https://community.intel.com/t5/Intel-Moderncode-for-Parallel/Strange-perfomance-variations-on-Xeon-with-hyperthreading-MMX/m-p/960126#M5305 <DIV></DIV> <DIV></DIV> <DIV></DIV> <DIV></DIV> <DIV>I have optimized implementation of string alignment algorithm. I am using SSE2 intructions heavily that gives me on average 7 fold speedup. There is about 32 kb read only buffer thatis shared among multiple threads. Each thread requires individual read/write bufferof at most128Kb. Each thread also uses individual read only buffer with the size of about 64Kb.</DIV> <DIV></DIV> <DIV>Now, the numbers I get:</DIV> <DIV></DIV> <DIV>Higher number better (linear to the time)</DIV> <DIV></DIV> <DIV>Single Xeon 2.8 box with HT enabled:</DIV> <DIV>single threadperfomance 312</DIV> <DIV>two thread perfomance: 604</DIV> <DIV></DIV> <DIV>Dual Xeon 2.8 box with HT enabled:</DIV> <DIV>single thread perfomance: 295</DIV> <DIV>dual thread perfomance: 500</DIV> <DIV>three threads: 445</DIV> <DIV>four threads: 390</DIV> <DIV></DIV> <DIV>Single P4 3.0 Box 2Mb cache with HT delivers:</DIV> <DIV>single threadperfomance: 295</DIV> <DIV>two thread perfomance: 500</DIV> <DIV>four thread pefomance: 460</DIV> <DIV></DIV> <DIV>My 1.7 Centrino laptop:</DIV> <DIV>single thread: 439</DIV> <DIV>two threads: 425</DIV> <DIV>three threads:425</DIV> <DIV>four threads:410</DIV> <DIV></DIV> <DIV>Is there good explanation?</DIV> <DIV></DIV> <DIV>I am buffled. Something to do with cache?<BR />Please help me to understand it.</DIV> <P>Message Edited by chum on <SPAN class="date_text">11-28-2005</SPAN> <SPAN class="time_text">11:20 PM</SPAN></P><P>Message Edited by chum on <SPAN class="date_text">11-28-2005</SPAN> <SPAN class="time_text">11:23 PM</SPAN></P> Tue, 29 Nov 2005 15:12:08 GMT chum 2005-11-29T15:12:08Z https://community.intel.com/t5/Intel-Moderncode-for-Parallel/Strange-perfomance-variations-on-Xeon-with-hyperthreading-MMX/m-p/960126#M5305 <DIV></DIV> <DIV></DIV> <DIV></DIV> <DIV></DIV> <DIV>I have optimized implementation of string alignment algorithm. I am using SSE2 intructions heavily that gives me on average 7 fold speedup. There is about 32 kb read only buffer thatis shared among multiple threads. Each thread requires individual read/write bufferof at most128Kb. Each thread also uses individual read only buffer with the size of about 64Kb.</DIV> <DIV></DIV> <DIV>Now, the numbers I get:</DIV> <DIV></DIV> <DIV>Higher number better (linear to the time)</DIV> <DIV></DIV> <DIV>Single Xeon 2.8 box with HT enabled:</DIV> <DIV>single threadperfomance 312</DIV> <DIV>two thread perfomance: 604</DIV> <DIV></DIV> <DIV>Dual Xeon 2.8 box with HT enabled:</DIV> <DIV>single thread perfomance: 295</DIV> <DIV>dual thread perfomance: 500</DIV> <DIV>three threads: 445</DIV> <DIV>four threads: 390</DIV> <DIV></DIV> <DIV>Single P4 3.0 Box 2Mb cache with HT delivers:</DIV> <DIV>single threadperfomance: 295</DIV> <DIV>two thread perfomance: 500</DIV> <DIV>four thread pefomance: 460</DIV> <DIV></DIV> <DIV>My 1.7 Centrino laptop:</DIV> <DIV>single thread: 439</DIV> <DIV>two threads: 425</DIV> <DIV>three threads:425</DIV> <DIV>four threads:410</DIV> <DIV></DIV> <DIV>Is there good explanation?</DIV> <DIV></DIV> <DIV>I am buffled. Something to do with cache?<BR />Please help me to understand it.</DIV> <P>Message Edited by chum on <SPAN class="date_text">11-28-2005</SPAN> <SPAN class="time_text">11:20 PM</SPAN></P><P>Message Edited by chum on <SPAN class="date_text">11-28-2005</SPAN> <SPAN class="time_text">11:23 PM</SPAN></P> Tue, 29 Nov 2005 15:12:08 GMT https://community.intel.com/t5/Intel-Moderncode-for-Parallel/Strange-perfomance-variations-on-Xeon-with-hyperthreading-MMX/m-p/960126#M5305 chum 2005-11-29T15:12:08Z