On Saturday 24 January 2004 16:59, Andreas Winkelmann wrote:

Am Samstag, 24. Januar 2004 14:52 schrieb Andreas Winkelmann:

...

Look at the Features-Line, the "ht" stands for "HyperThreading". But if you see only one cpu, check for the right kernel or maybe the kernel-options.

Hmm, i overlooked something. ht works with 2.4.x. You only need an smp-kernel.

I have a little question on that: Do i have to self-compile Pentium-4-SMP optimized kernel 2.6x (say, the one from /people/kraxel) or the ordinary pentium-based SMP kernel will be enough? I have Pentium 2.6Hz with HT on Asus P4P800 Deluxe Some time ago, when i just installed SuSE 9.0, running with HT enabled, with defaults 2.4.21-SMP kernel i had numerous faults in the filesystem (reiserfs), which resulted in complete destroying of data. -- this seemed pretty strange to me, since i have dual-P-III with reiser and it runs OK. (now i run ordinary default 2.6.1 kernel with HT disabled -- it weemd to work fine.)

-- Andreas

On Saturday 24 January 2004 09:47, Alexandr Malusek wrote:

"Steven T. Hatton" writes:

...

One reason I want to try the 2.6 kernel is because I hear it has support for hyperthreading, and I think my CPU has the feature.

2.4 supports HT too but its scheduler is a bit confused by this feature. (It schedules jobs as if two CPUs were present. Low priority jobs thus get more CPU cycles per unit time than they would get if only a single CPU was present). Perhaps this is fixed in 2.6.

That makes Andraes's comment about using the smp kernel more understandable. I guess this is kind of like internal smp or something?

...

Can someone tell me if the CPU with the following /sbin/hwinfo --cpu output is hyperthreaded? ... fpu,vme,de,pse,tsc,msr,pae,mce,cx8,apic,sep,mtrr,pge,mca,cmov,pat,pse36,c lflush,dts,acpi,mmx,fxsr,sse,sse2,ss,ht,tm

IMHO ht stands for it.

That was my guess, but Intel doesn't make it easy to match the dump the chip gives with the data on their site.

...

Will the 2.6 kernel take advantage of that?

Yes, but I had to activate HT in BIOS and use the SMP kernel.

I looked in my BIOS and didn't see a setting for it. I even upgraded the BIOS hoping that would change the options, but still see no setting. It's an Intel board, so I would like to think it's supported. How can you tell it's running in HT mode?

...

Will it make much difference?

I don't see any significant increase in speed. But I don't use applications optimized for HT.

I was under the impression this was below the application level. That the CPU and OS would 'automagically' handle the HT. Then again, my knowledge of the technology is limited to reading the sales glossy. I have a celeron in another box, and one place I see a big difference between it and a comperable Pentium is with ssh, or so it seems. If its simply a question of better utilization of time slices, perhaps it will boost the compile rate for mozilla and (12 hour) KDE builds.

-- A.M.

STH

On Saturday 24 January 2004 13:43, Andreas Winkelmann wrote:

Am Samstag, 24. Januar 2004 17:46 schrieb Steven T. Hatton:

...

...

Yes, but I had to activate HT in BIOS and use the SMP kernel.

I looked in my BIOS and didn't see a setting for it. I even upgraded the BIOS hoping that would change the options, but still see no setting. It's an Intel board, so I would like to think it's supported. How can you tell it's running in HT mode?

www.intel.com ?

That's where I was looking. I'll have to look a the BIOS next time I reboot. I can't seem to find a way to get a description I can reliably match 1-1 to the codes on their web site. I _just_now_ booted with the smp kernel for the first time. I haven't looked into setting, etc. at all.

Which kernel do you use?

# uname -r

2.4.21-144-smp

Which kernel-options?

# cat /proc/cmdline

root=/dev/hdb3 vga=0x317 hdc=ide-scsi hdclun=0 desktop splash=silent I don't see any indication of HT.

-- Andreas

STH

"Steven T. Hatton" writes:

...

Yes, but I had to activate HT in BIOS and use the SMP kernel.

I looked in my BIOS and didn't see a setting for it. I even upgraded the BIOS hoping that would change the options, but still see no setting.

The MicroStar boards which I use (http://www.msicomputer.com/) have the setting. If HT is enabled in their BIOS then I see 4 CPUs in Linux, otherwise I see just 2 CPUs there.

...

I don't see any significant increase in speed. But I don't use applications optimized for HT.

I was under the impression this was below the application level.

Yes, but some instructions break the simultaneous execution of two "streams" of instructions in the CPU. If for instance a multithreaded application is optimized for HT, then these collisions are rare and both threads run at "full" speed. In my case, these collisions are frequent and therefore only one stream runs at full speed, the other is blocked. -- A.M.

Anders Johansson writes:

On Saturday 24 January 2004 20:40, Alexandr Malusek wrote:

...

Yes, but some instructions break the simultaneous execution of two "streams" of instructions in the CPU. If for instance a multithreaded application is optimized for HT, then these collisions are rare and both threads run at "full" speed. In my case, these collisions are frequent and therefore only one stream runs at full speed, the other is blocked.

But surely HT isn't fully dual core?!

Yes, the CPU isn't fully dual core. Actually, it's a single core but some execution units may work in parallel. Anyway, the description above is oversimplified.

I was led to believe that HT meant that when one thread was blocking for I/O, another thread could use the idle CPU cycles while the first thread completed, thus increasing the CPU utilization

HT is much more than that, see e.g. http://www.intel.com/technology/hyperthread/download/25000802.pdf This document doesn't contain the marketing rubbish and is easy to read and understand.

I'm not an expert in CPU design so

Neither I. I just read the document above several years ago. -- A.M.

On Saturday 24 January 2004 22:49, Alexandr Malusek wrote:

Alexandr Malusek writes:

...

...

I'm not an expert in CPU design so

Neither I. I just read the document above several years ago.

Having checked previous e-mails in this thread I conclude that here I unknowingly responded to myself and not to Anders or anyone else. I think the disease is called schisofrenia :-).

'sok, it was an accurate description of me anyway :)