Hyper-Threading performance Test (on vs off)

[exar333]

The article spent a lot of time talking about how Intel duals-with-HT will never be priced against AMD quads, yet within a few months Clarkdale was doing just that.

That is because it was/is just as fast some 3/4-core AMD offerings for many uses.

HT was never to replace extra cores, but it is a great way to get extra performance for relatively low die-space and/or power usage. It helps the CPU better utilize it's resources in certain applications.

Think of it this way. You can have a 3 ghz dual-core w/ HT that uses 45w or a 2 ghz quad-core w/out HT that uses 65w. The HT gives the dual-core an extra 10-20% of CPU power for just a few watts, helps with multi-tasking, AND allows the clocks to stay higher (3ghz vs 2ghz) for better single-threaded performance.

 

[greenhawk]

iirc intel was forced to use ht do to the out of order instruction and or the
cache hit and misses but again I dont remember.

It was the long pipelines that forced the original need. A long pipeline that misses the expected path of a section of code takes a longer time to re-set back onto the correct path. Tho with the shorter pipelines now present, it makes the biggest advantage of HT mute in my opinion.

Would the results be different on win 7 .

If the OS knew which was a read core and which virtual, it might be able to take advantage of that (ie: assign a light load to the virtual and a heavy to the actual), but I have not read any information on that topic to be sure. If it appears in a similar fashion as AMD's future bulldozer's cpu pairing, then the difference between virtual and real is not worth coding for.

Personally, the one question for HT that I can not find a definit answer for is with the different turbo modes the i7 uses. With normal HT, it either helped or hindered the overall cpu based on the code being run, with Turbo being able to clock single cores to high speeds, then having HT on that core can mean better single thread performance (one fast core, windows sees 2 fast cores), but that assumes windows can tell which "pair" of cores (read+HT) is which so as not to force 2 real cores to full speed when a real core and HT can do the job.

Guess I need to find someone that has a i7 2600 and can check clock speeds with 1,2,3,4 copies of prime , both with HT on and HT off, then report clock speeds.

 

[Voo]

If the OS knew which was a read core and which virtual, it might be able to take advantage of that (ie: assign a light load to the virtual and a heavy to the actual), but I have not read any information on that topic to be sure

Since Vista the OS can distinguish physical/logical cores and also does take that into account when scheduling. The Win7 scheduler actually quite improved on this, so yes the fact WHICH OS was used makes a huge difference for this kind of test.

If this test was done on XP the results are completely meaningless, less so if it was done on Vista, but really the only interesting results would be if the tests were run under Win7..