Karol Herbst
2016-Jan-13 12:49 UTC
[Nouveau] [PATCH 0/2] allow partly reclocking on chipset
> Samuel Pitoiset <samuel.pitoiset at gmail.com> hat am 13. Januar 2016 um 13:43 > geschrieben: > > Hi! > > Did you check on different Fermi chipsets or only with one variant?currently I only checked that on my nvc1, but I thought I could just send the patches and it is easier for others to try it out this way.> > Are you sure that engine reclocking works as expected on Fermi? Because > enabling it without a strong inspection sounds like a prediction and it > might not work.It seems to work, because I got a huge performance increase in gputest_pixmark_piano, check the second commit for details ;)> > On 01/13/2016 01:25 PM, Karol Herbst wrote: > > some chipset have working engine reclocking, but broken memory reclocking > > like > > Fermi. We should for now, add the functionality to allow partly reclocking > > for > > those. > > > > Allthough this doesn't give as much performance as one might wish, it is > > till > > noticeable and may improve performance enough to be noted. > > > > Karol Herbst (2): > > clk: seperate engine and memory reclock toggles > > clk: allow engine reclock on fermi > > > > drm/nouveau/include/nvkm/subdev/clk.h | 3 ++- > > drm/nouveau/nvkm/subdev/clk/base.c | 21 ++++++++++++++------- > > drm/nouveau/nvkm/subdev/clk/gf100.c | 3 ++- > > drm/nouveau/nvkm/subdev/clk/gk104.c | 3 ++- > > drm/nouveau/nvkm/subdev/clk/gk20a.c | 2 +- > > drm/nouveau/nvkm/subdev/clk/gt215.c | 2 +- > > drm/nouveau/nvkm/subdev/clk/mcp77.c | 3 ++- > > drm/nouveau/nvkm/subdev/clk/nv04.c | 2 +- > > drm/nouveau/nvkm/subdev/clk/nv40.c | 2 +- > > drm/nouveau/nvkm/subdev/clk/nv50.c | 2 +- > > drm/nouveau/nvkm/subdev/clk/priv.h | 6 ++++-- > > 11 files changed, 31 insertions(+), 18 deletions(-) > > > > -- > -Samuel
Samuel Pitoiset
2016-Jan-13 12:54 UTC
[Nouveau] [PATCH 0/2] allow partly reclocking on chipset
On 01/13/2016 01:49 PM, Karol Herbst wrote:>> Samuel Pitoiset <samuel.pitoiset at gmail.com> hat am 13. Januar 2016 um 13:43 >> geschrieben: >> >> Hi! >> >> Did you check on different Fermi chipsets or only with one variant? > > currently I only checked that on my nvc1, but I thought I could just send the > patches and it is easier for others to try it out this way.Yeah, you need to have more feedbacks on different chips. What about fixing memory reclocking instead of this series which is going to be somehow useless when this will be done? I mean memory reclocking probably offers more performance increase that engine reclocking, no?> >> >> Are you sure that engine reclocking works as expected on Fermi? Because >> enabling it without a strong inspection sounds like a prediction and it >> might not work. > > It seems to work, because I got a huge performance increase in > gputest_pixmark_piano, check the second commit for details ;)I did, but 5.1->6.4 for heaven is a good start, but definitely not a huge performance increase in my humble opinion. :-)> >> >> On 01/13/2016 01:25 PM, Karol Herbst wrote: >>> some chipset have working engine reclocking, but broken memory reclocking >>> like >>> Fermi. We should for now, add the functionality to allow partly reclocking >>> for >>> those. >>> >>> Allthough this doesn't give as much performance as one might wish, it is >>> till >>> noticeable and may improve performance enough to be noted. >>> >>> Karol Herbst (2): >>> clk: seperate engine and memory reclock toggles >>> clk: allow engine reclock on fermi >>> >>> drm/nouveau/include/nvkm/subdev/clk.h | 3 ++- >>> drm/nouveau/nvkm/subdev/clk/base.c | 21 ++++++++++++++------- >>> drm/nouveau/nvkm/subdev/clk/gf100.c | 3 ++- >>> drm/nouveau/nvkm/subdev/clk/gk104.c | 3 ++- >>> drm/nouveau/nvkm/subdev/clk/gk20a.c | 2 +- >>> drm/nouveau/nvkm/subdev/clk/gt215.c | 2 +- >>> drm/nouveau/nvkm/subdev/clk/mcp77.c | 3 ++- >>> drm/nouveau/nvkm/subdev/clk/nv04.c | 2 +- >>> drm/nouveau/nvkm/subdev/clk/nv40.c | 2 +- >>> drm/nouveau/nvkm/subdev/clk/nv50.c | 2 +- >>> drm/nouveau/nvkm/subdev/clk/priv.h | 6 ++++-- >>> 11 files changed, 31 insertions(+), 18 deletions(-) >>> >> >> -- >> -Samuel-- -Samuel
