Hi,

I have a P4 2.6c with hyperthreading. I can't use a kernel without multi-cpu support activated (the kernel boot sequence stops). Have you heard about problem with hyperthreading? In fact I tried to compile my kernel without multi-cpu support because linux was very slow...

Thanks

Hyperthreading enabled technically requires an SMP kernel, it might just be a bit of age to your kernel, what 2.4 are you using?

Cheers,

Finegan

turn off hyperthreading in the bios, if you wish.

i asked a friend of mine who was working on it at the
time for intel, if it would be possible to turn it on and
off on the fly, with a utility. he said that idea had popped
up in meetings, but the way hyperthreading was
implemented, that wouldn't be possible without a chip
redesign.

i was trying to get him to bitch the guys into making
hyperthreading where you could turn it on or off on
the fly for testing software with it.

Yvain : what distribution do you use?

Thanks for your answers... I have a gentoo distribution and the kernel must be a 2.4.30 or something like that (I am not on my computer right now and I can't check that more precisely). As whansard told me, I will turn off hyperthreading for the moment and wait for kernel's updates. Do you know how I could measure global performances of my computer under linux? (a benchmark or something like that)

Cheers,
Yvain

I just downloaded kernel-2.4.21 from kernel.org, compiled it, and the hyperthreading works just fine on my system. I have RedHat, and I downloaded the smp kernel from their rpm database, installed it, and tried to reboot - it was a no go. Not sure, but the hyperthreading seems a little unforgiving when it comes to hardware setup.

You can find a lot of different benchmarks on the following site :
http://freshmeat.net/browse/138/?topic_id=138

Nbench, a port of Byte magazine's benchmark, is nice (simple), but dated :
http://www.tux.org/~mayer/linux/bmark.html

You may find the following article interesting :
http://www-106.ibm.com/developerwork...?ca=dgr-lnxw01

a friend of mine wrote this.


http://cedar.intel.com/cgi-bin/ids.d...AL&catCode=BBX

and recommends this

http://cedar.intel.com/cgi-bin/ids.d...sp?catCode=DCK

and he told me this.


We expect 20-30%

>gain on applications that scale well with multiple processors. That
>means single threaded applications (word, games, _most_ consumer apps)
>will get little if any gain and might even degrade a percent or so.
>Applications that we've seen perform the best with HT have been DCC
>(digital content creation) applications like 3ds max, Maya, and
>Photoshop where gains have been up to 70% with some workloads. Gains
>seen are VERY workload dependent. You really can't say that with HT,
>Maya gains 40% because there are some workloads that have no gain and
>some that even degrade. My job have been trying to mitigate those
>losses and help others do the same so that no application slows down
>with HT. One of the big issues that will come up is that some consumer

>applications fail with multiple processors because they were never
>tested for thread safety. When they crash, we hope it's not seen as an

>HT problem...
>
>Unreal ~3%, Quake ~2% gain. Internally, we've had a team that has
>threaded the LAME encoder and we see around a 50% increase.

Many thanks to all of you for your help. Whansard, thanks for your useful links about hyperthreading : it was really interesting.
Cropcircle, I will try your benchmarks later : I have to compare my old athlon with a p4...

Bye