New SPARC 20X Faster Than P4/Athlon??

[Bozo Galora]

or is that benchmark screwy?


link

 

[Diffusion]

That benchmark seems to optimize particularly well, if I remember correctly, it is a neural networking bench. The USIII at 1.05ghz does seem unusually fast though, it scaled up about 10% clock wise from the USIII 900, and shows speed gains of 30-40%, supposedly they redesigned the core, and its REALLY helped that chip.

 

[Sohcan]

<< As the regulars point out at comp.arch this is five times faster than the fastest Alpha chip, more than three times faster than the 900Mhz UltraSPARC predecessor, and 20 times faster than the speediest Athlon or P4. >>

Those numbers are for one of the 13 tests that are part of SPECfp, which you will notice are much higher than the rest. The overall SPECfp score for the 1.05GHz US-III, while very respectable given their previous designs, is by no means whatsoever 5 times faster than the 1GHz Alpha or 20 times faster than the 2GHz P4 or 1.6GHz Athlon XP.

Like Paul DeMone said in the comp.arch post, either Sun performed some weird compiler trick, or there was a bug in reporting the scores. If it is due to an optimization that SPEC doesn't allow, the score may be disqualified.

 

[Mats]

Interesting article, keep us posted BOZO.

 

[Bozo Galora]

but of course

 

[Bozo Galora]

link

 

[MrHelpful]

Too bad it isn't an x86 processor...