Low power operational mode of nv50

Low power operational mode of nv50

[Jan Kiszka]

[-- Attachment #1.1: Type: text/plain, Size: 1087 bytes --]

Hi,

what are the missing bits to get nv50-class hardware into lowest power
mode? Even when clocking down to perflvl 0, the GPU temperature of my
Quadro FX 880M is still quickly climbing until it reaches 60 °C and the
fan goes on for a while (I tried up to 2.6.38-rc1 so far - any progress
in nouveau-next?). This is not only annoying but also sucking several
watts more than what is feasible - unfortunate for mobile operation.

The benchmark is still the blob which runs the card at level 0 around 54
°C, and that means fan-less. But as the blob is causing increasing
troubles across suspend/resume, not to speak of all its other
limitations. So I'd rather like to switch ASAP now.

What may cause this different power consumption? As the clocking is
apparently the same (the blob even claims to use a few more MHz), I was
wondering if the less sophisticated firmware of nouveau could be the
reason. Or is the host driver keeping the GPU too busy even in idle state?

Is there anything I could do to help analyzing or even resolving this issue?

Thanks,
Jan



[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 259 bytes --]

[-- Attachment #2: Type: text/plain, Size: 181 bytes --]

_______________________________________________
Nouveau mailing list
Nouveau-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
http://lists.freedesktop.org/mailman/listinfo/nouveau

Re: Low power operational mode of nv50

[Martin Peres]

Le 22/01/2011 12:13, Jan Kiszka a écrit :
> Hi,
>
> what are the missing bits to get nv50-class hardware into lowest power
> mode? Even when clocking down to perflvl 0, the GPU temperature of my
> Quadro FX 880M is still quickly climbing until it reaches 60 °C and the
> fan goes on for a while (I tried up to 2.6.38-rc1 so far - any progress
> in nouveau-next?). This is not only annoying but also sucking several
> watts more than what is feasible - unfortunate for mobile operation.
>
> The benchmark is still the blob which runs the card at level 0 around 54
> °C, and that means fan-less. But as the blob is causing increasing
> troubles across suspend/resume, not to speak of all its other
> limitations. So I'd rather like to switch ASAP now.
>
> What may cause this different power consumption? As the clocking is
> apparently the same (the blob even claims to use a few more MHz), I was
> wondering if the less sophisticated firmware of nouveau could be the
> reason. Or is the host driver keeping the GPU too busy even in idle state?
>
> Is there anything I could do to help analyzing or even resolving this issue?
>
> Thanks,
> Jan
Hi,

To put it simply, I have no idea :s I spotted some strange behaviour 
(that could indicate some power management) when tracing the kernel but 
my attempts to mimic it led to ..... 0.1W less (I guess it's just noise).

I must admit I'm not working enough on it yet, this is because I would 
like to get safe recloking and fan management out of the door before 
looking at real power management, but if you have some knowledge in the 
field, I would love if you could give me a hand on this :)

I'm pretty sure the difference in power consumption is due to un-needed 
engines running. I may have a look at it this week as I'll be away from 
my desktop computer.

Anyway, I'll be in Germany next week, are you living close to Bochum? We 
could talk about this if you want/can.

Take care,
Martin

Re: Low power operational mode of nv50

[Jan Kiszka]

[-- Attachment #1.1: Type: text/plain, Size: 2806 bytes --]

[please keep me on CC]

On 2011-01-22 12:49, Martin Peres wrote:
> Le 22/01/2011 12:13, Jan Kiszka a écrit :
>> Hi,
>>
>> what are the missing bits to get nv50-class hardware into lowest power
>> mode? Even when clocking down to perflvl 0, the GPU temperature of my
>> Quadro FX 880M is still quickly climbing until it reaches 60 °C and the
>> fan goes on for a while (I tried up to 2.6.38-rc1 so far - any progress
>> in nouveau-next?). This is not only annoying but also sucking several
>> watts more than what is feasible - unfortunate for mobile operation.
>>
>> The benchmark is still the blob which runs the card at level 0 around 54
>> °C, and that means fan-less. But as the blob is causing increasing
>> troubles across suspend/resume, not to speak of all its other
>> limitations. So I'd rather like to switch ASAP now.
>>
>> What may cause this different power consumption? As the clocking is
>> apparently the same (the blob even claims to use a few more MHz), I was
>> wondering if the less sophisticated firmware of nouveau could be the
>> reason. Or is the host driver keeping the GPU too busy even in idle
>> state?
>>
>> Is there anything I could do to help analyzing or even resolving this
>> issue?
>>
>> Thanks,
>> Jan
> Hi,
> 
> To put it simply, I have no idea :s I spotted some strange behaviour
> (that could indicate some power management) when tracing the kernel but
> my attempts to mimic it led to ..... 0.1W less (I guess it's just noise).

