>>19,20
POWER is Pathetic, but the SPARC is Shit.
https://blogs.oracle.com/BestPerf/entry/20130326_sparc_t5_speccpu2006_rate
Look at the 8-CPU results; the POWER7 at 3.92GHz gets 2770 SPECint while the Xeon at 2.4GHz gets 2180 - so the raw score for that POWER7 system was higher, but notice that it needs 1.6x as many threads to get a 1.27x higher result. Looking at execution efficiency, we have
SPARC T5 1.02 result/thread/GHz
POWER7 2.76 result/thread/GHz
POWER7+ 3.30 result/thread/GHz
Xeon 5.68 result/thread/GHz
Opteron 6.96 result/thread/GHz
The 1-CPU SPECint results are...
SPARC T5 1.01 result/thread/GHz
POWER7 2.54 result/thread/GHz
Opteron 6.90 result/thread/GHz
Xeon 7.69 result/thread/GHz
Pretty similar to results I calculated a year ago:
http://dis.4chan.org/read/prog/1359605422/108
Where are the ARMs? There still hasn't been a SPECint'06 result submitted for an ARM-based server or desktop, so I can't make any concrete statements here, but if RISC vs CISC is the differentiator then the trend is clear:
x86 is still superior.
Another thing about those systems above is that the SPARCs and POWERs are very instruction-hungry, have low code density, and require massively wide memory busses to run at peak performance.