NUC machines and distributed computing performance

Celeron-based NUC's (and equivalent uSFF boxes from HP, Dell, Lenovo, etc. with the same CPU) are all pretty low power draw so they should not have cooling issues that might limit performance. You will probably see them hitting designed power limits but that's normal behaviour as designed as necessary as opposed to signs of a thermal issue. (desktops often do not enforce power limits, for perspective)

If you are looking at traditional mainstream laptop line-up NUCs based on Intel Core or AMD Ryzen processors, then yes, you might see some thermal limitations.

i..e: Alder Lake and Raptor Lake:


vs. Jasper Lake:

Not a direct on topic answer, but there will be various levels of SIMD available to you. A CPU that only supports SSE2 will be much slower than a CPU that supports AVX2 or AVX512 250% at the same clock speed

And apologies, I just realized I am somewhat unclear in my initial reply. Will elaborate later, but if you don't want to wait, Anandtech has several long articles on turbo in modern CPUs that are likely even linked from the articles I linked already....

ZeroZanzibar said:

The internet isn't what it used to be to find out basic stuff.

Click to expand...

"We review the top 17 ways to find out basic stuff on the internet [2023] Number 14 will shock you!"

No. It most assuredly is not.

I am curious about the NUC machines but wonder about their performance.

Click to expand...

They are best suited to desktop use, small periods of burst performance within the thermal inertia of the system, and are not well suited to the sustained performance demands of compute. Yet, with some possible creativity, they can often be made to work.

But, I've also read a handful of suggestions that heat limits their performance and so it isn't just a matter of looking up CPU specs. Are these just another flat earther website, or is it a real issue?

Click to expand...

It is a real issue. A typical NUC will throttle and significantly limit performance in almost all cases. However, there are ways to mitigate it. First, most have various settings of thermal behavior in the firmware configuration utility: You must set performance. This ramps fans rapidly and loudly, but will maintain better performance.

Beyond this, if you are free to modify them, consider some of the additional heatsink modifications one may find on Reddit and other places. Simply, you remove the case, cut additional holes, and mount a standard heat pipe based heatsink either in place of the stock heatsink fan combination, or, similarly effective, on top of the stock heatsink. Place the auxiliary heatsink on top of the CPU, directly on top of the heat pipe, with thermal grease. With good fan flow, this will pull additional heat out of the system, allowing better sustained performance.

But, simply, if maximum performance is a concern, a typical desktop chassis and heatsink will radically outperform a NUC. They are not built for sustained high performance compute. Or, at least, older ones are not. I cannot speak to newer ones.

Yes. Tens of minutes of sustained compute on a standard NUC platform will throttle in any reasonable conditions. It may not be massive throttling depending on performance and fan settings, but it will not be maximum sustained performance.

I cannot speak to the small "monitor on top" thin client size desktops, but I expect they suffer similarly.

A midtower with a good heatpipe based heatsink and reasonable fans should run at maximum performance without limits, until you get into exceedingly high power chips.

Yet, does a 10% throttle reduction in speed matter for your workloads if you are looking at older hardware?

I have a cheapo NUC-like i3 jobby and in Windows 10 the fan will spin up within 10 seconds of any hard work, and it will throttle within a minute. I imagine an i7 would just basically be fan-on all the time.

Windows 11 demonstrates less overall fan activity and gives more headroom before throttling.
ChromeOS was even better but will still start throttling in far less than 10 minutes. And I'm guessing that ChromeOS isn't going to be useful for your processing jobs.

The little boxes and their tiny fans just aren't meant for intense, sustained processing. They're not even that good as media servers for occasional transcoding.