Subject: Re: RFRAC: p4tcc
To: matthew green <mrg@eterna.com.au>
From: Juan RP <juan@xtrarom.org>
List: tech-kern
Date: 03/15/2007 02:18:06
On Thursday 15 March 2007, matthew green wrote:
> http://www.netbsd.org/~xtraeme/p4tcc.diff
>
> BTW, I added in arch/x86 because some Pentium 4 models have
> the EM64T extension, so that you can run it with NetBSD/amd64
> (this is my case).
>
> Any comment before I commit this?
>
>
> this patch should be optional and disabled by default. in
> testing on P4-M 2.2 systems i was able to only observe less
> than 1W usage drop when setting my clock mod to the lowest
> availble. that's reducing my cpu speed to 1/8th (275mhz
> effectively.) when switching (old) speedstep from 2.2ghz
> to 1.2ghz i observed 5-6W benefit.
Counting that it has the same problems than est and powernow,
and there's very little gain with it, I backed it out.
> at the time i discussed with some x86 folks i know and other
> information i found on the web indicated that my results
> were to be expected. ie, p4 clock modulation is pretty
> useless as a power saving feature, and i believe people
> should have to choose to enable it.
Maybe in your case you only saved 1W, but perhaps you'll save more W in other
modern CPUs? I would like to see some power saving tests.
I don't have a laptop, so anyone else could try that.
> i just found the patch i wrote for this back then:
> http://mail-index.netbsd.org/port-i386/2004/08/17/0002.html
> it makes me realise me that your code is also not MP
> friendly. it will run p4tcc_init() ncpu times on cpu0.
I didn't know that this patch existed :-)
--
http://plog.xtrarom.org/
Juan RP's blog - NetBSD/pkgsrc news in Spanish