A while back I had my Ryzen pre-built overclocked to 3.7, then 3.75GHz. Everything seemed fine for months. I played a lot of games, and the hardest duty I gave it was running the occasional VM or compiling WINE every once in a while.
Then I thought, maybe I really ought to make sure this thing is stable. So I ran Prime95, and tried its equivalent Linux executable from mersenne.org, and quickly found myself looking at 85C+ temperatures. So I dialed back to stock clock speeds fearing for the life of my CPU.
But I started thinkingâŚI ran this thing stably for months. How useful is such a test if it doesnât use the machine realistically? So is there something out there that puts a more reasonable load on the CPU?
A little and a lot; or, as you may have guessed, that depends
Prime95/MPrime is a lot more demanding than needed for most folks out there, but itâs also dependable, which isnât something many âstressâ tests can say.
(have passed test after test in the past, only to fail at something as simple as unzipping huge files for example).
The trick is using it to your advantage;
Stock settings involve a lot of things, including thermal stressing in levels akin to Furmarkâs; you donât need that.
You want a custom set with predefined FTTs so as to check your stability and only your stability (before you say it, no, these are enough for thermals as well). Start with its instructions text, provides enough information.
Now if you wanted something more⌠realistic, it would unfortunately depend on the platform:
For Intel, download Intel XTU (not burning test), trust it for both CPU and RAM. Donât look back.
For AMD⌠nothing similar, sorry. Your best bet is OCCT with AVX unchecked (unless of course it is something you employ); sucks, but, nothing else.
(occt is more demanding than XTU [ie weâre yet again verging on the ânon realisticâ], but beggars canât be choosers)
Whatever you go for, a bit of advice; do not trust AIDA, for anything. Donât even bother.
As to my opinion on your way of handling things? If you can afford a potential shutdown while in the middle of âxâ operation or function or a corruption of Lord knows how many files and folders, by all means. Otherwise, get scientific about this.
Thanks for the thoughts. Firstly, this is not a mission critical machine. Itâs mainly used for gaming and/or tinkering with Fedora. OTOH, it would take a lot of effort to re-create the setup Iâve got. So maybe it is worth being scientific.
What I really need is better cooling. Itâs got a crapy OEM cooler that may not even be up to AMDâs standard. Iâve checked with the OEM, and all theyâll tell me is they use an âOEM AMD mounting solution.â Iâm not sure if that would fit a retail heatsink; Iâd love to go on eBay and pick up someoneâs unloved Wraith Spire.
I know Dell is sub-optimal. But at the time I bought it, RX 580s were selling at insane prices and the only way to get one affordably was to buy a complete pre-built.
Never touched a Dell in my life, googleing Inspiron now; in the meantime, check edited post above, lol
edit 1: their website is cancer⌠i canât find anything definitive about CPU cooler clearance⌠if you know it, do post.
edit 2: apparently they ship with locked BIOSes?.. hmmâŚ
O.K. Unless you can find i) the max clearance and ii) whether itâs even possible to upgrade ( i donât buy ready stuff, so no clue what they do in there)?
Leave it as is.
Youâre 99% running it at a higher voltage than necessary (their only way of ensuring âit worksâ without individually tweaking each and every unit they ship, would take years); it will be a ridiculously high voltage, sufficient enough to keep it stable for you, but hopefully (sorryâŚ) not high enough to be dangerous.
So your sole concern currently is the temp(s) youâre getting. If acceptable, call it a day for now.
Assuming advice is welcome? Next time, steer clear from ready-made stuff. You either overpay, are shoe-horned in some way or other, or are left to deal with situations you wouldnât have had to otherwise.
It is possible to upgrade the CPU to a 2000-series Ryzen. The BIOS is locked, but I can control clock speed and voltage via software (Ryzen Master in Windows, zen-states.py in Linux) - assuming that makes a difference to your conclusion.
Iâve contemplated gutting the Dell, buying an X470 motherboard and transplanting most of the components. Iâd enjoy the increased, um, freedom to do what I want, but Iâm not sure I want to part with an extra $150 - $250 (motherboard, case, HSF, etc.)
Iâve never used RM (for multiple reasons); i do everything from my BIOS, so not in a position to tell you what you can and cannot do with RM (there are settings that become essential for higher OC stability which you need and i doubt RM covers due to their nature, but⌠donât quote me on that). You can definitely improve it some with it though, so good news there.
Youâll almost definitely need a better cooler though.
The motherboard is a secondary thing in your case;
In terms of CPU, what an X470 would offer in terms of improved performance, you (again) need the relevant dissipation to take advantage of.
