ccookseywork
05-05-2004, 10:53 AM
I am in the process of sizing servers and found that they are hyper threaded. Most documents claim that you get a 10 to 20 percent gain in CPU resources with this enabled.
My issue: Currently I use industry standard SpecInt ratings to model resource behavior, however, they do not have a SpecInt rating for Hyper threaded machines that take into account logical processors.
All the documents I have read specifically mention that it is highly unlikely that you would ever see no more than a 20 percent gain.
Suppose you see 50 percent out of 200. All you really have is One processor.
My thought is that you could make the assumption that no more than 20 percent could be gained according to Intel white papers. This would give you an additional 40 percent over the 100 resulting in 140 percent total that could be used.
The result would be approximately 36%. (50/140)
Does anyone have any ideas on this?
My issue: Currently I use industry standard SpecInt ratings to model resource behavior, however, they do not have a SpecInt rating for Hyper threaded machines that take into account logical processors.
All the documents I have read specifically mention that it is highly unlikely that you would ever see no more than a 20 percent gain.
Suppose you see 50 percent out of 200. All you really have is One processor.
My thought is that you could make the assumption that no more than 20 percent could be gained according to Intel white papers. This would give you an additional 40 percent over the 100 resulting in 140 percent total that could be used.
The result would be approximately 36%. (50/140)
Does anyone have any ideas on this?