Hi All,
I am not much impressed with the uncritical hype in this thread about hyperthreading. Although the latest P4 technology is indeed faster than any earlier Intel product, much doubt has been expressed about the true contribution of hyperthreading.
I pointed this out in an earlier post. Whereas I could definitely see decent load sharing in my older multiprocessor systems, not much benefit seemed to show up in my current single processor P4 with hyperthreading enabled.
I had observed this for more than 6 months now and decided to google around a bit. I did not encouter much of the uncritical "yeah yeah hype" of this current ET thread. In fact several articles point out that turning hyperthreading off may actually speedup things:
http://www.newsfactor.com/perl/story/21119.html
What is quite relevant for serious scientific kind of computation is that especially floating point operations don't seem to benefit at all. The article points out that in fact only a single floating point processor is available.
This would make it definitely worthwile to go for true multiprocessor solutions. (nitro keep on going!). This less or more confirms what I seemed to have found out.
P4: better speed? - by all means.
Due to hyperthreading? - doubtful, lots of hype certain.
P.S. My experience is based on the 2.8GHz of 6 months back. Could Intel have improved/corrected something in the latest 3.2GHz?
Be good,
nononsense
I am not much impressed with the uncritical hype in this thread about hyperthreading. Although the latest P4 technology is indeed faster than any earlier Intel product, much doubt has been expressed about the true contribution of hyperthreading.
I pointed this out in an earlier post. Whereas I could definitely see decent load sharing in my older multiprocessor systems, not much benefit seemed to show up in my current single processor P4 with hyperthreading enabled.
I had observed this for more than 6 months now and decided to google around a bit. I did not encouter much of the uncritical "yeah yeah hype" of this current ET thread. In fact several articles point out that turning hyperthreading off may actually speedup things:
http://www.newsfactor.com/perl/story/21119.html
What is quite relevant for serious scientific kind of computation is that especially floating point operations don't seem to benefit at all. The article points out that in fact only a single floating point processor is available.
This would make it definitely worthwile to go for true multiprocessor solutions. (nitro keep on going!). This less or more confirms what I seemed to have found out.
P4: better speed? - by all means.
Due to hyperthreading? - doubtful, lots of hype certain.
P.S. My experience is based on the 2.8GHz of 6 months back. Could Intel have improved/corrected something in the latest 3.2GHz?
Be good,
nononsense
