[Computer-go] Fuego 04 native Windows

Matthew Woodcraft matthew at woodcraft.me.uk
Sun Jun 13 09:54:16 PDT 2010


Petr Baudis wrote:
> This is not what I see; for me and Pachi, i7 performs about 15% faster
> than Core2 in single thread.  This is "Q9300 @ 2.50GHz" vs. "i7 920 @
> 2.67GHz" so the clock difference certainly is not the sole reason.

Did you disable turbo on the i7? If not, it's likely to be running at
2.8GHz (the TSC still ticks at 2.67 GHz, so libego's kpps/GHz
calculation doesn't take this into account).


> Pachi:
>
> [i7]    2.3kpps
> [core2] 1.9kpps
>

So pachi likes Nehalem a bit better than core 2. That isn't surprising:
these look like much heavier playouts than the ultra-light ones I was
measuring. Is this just playouts, or tree too?



> I also just did a quick test with latest libego:
>
> [i7]
> 27.2497 kpps
> 10.2013 kpps/GHz (clock independent)
>
> [core2]
> 23.043 kpps
> 9.26901 kpps/GHz (clock independent)

> ...and Erik's ancient libego build:
>
> [i7]
> Performance:
>   1000000 playouts
>   7.69283 seconds
>   129.991 kpps
>
> [core2]
> Performance:
>   1000000 playouts
>   8.46945 seconds
>   118.071 kpps

If turbo is on, these two look like equal performance per (real) clock
tick.

The 'ancient' one closely matches the i7 performance I get from a 2008
libego build, but I got rather higher core2 performance. Perhaps just
different compiler behaviour.

-M-



More information about the Computer-go mailing list