PM issues r600 driver

[jrch2k8]

hi guyz, first my kudos this driveris geeting great and fast (at least for my needs) but im having a weird pm issue wich i like to report, so i need your help to enlight me about how to debug it.

my specs
phenom II X4 955 BE @ 4ghz
AMD Radeon HD 4850x2
1 32" HDTV connected through DVI-1
1 22" Monitor connected through DVI-0

OS Kubuntu 10.04.1 + ppa motu:xorgedgers
kernel 2.6.36-rc4 from ubuntu kernel team
PM settings dynpm/profile low(aka tested with both)

ok here is the issues i discover in my pc running r600 driver(i know pm is almost KMS exclusive, r600 is just to make it clearer to read) that when i have only 1 monitor output active the driver with almost any flicker anymore keep a decent temperature close to fglrx(this is normal desktop composite + chromium hardware accelerated) but when i activate the second monitor output in clone or extended mode the gpu goes to maximun settings(aka high and 75?) whatever the option i pass to the pm system

when i disable the second monitor output the pm system goes back to normal.

Notes:

1.) 4850x2 card have 4 DVI outputs aka 2 per GPU, im using the DVI0 and DVI1 which are dependant on GPU0

2.) is amazing but KMS recognize my second GPU(GPU1) and apply PM and keep it COOL , btw when i trigger the bug only GPU0 is affected GPU1 keeps low temps no matter what happens with GPU0

3 i heard that pm code in KMS shut down many PCI-e lanes depending the configuration, now since X2 cards are a crossfire system but pasted in a single PCB, is possible that when i set GPU1 to profile:low it kills many pci-e lanes no matter that GPU0 is on dynpm or profile:high, since the pci-e port is the same?
----------is kms aware of this kind of setup?
----------i can help to debug or provide test info, if you send me some instruction of what to look at

4 since KMS recongnize and provide acceleration for both of my GPU's(im impressed btw), it means some work has been done in the multiGPU arena? (mmm in fact xorg logs show me that my second GPU is getting DRI accel too and it works but not in parallel yet)

[droidhacker]

You need to be aware that "dynpm" doesn't work on multi-head configurations.

Since you have a multi-gpu card, is it possible to put the second monitor on to the second GPU? You might be able to get dynpm working that way by configuring them both to dynpm.....

I.e., the reason why dynpm only works with single-head mode is because it tries to reclock the GPU during blanking, so if you have multiple monitors, the blanking doesn't match up and it doesn't get to reclock.

If you're running two GPUs, each with one monitor, you should be able to reclock the GPUs since each GPU is only responsible for one monitor, and it can therefore do so during blanking.

[jrch2k8]

You need to be aware that "dynpm" doesn't work on multi-head configurations.

Since you have a multi-gpu card, is it possible to put the second monitor on to the second GPU? You might be able to get dynpm working that way by configuring them both to dynpm.....

I.e., the reason why dynpm only works with single-head mode is because it tries to reclock the GPU during blanking, so if you have multiple monitors, the blanking doesn't match up and it doesn't get to reclock.

If you're running two GPUs, each with one monitor, you should be able to reclock the GPUs since each GPU is only responsible for one monitor, and it can therefore do so during blanking.

i tried that but the the GPU1 don't accept to move a window from the GPU0 or keep 3d working between monitors, maybe i made something wrong? you any tip that can help me with this setup?

[pingufunkybeat]

I have a question.

Does dynpm work on Windows/MacOS using the official drivers in multi-screen configuration?

If so, how do they deal with the reclocking/vsync issues?

It can't be a problem unique to Linux.

[bridgman]

I don't think the open drivers run on Windows/MacOS, do they ? The proprietary drivers have much more complex power management code to deal with those scenarios.

The technical challenges are not unique to Linux, but the attempt to solve all of those challenges with somewhere between 1/5th and 1/50th the resources is unique to Linux. That is why proprietary drivers appeared in the first place and why they still exist today... they allow core code to be shared across 100% of the PC market rather than just the market share of a single OS family.

[pingufunkybeat]

I don't think the open drivers run on Windows/MacOS, do they ? The proprietary drivers have much more complex power management code to deal with those scenarios.

This is basically what I was asking.

So, in other words, it is really difficult to do dynpm with two (or more) separate screens, and a lot of work went into proprietary drivers to make it work.

So, it is rather unlikely that we'll see dynpm work in this configuration any time soon?

[bridgman]

Depends what you mean by "any time soon". Ask again in a couple of months.

If you mean "will it happen inside a month", probably not...

[droidhacker]

i tried that but the the GPU1 don't accept to move a window from the GPU0 or keep 3d working between monitors, maybe i made something wrong? you any tip that can help me with this setup?

You might want to look into DMX/Xdmx.. but yes, that is a downside to multiple GPUs.

[pingufunkybeat]

Depends what you mean by "any time soon". Ask again in a couple of months.

If you mean "will it happen inside a month", probably not...

With "any time soon", I meant a couple of years, and I'm pleasantly surprised that you're mentioning months!

[droidhacker]

With "any time soon", I meant a couple of years, and I'm pleasantly surprised that you're mentioning months!

Way to hijack a simple thread...

And FYI: he said to ASK AGAIN in a couple of months... not that things would be changing within a couple of months. In a couple of months you ask him and the answer might be that they've contracted an 18 year old to work on improving the power management once he has earned his PhD in computer science (about 10-12 years).