| 1 |
On 12/30/2010 12:21 AM, Mick wrote: |
| 2 |
> On Thursday 30 December 2010 03:16:05 Bill Longman wrote: |
| 3 |
> |
| 4 |
> This is what my i7 Q is showing: |
| 5 |
> |
| 6 |
> Handle 0x0005, DMI type 4, 42 bytes |
| 7 |
> Processor Information |
| 8 |
> Socket Designation: U2E1 |
| 9 |
> Type: Central Processor |
| 10 |
> Family: <OUT OF SPEC> |
| 11 |
> Manufacturer: Intel |
| 12 |
> ID: E5 06 01 00 FF FB EB BF |
| 13 |
> Version: CPU Version |
| 14 |
> Voltage: 3.3 V |
| 15 |
> External Clock: 133 MHz |
| 16 |
> Max Speed: 4096 MHz <--my max speed with turbo should be 2.8GHz?--> |
| 17 |
> Current Speed: 1600 MHz <--my max speed without turbo--> |
| 18 |
> Status: Populated, Enabled |
| 19 |
> Upgrade: ZIF Socket |
| 20 |
> L1 Cache Handle: 0x0006 |
| 21 |
> L2 Cache Handle: 0x0007 |
| 22 |
> L3 Cache Handle: 0x0008 |
| 23 |
> Serial Number: Not Specified |
| 24 |
> Asset Tag: Not Specified |
| 25 |
> Part Number: Not Specified |
| 26 |
> Core Count: 4 |
| 27 |
> Core Enabled: 4 |
| 28 |
> Thread Count: 8 |
| 29 |
> Characteristics: |
| 30 |
> 64-bit capable |
| 31 |
> |
| 32 |
> My turbo reading leads me to think that the dmidecode is not necessarily |
| 33 |
> reporting what the CPU can do. |
| 34 |
I've always laughed at the shoddy workmanship of DMI. I don't know the |
| 35 |
percentage of boards with "To Be Filled By O.E.M." in them, but most of |
| 36 |
it is useful information. Here's another snapshot of idle stats, so it |
| 37 |
looks like it's handling the idle routines okay: |
| 38 |
/sys/devices/system/cpu/cpu0/cpuidle |
| 39 |
09:05:05# for a in {0..3}; do echo === state$a ===; ls state$a;cat |
| 40 |
state$a/*; done |
| 41 |
=== state0 === |
| 42 |
desc latency name power time usage |
| 43 |
CPUIDLE CORE POLL IDLE |
| 44 |
0 |
| 45 |
C0 |
| 46 |
4294967295 |
| 47 |
24712609 |
| 48 |
9234 |
| 49 |
=== state1 === |
| 50 |
desc latency name power time usage |
| 51 |
MWAIT 0x00 |
| 52 |
3 |
| 53 |
NHM-C1 |
| 54 |
4294967294 |
| 55 |
530943161 |
| 56 |
2079744 |
| 57 |
=== state2 === |
| 58 |
desc latency name power time usage |
| 59 |
MWAIT 0x10 |
| 60 |
20 |
| 61 |
NHM-C3 |
| 62 |
4294967293 |
| 63 |
4419507305 |
| 64 |
4950330 |
| 65 |
=== state3 === |
| 66 |
desc latency name power time usage |
| 67 |
MWAIT 0x20 |
| 68 |
200 |
| 69 |
NHM-C6 |
| 70 |
4294967292 |
| 71 |
49546556808 |
| 72 |
7879505 |
| 73 |
|
| 74 |
It still points to that "max cpu freq" as the culprit. |
| 75 |
|
| 76 |
I don't recall what kicks in the turbo mode on the i7's. I thought it |
| 77 |
was simply bumping the speed of one core if it could keep the other |
| 78 |
core(s) in the deep sleep state. So, a single threaded process would get |
| 79 |
a speed improvement. If your other cores were running you'd be out of |
| 80 |
luck. Could be wrong. Brain's been running about half a century now, so |
| 81 |
it's wearing out. |
Archives