Followup: Losing time like crazy (1s short per 10s)

Arthur Chance freebsd at qeng-ho.org
Tue Aug 16 12:42:40 UTC 2016


On 16/08/2016 13:04, Martin S. Weber wrote:
> following up to :
> 
> On 2016-08-16 13:39:33, Martin S. Weber wrote:
>> On 2016-08-16 12:33:47, Matthew Seaman wrote:
>>> Could you open a PR with the information you've included in this mail
>>> thread please?  
>>
>> done, #211897. I've included my dmesg and service -e output.
> 
> ..and now my p7 dmesg as well as the p4 one, and a diff between the two,
> which highlights two, peculiarities, as Matthew called them:
> 
> -FreeBSD 10.3-RELEASE-p4 #0: Sat May 28 12:23:44 UTC 2016
> +FreeBSD 10.3-RELEASE-p7 #0: Thu Aug 11 18:38:15 UTC 2016
> (...)
> -CPU: Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz (4000.07-MHz K8-class CPU)
> +CPU: Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz (4439.49-MHz K8-class CPU)
> (...)
> -Timecounter "TSC-low" frequency 2000036900 Hz quality 1000
> +Timecounter "TSC-low" frequency 1109872466 Hz quality 1000
> 
> Not sure which I like better for the culprit, the 4439.49 MHz equals 4.0 Ghz
> or the different timecounter frequencies. The former nicely corresponds with
> the "10%" of time I lose.

For comparison, the same processor under 10.2 (really must find time to
upgrade)

CPU: Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz (3997.77-MHz K8-class CPU)

Timecounter "TSC-low" frequency 1998885348 Hz quality 1000

The only thing I can think of is that I don't have turbo mode enabled:

dev.cpu.0.freq_levels: 4001/88000 4000/88000 3800/81704 etc...
dev.cpu.0.freq: 4000

What does "sysctl dev.cpu.0.freq" show on your machine?

-- 
Schrödinger's cat had 18 half lives.


More information about the freebsd-questions mailing list