alience 說:期待Dothan的數據+1
不過V大
比同時脈性能不是光靠L2 cache,momery BW/latency就可以看出端睨
還有a little more efficient integer core出自何處哩??
Vanderpool 說:every uArch has his weakness and strength, because different workload has diff requirements to the processor. For this kind of workload, cache size/latency, memory latency and integer core is the most important. So I just mentioned these factors.
Dothan core is very efficient in integer code, because of the uops fusing, dedicated stack engine and very accurate branch prediction unit. All these features will not benefits FP performance much. you can check out the Speccpu2000 score, Dothan 2.26G can compete with 2.6G Opteron in specint2k. it's the most efficient integer core in the world now.
alience 說:SPEC的測試跟cache/momery BW都有很大關係
且還有compiler最佳化的差異
光用這個告訴我Dothan整數運算較K8強我是無法同意的
不同設計的cpu core之間同時脈性能比較不能用這些表面上的東西來比
用cache/memory BW&latency,branch prediction比都很奇怪
且實測數據出來了,Dothan大敗
請問這又要怎麼解釋??
jasonyang 說:http://www.aceshardware.com/SPECmin...&nct=256&cpcf=1&cpct=4&mf=250&mt=3800&o=0&o=1
近年來,許多 cpu 在架構沒有改進的情況下,只是增加 L3 cache 而能夠在 SPECint2000 & SPECfp2000 獲得不小的性能增長,尤其又加上 compiler 針對 SPEC 來最佳化取得較高的成績,SPEC 反映出 cpu 真實性能越來越不準了。像是 Itanium2 與 xeon 就是很好的例子。
雖然一般的評比網站上測試 dothan 與 939 k8 是互有勝負,但 dothan 輸多贏少,不過 dothan 2.6Ghz 性能應該不會輸 FX-60 這麼多才是,除非是浮點性能,這次輸了這麼離譜,可能需要檢查是否測試環境上有無問題。