2018 Mac mini only as fast as Dell XPS 13?

[Ihchansg]

I just bought a refurb 2018 Mac mini i5 with 8GB of RAM and 512 GB of storage.
The first thing I did was run some benchmarks, and I found that the floating point performance seems lacking.
Specifically, running the Mersenne Prime95 benchmark with 2048k FFT, all cores for 60 sec
Mac mini: 274 iterations/sec
linux desktop w intel i5 8500 cpu: 420 iter/sec
Intel nuc with i5 8259u: 360 iter/sec
2016 Dell XPS 13 with i5 8250u: 261 iter/sec

The mini (65w cpu) is only doing 65% the speed of the 65w i5 8500 linux desktop, which has been set to limit power to 65w in the bios.
The Intel NUC 8i5 with a 28w cpu handily beats the Mac mini, and the XPS 13 with a 15w TDP cpu running on battery power is almost as fast!

Do I have bad hardware?
Or is this expected for the 2018 Mac mini i5?
Can another 2018 Mac mini i5 owner try running the Prime95 benchmark and report what speed they get?
I just want to know if my hardware is bad.

The mini seems fine on integer performance though (Handbrake, compiling software), but lacks in FP test (prime95, tensorflow), always beaten by the 28w NUC.
SSD speed is about 1.4GB/s, which is fine.

 

[3SQ Machine]

End up keeping it? The refurb i5 '18 Mac Mini I had seemed a bit slow, too, but I didn't benchmark it. My issue was the lack of dGPU, and possibly T2, turned out to be a bigger bottleneck than expected for my workflow.

I also wonder if the mini having the 8500B might mean some other gimping

 

[Krevnik]

Do I have bad hardware?
Or is this expected for the 2018 Mac mini i5?
Can another 2018 Mac mini i5 owner try running the Prime95 benchmark and report what speed they get?
I just want to know if my hardware is bad.

The difference between the Mini and the Linux desktop is a little concerning, but it's not clear you've controlled for variables in your testing, or ruled out Prime95 itself as the possible source of the discrepancy.

• What OS are you using on the Mini?
• Can you try installing Linux on the Mini and running the benchmark so it more closely matches the Linux desktop?
• Is the system properly loaded down on the Mac when it runs, or is it maybe using 4 of the 6 cores?

I suspect in this case, because Prime95 is dependent on AVX hardware, if you got a Prime95 build on the Mini that isn't fully AVX enabled on the Mac, you will see differences like what you see here. Because Prime95 isn't exactly written as a benchmark tool, there's not a huge reason for them to be worried too much about slower Mac builds, beyond the fact that Mac users will return fewer results to the overall project. So fixing AVX bugs for Mac users may not even be a big priority.

Unfortunately, since it looks like the latest Mac build is 32-bit only, I can't even run it on my system to even play around with it.

But in general, I warn against using single benchmarks when comparing across platforms. Bugs and code differences between the two builds make comparisons iffy in the best of times, and not useless in the worst of times. Unless you're buying this to run Prime95 for the GIMPS project, it's not even really measuring your workflow.

EDIT: I notice this is an older thread, but I mostly still want to point this out for others as well.

I also wonder if the mini having the 8500B might mean some other gimping

The only difference is the 8500B is using a BGA mounting, not LGA, which requires a socket. Otherwise, it's the same as the regular 8500.

Geekbench in particular gives very similar results:
8500: https://browser.geekbench.com/processors/2199
8500B: https://browser.geekbench.com/processors/2553

And looking at specific results suggests floating point is pretty similar between the two:
i5 Mini: https://browser.geekbench.com/v5/cpu/773400
Lenovo i5 8500: https://browser.geekbench.com/v5/cpu/678229

But even that benchmark starts to show some differences between platform implementations if you look closely enough, such as the image inpainting test. But it's close enough to show that the two CPUs are within the margin of error of these sorts of benchmarks that they are likely the same thing.