From: Dominik Brodowski Subject: Re: Frequency switch latency Date: Sun, 8 Feb 2004 19:16:02 +0100 Sender: cpufreq-bounces@www.linux.org.uk Message-ID: <20040208181602.GA4891@dominikbrodowski.de> References: <200401291206.06389.stefan@buettcher.org> Mime-Version: 1.0 Content-Type: multipart/mixed; boundary="===============91101450205053958==" Return-path: In-Reply-To: <200401291206.06389.stefan@buettcher.org> List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: cpufreq-bounces+glkc-cpufreq=gmane.org@www.linux.org.uk To: Stefan Buettcher Cc: cpufreq@www.linux.org.uk --===============91101450205053958== Content-Type: multipart/signed; micalg=pgp-sha1; protocol="application/pgp-signature"; boundary="PEIAKu/WMn1b1Hv9" Content-Disposition: inline --PEIAKu/WMn1b1Hv9 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Content-Transfer-Encoding: quoted-printable On Thu, Jan 29, 2004 at 12:06:06PM +0100, Stefan Buettcher wrote: > This is a little off-topic. In case it does not belong here: Sorry. >=20 > I am currently working on a kernel-compiler cooperation for > dynamic voltage/frequency scaling. I am doing this for Intel's > XScale PXA architecture. It has turned out that changing the > frequency on the XScale takes some hundred microseconds, > but voltage scaling takes 5-8 milliseconds (I am working with > some evaluation board with DVS support). Are there numbers > for PowerNow or SpeedStep on the web? I could not find any. AFAIK, there are no such numbers available even though this is a well-known phenomenon. SpeedStep and PowerNow (usually) have fixed frequency/voltage pairs with different voltages for different frequencies, so there is no difference in latencies between different "states"... > Another question: I have seen that there are cases in which > energy can be saved by not only changing the CPU frequency/ > voltage but also the memory frequency. This is especially true > when a process with high cache efficiency is executed. We > simply do not need the memory then. Are there any plans for > integrating memory speed into the CPUfreq code sometime in > the future? Not yet... it hasn't even been discussed yet on this list AFAIK. Nonetheless, IMO it should at least be interconnected to cpufreq or=20 even become part of cpufreq. Dominik --PEIAKu/WMn1b1Hv9 Content-Type: application/pgp-signature Content-Disposition: inline -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.4 (GNU/Linux) iD8DBQFAJnziZ8MDCHJbN8YRAiftAJwPe3i+OLEoZqiwLdUwKdgsytTP3wCfS9OV C4z1ebp7mNYkHIqJjsiWLfA= =DRXH -----END PGP SIGNATURE----- --PEIAKu/WMn1b1Hv9-- --===============91101450205053958== Content-Type: text/plain; charset="us-ascii" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Disposition: inline _______________________________________________ Cpufreq mailing list Cpufreq@www.linux.org.uk http://www.linux.org.uk/mailman/listinfo/cpufreq --===============91101450205053958==--