JC heeft een kort artikel gepost over het verhaaltje van een of andere stoere Japanner die z'n Athlon van DDR SRAM voorzien heeft. Dit systeem zou op gelijke kloksnelheid een 35% hogere CPUMark99 score geven als een normale Athlon, wat nogal onwaarschijnlijk klinkt:
According to here, K7-600 is 56.0, K7-650 is 60.0. According to here, K7-600 is 55.9, K7-650 is 57.8, K7-700 is 64.0, K7-800 is 69.4. A score of 95.3 would be boosting the Athlon's CPUMark by a factor greater than 35%, clock for clock. According to here, changing the 512K L2 on the PIII to 256K, then doubling the clock and better than halving the latency (among some other stuff) yielded an improvement of 24%.The numbers here pressure us into believing that keeping Athlon's L2 at 512K, but only doubling the clock (not changing the latency) will result in performance increasing by a greater margin. Possible. I've heard that Athlon is hungrier for better caching, as it has more functional units and pipelines and suchlike, so maybe it'll get a greater than expected boost from faster caches. Nonetheless, this is one of those "too good to be true" deals -- If the additional cache improvements on the Thunderbird (lower latency, maybe higher width) still leave it "only" as fast as this Frankenthlon, then you'd find, umm,
, a Tb-800 beating a PIII-1000 in this benchmark. I'm not quite ready to accept all this.