Now this is interesting:

If the tools that come with openSUSE 11.2 are right, then on my AMD 
Phenom QuadCore box single-threaded apps (e.g. MegaPOV or MCPov ;-)) 
perform doubly poor (or actually something around 6.5-fold): Not only do 
they use just 25% of the total available processing power - by their 
failure to do so they also motivate some instance in the system to 
further reduce that total, by throttling the CPU clock from 2.3 GHz to 
as low as 1.4 GHz (actually 1.15 GHz when idle)...

Does someone have an idea whether this is controlled by the OS or by the 
CPU itself - and where I could turn it off?

Post a reply to this message

clipka <ano### [at] anonymousorg> wrote:
> If the tools that come with openSUSE 11.2 are right, then on my AMD 
> Phenom QuadCore box single-threaded apps (e.g. MegaPOV or MCPov ;-)) 
> perform doubly poor (or actually something around 6.5-fold): Not only do 
> they use just 25% of the total available processing power

  Isn't that the correct number for a single-threaded app running on a
quad-core?

> - by their 
> failure to do so they also motivate some instance in the system to 
> further reduce that total, by throttling the CPU clock from 2.3 GHz to 
> as low as 1.4 GHz (actually 1.15 GHz when idle)...

  Are you sure it's not doing that because of overheating instead?

-- 
                                                          - Warp

Post a reply to this message

Warp schrieb:
> clipka <ano### [at] anonymousorg> wrote:
>> If the tools that come with openSUSE 11.2 are right, then on my AMD 
>> Phenom QuadCore box single-threaded apps (e.g. MegaPOV or MCPov ;-)) 
>> perform doubly poor (or actually something around 6.5-fold): Not only do 
>> they use just 25% of the total available processing power
> 
>   Isn't that the correct number for a single-threaded app running on a
> quad-core?

Yes - so far, so good...

>> - by their 
>> failure to do so they also motivate some instance in the system to 
>> further reduce that total, by throttling the CPU clock from 2.3 GHz to 
>> as low as 1.4 GHz (actually 1.15 GHz when idle)...
> 
>   Are you sure it's not doing that because of overheating instead?

Pretty bloody likely not, as the CPU clock indicator goes /up/ when load 
increases - not the typical reaction to thermal issues, is it? :-P

But I found the solution to the riddle: What I saw was only the 
/average/ clock; when not under full load, openSUSE will apparently put 
some of the cores in a kind of standby mode, throttling their clock to 
half speed, while happily running the processor-hungry single-thread app 
on some other core still clocked at full speed.

Post a reply to this message

> But I found the solution to the riddle: What I saw was only the /average/ 
> clock; when not under full load, openSUSE will apparently put some of the 
> cores in a kind of standby mode, throttling their clock to half speed, 
> while happily running the processor-hungry single-thread app on some other 
> core still clocked at full speed.

Is it really possible to put individual cores into standby or clock a core 
individually?

Is this a feature of AMD or does it work with Intel, too?

What about Intel Quadcores, which are basically two double cores?

Post a reply to this message

TC wrote:
> Is it really possible to put individual cores into standby or clock a core 
> individually?

There's no reason it couldn't be. The parts in your cell phone all have 
individually-controlled power so they can be turned off for the miliseconds 
you're not using them to make the battery last longer. I wouldn't be 
surprised if people worried about it enough (at least the heat if not the 
power consumption) that Intel went and added such to the separate cores.

-- 
Darren New, San Diego CA, USA (PST)
   Is God willing to prevent naglams, but unable?
     Then he is not omnipotent.
   Is he able, but not willing, to prevent naglams?
     Then he is malevolent.

Post a reply to this message

Darren New wrote:
> 
> There's no reason it couldn't be. The parts in your cell phone all have
> individually-controlled power so they can be turned off for the
> miliseconds you're not using them to make the battery last longer. I
> wouldn't be surprised if people worried about it enough (at least the
> heat if not the power consumption) that Intel went and added such to the
> separate cores.
> 

Well, my laptop from early 2007 (C2D T5600) could do it. I could even
fix the other core to run at 1GHz and let the other one throttle between
1 and 1,83GHz due to need of cycles.

So yes, I'd it's common on today's processors.

-Aero

Post a reply to this message

TC schrieb:

> Is it really possible to put individual cores into standby or clock a core 
> individually?

Apparently so. If the interface to the other CPU components run at a 
different frequency than the cores anyway, then it should actually be 
quite straightforward.

> Is this a feature of AMD or does it work with Intel, too?
> 
> What about Intel Quadcores, which are basically two double cores? 

 From what I see on the internet, Intel CPUs can do the same only for 
pairs of cores (at least with the CoreDuo technology). They have some 
alternative called "duty cycle modulation" though, in which they appear 
to stall a core comletely at regular intervals; maybe they can do that 
for individual cores, too.


(Random fun fact: At the frequencies modern CPU cores run at, signals 
can travel no faster than about 10-15 cm per clock cycle: Anything 
faster than that would exceed the vacuum speed of light.)

Post a reply to this message