| 1 |
On Tue, May 10, 2011 at 9:38 AM, Bill Longman <bill.longman@×××××.com> wrote: |
| 2 |
> On 05/10/2011 09:34 AM, James wrote: |
| 3 |
>> Alan McKinnon <alan.mckinnon <at> gmail.com> writes: |
| 4 |
>> |
| 5 |
>> |
| 6 |
>>> otherwise. Just enable ondemand, disable everything else, and et the kernel |
| 7 |
>>> get on with doing what it does best: |
| 8 |
>> |
| 9 |
>> So this is what you are saying? |
| 10 |
>> |
| 11 |
>> |
| 12 |
>> [*] CPU Frequency scaling │ │ |
| 13 |
>> │ │ [*] Enable CPUfreq debugging │ │ |
| 14 |
>> │ │ <*> CPU frequency translation statistics │ │ |
| 15 |
>> │ │ [ ] CPU frequency translation statistics details │ │ |
| 16 |
>> │ │ Default CPUFreq governor (performance) ---> │ │ |
| 17 |
>> │ │ -*- 'performance' governor │ │ |
| 18 |
>> │ │ < > 'powersave' governor │ │ |
| 19 |
>> │ │ < > 'userspace' governor for userspace frequency scaling│ │ |
| 20 |
>> │ │ <*> 'ondemand' cpufreq policy governor │ │ |
| 21 |
>> │ │ < > 'conservative' cpufreq governor │ │ |
| 22 |
>> │ │ *** CPUFreq processor drivers *** │ │ |
| 23 |
>> │ │ < > Processor Clocking Control interface driver │ │ |
| 24 |
>> │ │ <*> ACPI Processor P-States driver │ │ |
| 25 |
>> │ │ < > AMD Opteron/Athlon64 PowerNow! │ │ |
| 26 |
>> │ │ < > Intel Enhanced SpeedStep (deprecated) │ │ |
| 27 |
>> │ │ < > Intel Pentium 4 clock modulation |
| 28 |
>> |
| 29 |
>> |
| 30 |
> |
| 31 |
> Yes but no. Yes, those are the correct choices, but the default governor |
| 32 |
> should be ondemand. |
| 33 |
|
| 34 |
Or in the case of the OP who is brave enough (or silly enough?) to |
| 35 |
risk the long term reliability of his CPU running it with no fan, |
| 36 |
possibly choose powersave with a specific low clock rate as the |
| 37 |
default and then switch to either ondemand or conservative manually |
| 38 |
when he needs more performance. In a machine such as he's playing with |
| 39 |
I wonder if he really wants ondemand (jumps to max and then slows down |
| 40 |
over time) vs conservative which more slowly ramps up the clock rate |
| 41 |
if the job at hand takes more time. |
| 42 |
|
| 43 |
It's all a trade off of performance vs power & heat. |
| 44 |
|
| 45 |
On my 12 thread server I've played with these two and frankly don't |
| 46 |
see a lot of difference doing any large job. They are both a bot |
| 47 |
slower than running performance, but I save a lot of power (and over |
| 48 |
time money) using them so I'm happy. |
| 49 |
|
| 50 |
- Mark |
Archives