A cause could also be

A cause could also be

Years ago I had a Raspberry Pi 1B in use as a server for a few small services. I later replaced it with a Fujitsu Futro s920, a thin client with an AMD G-series quad-core CPU (GX-415 GA). The AMD G-series is designed for embedded computers. That's why the Fujitsu Thinclient also has a fairly low power consumption. At least in idle this is not much higher than the Raspberry Pi, but with significantly more computing power.

A Raspberry Pi 4 also offers a quad-core CPU with a similar clock rate. Therefore I had the idea to benchmark the hardware a bit. And just for fun I compare it with a Sun Ultra 45 workstation, which has at least a similar CPU clock speed. This one has two UltraSPARC IIIi CPUs, I think from 2003.

Test hardware

– Raspberry Pi 1 Model B: CPUs (ARMv6): 1 x 700 MHz – Raspberry Pi 4 Model B: CPUs (ARMv8): 4 x 1500 MHz – Sun Ultra 45: CPUs (spar4u): 2 x 1600 MHz – Fujitsu S920: CPUs (amd64): 4 x 1500 MHz

Both the Raspberry Pi 1 and 4 are running a 32bit Raspbian. Perhaps not optimal for performance, but due to the small amount of RAM (2 GB) perhaps the better choice for productive use.


As integer benchmark the chess engine stockfish is used. Stockfish offers very good optimization for x86 and also ARM, but not for sparc. Therefore I compiled the whole thing without any special architecture optimizations, just normal compiler optimizations.

As floating point benchmark c-ray is used. This is a simple raytracer written in C.

Results Stockfish

The issue of stockfish bench looks like this:

For comparison, I'll just take Nodes/second. The greater the value here, the better:

Results C-Ray

In the end, C-Ray provides the number of milliseconds that rendering took.

Mean value with standard deviation:


The difference between Raspberry Pi 1 and 4 is gigantic. In the single core benchmark alone, an rpi4 is 14 times faster than an rpi1. But instead of one core, an rpi4 offers 4 CPUs. Also the FP speed is much higher. But that was to be expected.

More exciting is the battle between rpi4 and Fujitsu s920. These are about eye level. The AMD CPU leads in the Integer benchmark, the ARM CPU in the FP benchmark. Interestingly, in the Stockfish benchmark, the standard deviation is much lower for the s920 than for the rpi4. A possible cause could be the thermal problems with the rpi4, which could lead to a reduction of the cpu clock. But with C-Ray it's the other way around. I still don't have an explanation why the values are so much more stable with the rpi4 than with the other test candidates.

A bit weak I find the results at the Ultra 45. Of course, the hardware is getting on a bit, but since we're not talking about small embedded CPUs here, I would have expected more. But one cause could be that I run a quite old OS (Solaris 10) on the Ultra 45 and for compiling I had only gcc 5.5 available. On the rpi4 z.B. Came gcc 8.3 to use.

Now that was a simple CPU benchmark. An IO-heavy benchmark would be interesting. The Ultra 45 will not give up without a fight. Especially the bus architecture is still junk at rpi4, so the Ultra 45 could score here. To be continued (eventually).

Like this post? Please share to your friends:
Leave a Reply

;-) :| :x :twisted: :smile: :shock: :sad: :roll: :razz: :oops: :o :mrgreen: :lol: :idea: :grin: :evil: :cry: :cool: :arrow: :???: :?: :!: