Hi, I have recently built a new shuttle desktop and was comparing single app calculation times between it and an older machine. I got some very weird results, and was wondering if it was partially due to the opperating system or predominantly the hardware: Here are the specs: new machine Shuttle SB61G2, Intel 865G chipset, P4 3C, 1Gb DDR400 (two 512Mb Geil chips for dual channel at CAS2.5) Hyperthreading is on Running Suse 9.0 Professional with the distribution smp kernel old machine Asus P4P800 (I think), P4 2.4C, 1Gb PC1033 (RAMBUS two 512Mb DIMMs ) Runnning Suse 8.1 with the distribution kernel Now this application I am running is really memory intensive, but I was under the impression that dual channel DDR400 was not a lot slower than RAMBUS PC1033. Timing the old machine running my test job produces Real time 1m44 sec CPU time 1m23 sec Running the same job on the new machine give me Real time 4m03 sec CPU time 2m21 sec I am baffled. I have heard here and there that hyperthreading is not very good for single process dominated installations, but this difference is massive. Anothe weird thing, when I run xosview on the new machine it shows continuous system activity in CPU0 up near 100%. At the moment I am all for deactivating hyperthreading and re-installing the kernel. Help would be appreciated. How do I check whether dual channel memory is working properly? Should I use an updated kernel? Which one? Thanks Eugene
Upgrading to the latest kernel seems to have done the trick.
The shuttle now beats the rambus machine by a handfull of seconds.
Does anyone know how much better the 2.6 kernel will handle hyperthreading?
----- Original Message -----
From: "Eugene de Villiers"
participants (1)
-
Eugene de Villiers