0.1 and 0.4W when compared to what we had with i915_enable_rc6=1. But in any case, this is all highly experimental and I'll do more testing with it :). -- Eugeni Dodonov <http://eugeni.dodonov.net/> --f46d041702733b858704b2e83d24 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: quoted-printable <div class=3D"gmail_quote">On Tue, Nov 29, 2011 at 19:42, Ben Widawsky <spa= n dir=3D"ltr"><<a href=3D"mailto:ben at bwidawsk.net">ben at bwidawsk.net</a>&= gt;</span> wrote:<br><blockquote class=3D"gmail_quote" style=3D"margin:0 0 = 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;"> <div class=3D"im">On Tue, Nov 29, 2011 at 01:01:31PM +0000, Chris Wilson wr= ote:<br> > On Tue, 29 Nov 2011 10:55:04 -0200, Eugeni Dodonov <<a href=3D"mail= to:eugeni.dodonov at intel.com">eugeni.dodonov at intel.com</a>> wrote:<br> > > This should allow even more energy saving when GPU is not in use.= <br> > > According to the testing, this state results in around 0.1 - 0.4 = W better<br> > > power usage.<br> > ><br> > > No issues or regressions observed so far, but additional testing = is<br> > > certainly welcome.<br> ><br> > The docs I saw said "not implemented; do not use". Do we hav= e it on good<br> > authority that this is safe and useful to enable? And doesn't it<b= r> > require programming of more transition thresholds?<br> > -Chris<br> <br> </div>Yes, I think we do have to program more stuff to make this work. Perh= aps<br> the BIOS puts in decent values for these registers though? We would have<br= > to restore those on reset and resume I'd guess. If BIOS doesn't use= <br> anything there, you probably aren't even entering these states.<br> <br> Also, for posterity, there are 3 rc6 states, rc6, deep rc6, and deepest<br> rc6. I think deepest rc6 was recommended to avoid (though I don't recal= l<br> a specific root caused issue, just some data from Jesse, from the<br> windows team that it didn't seem stable). And I think deepest rc6 is<br= > also referred to as rc7 sometimes.<br></blockquote><div><br>We already setu= p the variables for both deep and deepest rc6 in our driver (GEN6_RC6p_* an= d GEN6_RC6pp_*), but we weren't using this additional state previously = - if I understood the documentation and the code correctly, we do enable pl= ain rc6 and deep rc6 currently. I haven't found any indications which would te= ll to avoid it in the latest docs, and I also haven't seen any regressi= ons or issues with it being enabled on any of the machines, so I thought it= would be worth trying that=20 additional state as well.<br><br>From the testing which QA did for this pat= ch, looks like we save between 0.1 and 0.4W when compared to what we had wi= th i915_enable_rc6=3D1.<br></div></div><br>But in any case, this is all hig= hly experimental and I'll do more testing with it :).<br clear=3D"all"> <br>-- <br>Eugeni Dodonov<a href=3D"http://eugeni.dodonov.net/"; target=3D"_= blank"><br></a><br> --f46d041702733b858704b2e83d24--
