performance of Intel "Harpertown" series

[ixfd64]

The technology website TG Daily claims that the X5355 (part of the "Clovertown" series) can sustain 50-60 GFLOPS, and Intel's official indeed confirms that the X5355 can perform up to 63.5 GFLOPS. (I know that FLOPS ratings are meaningless since the actual performance depends on the application, so we can assume that these values are theoretical peaks.) Since none of these sources mention dual-processor systems, it would be safe to assume that these figures refer to single processors. However, AMD's "Barcelona" series is supposedly 40% faster than the "Clovertown" series, although slower than Intel's 45 nm "Harpertown" series.

The Inquirer reports that two X5482 processors (part of the "Harpertown" series) are able to sustain 80-100 GFLOPS, implying that each X5482 can only sustain about 50 GFLOPS. Aren't these processors supposed to be faster than the "Barcelona" and the "Clovertown" series?

But then again, the tests were done on pre-production processors, which probably aren't fully optimized yet. I guess more details will become available once the "Harpertown" series is launched in about seven weeks from now.

[axn]

Originally Posted by ixfd64

Since none of these sources mention dual-processor systems, it would be safe to assume that these figures refer to single processors.

The intel official link you gave lists all dual processor systems only. (Scroll down and read the Configuration Details section)

Originally Posted by ixfd64

The Inquirer reports that two X5482 processors (part of the "Harpertown" series) are able to sustain 80-100 GFLOPS, implying that each X5482 can only sustain about 50 GFLOPS. Aren't these processors supposed to be faster than the "Barcelona" and the "Clovertown" series?

The theoretical FLOPS rating of these CPUs can be calculated by the formula:
Frequency * #Cores * 4 (for Core 2 and Barcelona) or
Frequency * #Cores * 2 (for pre-Core 2 intels and pre-Barcelona AMDs)

So for a 2.66 quadcore xeon, the theoretical peak is 2.66 * 4 * 4 or 42.66 GFLOPS. It'll take two processors to achieve an actual Linpack of 63.5 GFLOPS (yes, that is actual, not theoretical).