Yeah, I think we are looking something like 3-5 Watt.

> 
> I must admit I'm not working enough on it yet, this is because I would
> like to get safe recloking and fan management out of the door before
> looking at real power management,

I agree that being able to safely switch between power modes is
important. But, at least for the rather common mobile scenarios, it is
first of all important to have acceptable 2D support (I think that's
already the case) which does not suck your battery and steal your nerves
via noisy fans. And that's still preventing me to switch.

> but if you have some knowledge in the
> field, I would love if you could give me a hand on this :)

Yet I do not know much about the nouveau or the hardware it drives. But
I'm not really unfamiliar with messing around in drivers and kernels. So
if you tell me what might be worth looking at or trying out...

> 
> I'm pretty sure the difference in power consumption is due to un-needed
> engines running. I may have a look at it this week as I'll be away from
> my desktop computer.
> 
> Anyway, I'll be in Germany next week, are you living close to Bochum? We
> could talk about this if you want/can.

Not really close: Munich. Guess we have to stick with email or chat. :)

Thanks,
Jan



[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 259 bytes --]

[-- Attachment #2: Type: text/plain, Size: 181 bytes --]

_______________________________________________
Nouveau mailing list
Nouveau-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
http://lists.freedesktop.org/mailman/listinfo/nouveau

Re: Low power operational mode of nv50

[Martin Peres]

Le 22/01/2011 13:38, Jan Kiszka a écrit :
> [please keep me on CC]
ok, sorry about that.
> On 2011-01-22 12:49, Martin Peres wrote:
>>
>> To put it simply, I have no idea :s I spotted some strange behaviour
>> (that could indicate some power management) when tracing the kernel but
>> my attempts to mimic it led to ..... 0.1W less (I guess it's just noise).
> Yeah, I think we are looking something like 3-5 Watt.
Yes
>> I must admit I'm not working enough on it yet, this is because I would
>> like to get safe recloking and fan management out of the door before
>> looking at real power management,
> I agree that being able to safely switch between power modes is
> important. But, at least for the rather common mobile scenarios, it is
> first of all important to have acceptable 2D support (I think that's
> already the case) which does not suck your battery and steal your nerves
> via noisy fans. And that's still preventing me to switch.
Well, I was more pragmatic: "Let's do what we already 'know' we can do".
Also, I wouldn't know what are the normal instructions if I hadn't 
implemented
the reclocking support myself.

I understand your frustration though.
>> but if you have some knowledge in the
>> field, I would love if you could give me a hand on this :)
> Yet I do not know much about the nouveau or the hardware it drives. But
> I'm not really unfamiliar with messing around in drivers and kernels. So
> if you tell me what might be worth looking at or trying out...
Cool, well, what you can do is look at mmiotraces (I can provide you 
with some or you could make one yourself: 
http://nouveau.freedesktop.org/wiki/MmioTrace )

Send me your vbios and the trace when you're done. I'll tell you what to 
look for and what is already known :)

You'll also need to read 
https://github.com/pathscale/pscnv/wiki/Power-Management and 
https://github.com/pathscale/pscnv/wiki/Introductorycourse .

Then, you'll need the demmio tool to easily read the mmiotrace 
(https://github.com/pathscale/pscnv/wiki/RE_Tools).

$ demmio my_mmio_trace.txt | less -R

Then you need to look at the right place, I would suggest going to the 
position where the reclock is happenning (type "/0x4020 <=", less will 
search for "0x4020 <=" in the text).

I'll tell you more when I receive your mail (send it to mmio dot dumps 
at gmail dot com, my mailbox is already almost full of vbios).
>> Anyway, I'll be in Germany next week, are you living close to Bochum? We
>> could talk about this if you want/can.
> Not really close: Munich. Guess we have to stick with email or chat. :)
Indeed :)

Martin