Karol Herbst
2016-Jan-13 14:16 UTC
[Nouveau] [PATCH 0/2] allow partly reclocking on chipset
> Samuel Pitoiset <samuel.pitoiset at gmail.com> hat am 13. Januar 2016 um 13:54 > geschrieben: > > On 01/13/2016 01:49 PM, Karol Herbst wrote: > >> Samuel Pitoiset <samuel.pitoiset at gmail.com> hat am 13. Januar 2016 um 13:43 > >> geschrieben: > >> > >> Hi! > >> > >> Did you check on different Fermi chipsets or only with one variant? > > > > currently I only checked that on my nvc1, but I thought I could just send > > the > > patches and it is easier for others to try it out this way. > > Yeah, you need to have more feedbacks on different chips. > What about fixing memory reclocking instead of this series which is > going to be somehow useless when this will be done? > > I mean memory reclocking probably offers more performance increase that > engine reclocking, no? >yes you are right, but I was thinking, that enabling engine reclocking is kind of easy and it may help others already. Also it helps us to figure out problems with engine reclocking earlier.> > > >> > >> Are you sure that engine reclocking works as expected on Fermi? Because > >> enabling it without a strong inspection sounds like a prediction and it > >> might not work. > > > > It seems to work, because I got a huge performance increase in > > gputest_pixmark_piano, check the second commit for details ;) > > I did, but 5.1->6.4 for heaven is a good start, but definitely not a > huge performance increase in my humble opinion. :-) >right, but it is still 25% more> > > >> > >> On 01/13/2016 01:25 PM, Karol Herbst wrote: > >>> some chipset have working engine reclocking, but broken memory reclocking > >>> like > >>> Fermi. We should for now, add the functionality to allow partly reclocking > >>> for > >>> those. > >>> > >>> Allthough this doesn't give as much performance as one might wish, it is > >>> till > >>> noticeable and may improve performance enough to be noted. > >>> > >>> Karol Herbst (2): > >>> clk: seperate engine and memory reclock toggles > >>> clk: allow engine reclock on fermi > >>> > >>> drm/nouveau/include/nvkm/subdev/clk.h | 3 ++- > >>> drm/nouveau/nvkm/subdev/clk/base.c | 21 ++++++++++++++------- > >>> drm/nouveau/nvkm/subdev/clk/gf100.c | 3 ++- > >>> drm/nouveau/nvkm/subdev/clk/gk104.c | 3 ++- > >>> drm/nouveau/nvkm/subdev/clk/gk20a.c | 2 +- > >>> drm/nouveau/nvkm/subdev/clk/gt215.c | 2 +- > >>> drm/nouveau/nvkm/subdev/clk/mcp77.c | 3 ++- > >>> drm/nouveau/nvkm/subdev/clk/nv04.c | 2 +- > >>> drm/nouveau/nvkm/subdev/clk/nv40.c | 2 +- > >>> drm/nouveau/nvkm/subdev/clk/nv50.c | 2 +- > >>> drm/nouveau/nvkm/subdev/clk/priv.h | 6 ++++-- > >>> 11 files changed, 31 insertions(+), 18 deletions(-) > >>> > >> > >> -- > >> -Samuel > > -- > -Samuel
Possibly Parallel Threads
- [PATCH 0/2] allow partly reclocking on chipset
- [PATCH 0/2] allow partly reclocking on chipset
- [PATCH 0/2] allow partly reclocking on chipset
- [PATCH 1/2] fb/ramgf100: disable memory reclocking by default
- [PATCH] clk: allow config option to enable reclocking