In terms of RAM, they do sport a reduced latency/slightly higher frequency due to chipset improvements, but i) that also depends on what RAM youâre using (am not optimistic in this), ii) it entails some tweaking, iii) depending on what you do, you may or may not get a benefit out of that.
Start simple:
Download the latest HWinfo, or whatever relevant software exists for Linux if youâre on it.
Download OCCT, disable AVX, run a CPU test.
See where youâre at now thermally (what are your temps, is it throttling? etc.), take it from there.
edit: On a much broader level, iâd at best get me a new cooler, no more. Never âpatchâ, ultimately it ends up costing you even more. Either downright solve or start fresh, ie start saving. Given the period - prices, upcoming releases next year, etc etc - am not sure it would be smart for you to part with more money right now. Thatâs me though.
Depends what youâre doing with it and what your definition of âstableâ is.
A run-away process consuming all CPU that you donât notice for a while will maybe cause similar heat loads to stress-prime in a worst case scenario.
Do you want to build your machine to be OK in ALL circumstances, or borderline dodgy - depending on what is going on?
This is what stress tests are for - measuring how things are WORST CASE in order to ensure that your PC is OK in ALL circumstances.
Me? I have zero tolerance for instability. I donât care if i can get 5-10% more performance through overclocking or whatever, if it means i might get a crash or incorrect data once in a while. I consider modern hardware âfast enoughâ and the risk:reward from OC not worth it. For me.
Interesting results with OCCT. (There are two CPU stress tests; OCCT and Linpack. I assume you were referring to OCCT, but only Linpack has an option to disable AVX.) Dell seems to have set a pretty simple fan curve; as soon as the CPU temperature goes above 75C the fan kicks in until the temperature stabilizes.
Task Manager shows OCCT pegging all threads at 100%, yet it is not generating the same kind of intense heat as Prime95. This is what I was looking for, yet @throâs comment makes me wonder about his hypothetical scenario of a runaway process. OTOH, I kind of doubt that any random process or app on my system is going to generate the same kind of heat and stress as Prime95.
At any rate, the system seems able to successfully maintain 75C under full load (from OCCT) at 3.7GHz. Thanks for the Prime95 alternativeâŚ
Prime95 does produce somewhat unrealistic heat when itâs doing small FFTs (8k to 20kish) but the very high load is what makes it good as a stability test. The higher the load the faster you will uncover instability, so if you use something with a lower load you will have to run it much longer to be reasonably certain the system is stable.
The reason I say âsomewhatâ unrealistic is because Iâve seen similar thermals to Prime95 in real-world workloads like handbrake.
One solution to be able to still use it as a stress test but avoid the âunrealisticâ thermals is to just take the side panel off your case and blast it with a box fan while youâre running the test. Itâs probably a good idea anyway with that system because I doubt the VRM solution is very good, and Iâd be worried about the mosfets and caps.
All that said, Iâve found Ryzen to be different than most CPUs when it comes to stability testing. It can pass high stress loads for long periods but it will crash when basically idle. It seems to have issues going from load to idle in particular. Probably a power management thing. If you really want to be sure of stability you should also try to find a test that varies between high load and no load to give it that âyo-yoâ effect.
I do a custom massively multitasking stress test and let it âburn inâ for about 1/2 an hour. That seems to be enough to let temps stabilize and I donât have the patience to stress test for hours.
Start with the OCCT CPU stress test. Then I add Photoshop with at least 10 big picture files open, Firefox with a ton of tabs and YouTube running, Microsoft Office with a bunch of docs, MSI Kombuster, DOOM for Vulkan and I un-Zip a few files. All of that uses about 10GB RAM.
Adding a 3D ray tracing (try Cinebench, but I use Keyshot) is what usually crashes my system. I have an i5-4690K that has been O/C to 4.5Ghz for years. Ray tracing was somewhat unstable but when I ran my test and changed the O/C to 4.4Ghz, itâs been rock solid.
Thatâs an interesting observation. I found that Ryzen was crashing at stock while idle, but only when I had a GPU in passthrough mode. I recently revisited GPU passthrough on Fedora 28, and so far the system seems perfectly stable - at stock. Hopefully it continues to be so.
@Positron Thatâs certainly a lot of stress, but my goal was to try and model a more realistic workload than Prime95. My âworkloadâ is gaming, running a VM or two and the ever-so-occasional WINE compile. Compiling is probably the heaviest load I put on this CPU, and even that can be controlled with -j parameter. Now in the future I may build out my VM setup to include some virtual servers, but if I do that Iâll probably drop to stock speeds anyway.
