ARCHIV 1999-2006

ARCHIV :: # 1103

G5 Performance - Rumors vs Facts

Autor: wp - Datum: 30.10.2001

While I would be more than glad to see the performance values for the G5 you have published to come true, I have some serious doubts about their validity.

It would require a principal technological breakthrough, exceeding what was hitherto achieved by companies like alpha processor or MIPS. It is generally known that the efficieny of a CPU will decrease when the clock frequency is increased. From the www.spec.org site, I took the following SPEC peak values for three different CPUs, all representing a different "generation" and approach to boosting the performance. The MIPS CPU is finetuned for efficieny, without aiming at high clock speeds, the Intel P4 is vice versa, and the alpha (latest generation) is somewhat in between. The numbers are the highest I found for a particular CPU:

SPECint SPECint/MHz SPECfp SPECfp/MHz
MIPS 14k @ 500 MHz 427 0.85 463 0.93
alpha @ 1001 MHz 621 0.62 756 0.76
Pentium IV @ 2000 MHz 656 0.33 714 0.36

Pretty clear, the efficiency, here expressed in SPEC per MHz, decreases with increasing clock frequencies. You should also note that Intel has made tremendeous improvements with the P4 design, which has an even higher efficiency than e.g. the P3 at 1.6 GHz, iirc (bad benchmarks for the P4 are mostly due to badly adapted commercial software, the CPU has a very high potential). Even the new HP/Intel Itanium fits well into this scheme (0.46 and 0.72 @ 800 MHz).

Now come the alleged values for the G5

@ 1200 MHz 987 0.82 1005 0.84
@ 1400 MHz 1151 0.82 1173 0.84
@ 1600 MHz 1173 0.73 1359 0.85

As you can see, the efficiency almost reaches that of the MIPS at 500 MHz. It is hard to believe, I must say. The strange (too high) value for fp at 1600 MHz should also be noted.

Kommentare