intel noob question

Khannie

Ladies and gents:

I recently bought a P4 2.4GHz with HyperThreading. From what I understand, this is a precursor to multi-core. I was wondering if I should install an SMP kernel to take advantage of it? Or am I way off the mark?

Thanks in advance.

SouperComputer

have a look at this mail list:

http://lists.debian.org/debian-user/2005/09/msg03404.html

Khannie

Thanks. Interesting....though not conclusive.

I did some googling and checked my dual processor xeon box in work, and it says it only has two processors. Surely it should say four? (i.e. two with ht)

/proc/cpuinfo says:

processor       : 0
vendor_id       : GenuineIntel
cpu family      : 15
model           : 2
model name      : Intel(R) Xeon(TM) CPU 3.06GHz
stepping        : 9
cpu MHz         : 3057.018
cache size      : 512 KB
physical id     : 0
siblings        : 2
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 2
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe cid xtpr
bogomips        : 6029.31

processor       : 1
vendor_id       : GenuineIntel
cpu family      : 15
model           : 2
model name      : Intel(R) Xeon(TM) CPU 3.06GHz
stepping        : 9
cpu MHz         : 3057.018
cache size      : 512 KB
physical id     : 3
siblings        : 2
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 2
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe cid xtpr
bogomips        : 6094.84

Khannie

No "ht" on the xeon sticker....maybe I'm wrong with the ht assumption. WTF is the point of the xeon then? 512KB cache is the same as the P4. *grump*

At least I didn't pay for the work machine (though I did pay for the P4 at home).

HomunQlus

Hyperthreading means basically two processors. The information is correct. I have such a Pentium 4 HT Processor myself, in every OS it says 2 processors, so the information is correct.

Khannie

Well....that answer that. Thanks lads.

I think I confused you though Homun.....I ended up talking about two machines. 1 with 1 P4 in it, and another with 2 Xeons in it. I initially expected the xeon machine to show 4 processors with an SMP kernel.

WizZard

Yes, use an SMP kernel. If you're using a compile from source distro (like Gentoo) there are a few gcc/make options you can use aswell.

Khannie

Thanks wiz.

When you say gcc / make options; Is this apart from the actual kernel configuration? (make menuconfig, etc.)

I'd tend to just use the generic apt-get one out of laziness (apt-get....I love you), but I could be convinced to compile if it would give me some extra performance points.

WizZard

Yes, it's apart from the kernel config. Things like parallel make and certain optimisations that allow you to compile faster.
(I'm not going to get into which is faster, compile from source, or binary)
The extra options might give extra performance points - I know it does on Gentoo but I don't know if they're even possible on Debian. There probably is a way to manually use them, but according to this there seems to be no automagic way to do it.

deRanged

HomunQlus wrote:

Hyperthreading means basically two processors.

no it doesn't. it means that you can have two threads running on a single processor, giving 2 *virtual* cpus. This is grand in a machine with a low average load, but in a heavily loaded machine like a compute server it causes slowdown (due to thread swapping and such). HT is not recommended for servers.

One of the first things I do in all my compute servers is disable it.

this is the output from top on a machine I got yesterday - dual xeons with HT enabled.

top - 10:11:42 up 17:27, 1 user, load average: 0.00, 0.00, 0.00
Tasks: 71 total, 2 running, 69 sleeping, 0 stopped, 0 zombie
Cpu0 : 0.0% us, 0.0% sy, 0.0% ni, 100.0% id, 0.0% wa, 0.0% hi, 0.0% si
Cpu1 : 0.0% us, 0.0% sy, 0.0% ni, 100.0% id, 0.0% wa, 0.0% hi, 0.0% si
Cpu2 : 0.0% us, 0.0% sy, 0.0% ni, 100.0% id, 0.0% wa, 0.0% hi, 0.0% si
Cpu3 : 0.0% us, 0.0% sy, 0.0% ni, 100.0% id, 0.0% wa, 0.0% hi, 0.0% si
Mem: 2074668k total, 860864k used, 1213804k free, 20876k buffers
Swap: 1951856k total, 0k used, 1951856k free, 711024k cached


it shows 4 cpus, but if I run a big single threaded compile I'll see load across 2 cpus - 60/40 or whatever.

Khannie

deRanged wrote:

no it doesn't. it means that you can have two threads running on a single processor, giving 2 *virtual* cpus. This is grand in a machine with a low average load, but in a heavily loaded machine like a compute server it causes slowdown (due to thread swapping and such). HT is not recommended for servers.

One of the first things I do in all my compute servers is disable it.

It's clearly disabled on my local machine. I use it as a server box (i.e. others are free to use the resources on my machine, and do), but the typical load is low (on average, less than 1 full CPU is in use at any one time). It does see periods of intense activity, but they tend to be relatively short lived.

At night I have it running BOINC on the aids@home thing. Would that benefit from hyperthreading 4 processes? (Edit: My work machine already has a 2.6 SMP kernel)

deRanged

Khannie wrote:

At night I have it running BOINC on the aids@home thing. Would that benefit from hyperthreading 4 processes? (Edit: My work machine already has a 2.6 SMP kernel)

I doubt it - I don't know too much about how aids@home works but with HT you've still only got 1 (or whatever) physical cpu. so if you've two threads running on virtual cpus with HT, the most you can get is the performance from 1 cpu, minus a bit for swapping etc. so I'd say you'll get a slight degredation in performance.

I'm trying to think of a case where you might benefit - a sweep through a dataset could and there's other possibilities too.
I'm sure someone will prove me wrong.