An isolated case.
Exchange servers perform best with all kinds of optimizations that are not beneficial to most people. Go through the rest of mstechnet's optimizations, at least half of them are bad ideas for most uses.
I don't run an Exchange server, and I highly doubt more than a handful of people on OCN do either.
SQL Server & Peoplesoft recommend HT be turned OFF: [ MS ]
There are certainly a few cases where HT provides no help, or even hurts. They are very rare in desktop computing.
General usage, games, content creation, multimedia, archiving, multi-tasking, virtualization, and so on, and so on, are virtually never harmed by hyperthreading, and many of them are helped. Some benefit very significantly.
Linpack and SPEC are 100% synthetic. They exist as benchmarks only, with few to no equivalents that do real work.
Benchmarks like these are highly optimizied to load CPUs as much and as efficiently as possible.
Hyperthreading uses the gaps in an execution pipeline caused by inefficiencies. Without any gaps there is no lost performance to recoup.
I've pointed out in the past that the heat difference between HT on and off in LinX is no where near as extreme as with other stress tests. I've also used things like Perfmon to show how many instructions were retired by both programs. Linpack is a very efficient program, one should not expect much of a benefit from HT. The vast majority of real programs are not as efficient as Linpack. If everyone goes back to hand tuning everything in assembly language, then things might be different. However, this is very unlikely to happen.
Images from a test I did 18 months ago on my X3350:
See the difference? IBT (a linpack test) has almost double the IPC of Prime95, on the exact same setup. It's simply written to do more work. It leaves very little unused. Prime95, more efficient than most programs out there (and one that is part of a real distributed computing project), is a very different story.
For me, Avidemux using x264 to encode videos often takes 30% less time to complete the same task with HT enabled.
Archiving with 7-zip is often 40-50% faster with HT enabled, on my system. Putting 13,000 files into a 40GB archive on ultra compression takes hours on my 4.2GHz i7. Turning HT on can save me 90-120 minutes, easily.
Doing anything CPU intensive (for my CPU) along with anything else goes
much faster with HT enabled.
If you want screen shots and specific results, I'll do some later today. I am not at my desktop right now. My laptop has hyperthreading, but I can't disable it in Acer's BIOS.
Originally Posted by wuttz
Your examples of why HT isn't useful are either specialized and purely synthetic benchmarks, or are certain types of server loads that apply to almost no one here.
My examples of where HT is a big help, are things I, and many other people on OCN, do every day, to one degree or another.Edited by Blameless - 5/21/10 at 2:02pm