generation of CPU released works best for 1st/2nd gen Ryzen.

 

42 minutes ago, JorgeB said:

Multiple Ryzen 7XXX users also report stability issues, but those look unrelated to the power issue, possible just because the platform is too new for the kernel.

 

Such is to be expected with Linux and any brand new hardware model.  Microsoft and the motherboard manufactures get pre-production information and CPU samples well in advance.  They certainly sign non disclosure agreements, which is something difficult to de for Linux with it being open source.  Early adopter Linux users should expect to have issues.

On 2/4/2024 at 6:29 AM, ConnerVT said:

Early adopter Linux users should expect to have issues.

 

On 2/2/2024 at 4:04 AM, JorgeB said:

For better general stability I recommend Intel for Unraid/Linux.

Sounds like you two are in agreement with each other. It's a nuanced argument, but the broad strokes are clear. Want less hassle in general? Intel. Want possibly more performance per $? Research the crap out of the AMD solution you are targeting and don't expect the new hotness to be trouble free.

You're not wrong Jonathan.  But time flies - The 1st gen Ryzen released April 2017, which soon will be 7 years ago.  So Ryzen is no longer in the early adopter designation.

 

The funny thing is, soon AMD will likely have the more stable platform, as Intel starts updating its lineup with the mix of high performance and efficiency cores, as well as the rebranding of the Core processor line and rolling out more mobile processors as mainstream.  It will be fun to see how things look in a year or two.

43 minutes ago, ConnerVT said:

as Intel starts updating its lineup with the mix of high performance and efficiency cores

We're 3 generations deep into that already and there haven't been any major issues, at most lack of optimization but reliability's fine.