Linux GTX 675M support

[geophph]

Originally Posted by Gusar

Your fault for not informing yourself before shelling out the big bucks. Optimus isn't new and it's Linux situation is a known thing, so it's not like you can claim you couldn't have known.

If you spend big bucks based on assumptions and get burned, you only have yourself to blame.

The issue with Optimus under Linux has traditionally been Optimus does not work. You can only use your Intel GPU or the NVidia GPU with no power saving benefits. I'm ok with this and it has been known since Optimus first came out.

With the Ivy Bridge + 675M combo this has changed. Now Linux can no longer use the NVidia GPU at all. There is no clear information on this. In fact I was made aware when ordering a new laptop. The guys over at Power Notebooks informed me when I told them I would be putting Linux on it. After that I spent the better part of that Saturday searching and finding only one forum post (I believe here) with somebody not able to get their NVidia drivers to detect any screens.

If it wasn't for Power Notebooks telling me this problem I would have been unaware. I was lucky enough to find out before ordering a new notebook and purchased one with a supported hardware configuration. It is a lesson learned and I did not get burned on this one. NVidia will no longer be providing the Linux support they used to and will adjust making purchasing habits appropriately, more research. It's still not cool of NVidia.

[tmack]

It's telling that many vendors do not advertise the existence of Nvidia Optimus at all, for example:

http://www.msi.com/product/nb/GT60-0NC.html

How many times to they specify that the laptop has an additional video card or that Nvidia Optimus technology is present? This isn't uncommon, most OEMs do not specify the existence of Optimus in their Ivy Bridge products.

One can only wonder why that is the case.

[Gusar]

Originally Posted by geophph

The issue with Optimus under Linux has traditionally been Optimus does not work. You can only use your Intel GPU or the NVidia GPU with no power saving benefits. I'm ok with this and it has been known since Optimus first came out.

Incorrect. What you're talking about is switching graphic cards with a hardware mux. But Optimus is *not* about switching, never was. Pure Optimus laptops do not have a hardware mux, and this isn't new with the Ivy Bridge + 675M combo as you claim it is. Some laptops do have a mux and a BIOS option for "integrated/dedicated/Optimus". But that's a feature they provide in addition to Optimus. The point of Optimus is to operate without a mux.

Originally Posted by geophph

NVidia will no longer be providing the Linux support they used to

It's not on Nvidia alone to make Optimus happen. They need DMA-BUF. Texas Instruments, one of the major players in creating DMA-BUF, agrees that Nvidia should be able to have it. The rules of the Linux kernel, as quoted by the the Nvidia dev, agree that they should be able to have it. It's all here: https://lkml.org/lkml/2012/2/19/88

Originally Posted by tmack

One can only wonder why that is the case.

Because Optimus has become ubiquitous and as such isn't something that's worth pointing out. Don't go looking for conspiracy theories where there aren't any.

[geophph]

Originally Posted by Gusar

Incorrect. What you're talking about is switching graphic cards with a hardware mux. But Optimus is *not* about switching, never was. Pure Optimus laptops do not have a hardware mux, and this isn't new with the Ivy Bridge + 675M combo as you claim it is. Some laptops do have a mux and a BIOS option for "integrated/dedicated/Optimus". But that's a feature they provide in addition to Optimus. The point of Optimus is to operate without a mux.

Originally Posted by NVidia

Like a Hybrid car seamlessly transitions from gas to electric, NVIDIA Optimus technology automatically optimizes your computing experience so you can keep working without interruptions.

I may be wrong, but I'm going by what NVidia describes in layman's terms. I'm a software guy, so I never bothered to understand what is happening in depth with Optimus mainly because I didn't care. I want a desktop replacement laptop, a couple of hours of battery life is all I want.

I've never heard of Nvidia GPUs not working at all under Linux until now. I'm currently using a 420M with no problems. Just because I haven't heard of it doesn't mean it's not true, but I believe it is far less common. With the introduction of the Ivy Bridge + 675m combination it may end up being far more common.

Originally Posted by Gusar

It's not on Nvidia alone to make Optimus happen. They need DMA-BUF. Texas Instruments, one of the major players in creating DMA-BUF, agrees that Nvidia should be able to have it. The rules of the Linux kernel, as quoted by the the Nvidia dev, agree that they should be able to have it. It's all here: https://lkml.org/lkml/2012/2/19/88

I do agree with the fact that is a combination of Nvidia and the Kernel devs. As I stated previously it looks to be the common disagreements with the Kernel devs and proprietary players. In the end it's hurting us. I will appropriately switch some of my blame on the kernel devs as well if it makes you feel better.

[Gusar]

Originally Posted by geophph

I'm a software guy, so I never bothered to understand what is happening in depth with Optimus mainly because I didn't care.

But in order to discuss this topic properly, you do need to care and know how Optimus works.

Originally Posted by geophph

I've never heard of Nvidia GPUs not working at all under Linux until now.

Who said it doesn't work? Well ok, it may not with the current 295.53 and 302.11 drivers. But if you look at the release notes for 295.53, it added support for a whole bunch of 6xx cards, so I'm sure one of the next releases will add 675M support.
But that's support just for the GPU itself. Then still remains the usual problem that the GPU isn't connected to the internal display. So you need either proper support with DMA-BUF, or Bumblebee.

Originally Posted by geophph

I will appropriately switch some of my blame on the kernel devs as well if it makes you feel better.

Heh. But this isn't about me, I don't have an Optimus laptop. It's about properly understanding the issue at hand.

[geophph]

Originally Posted by Gusar

But in order to discuss this topic properly, you do need to care and know how Optimus works.

When I created this topic it was a last resort to figure out if the piece of hardware I want to buy will work or not.

Originally Posted by Gusar

Who said it doesn't work? Well ok, it may not with the current 295.53 and 302.11 drivers. But if you look at the release notes for 295.53, it added support for a whole bunch of 6xx cards, so I'm sure one of the next releases will add 675M support.
But that's support just for the GPU itself. Then still remains the usual problem that the GPU isn't connected to the internal display. So you need either proper support with DMA-BUF, or Bumblebee.

Heh. But this isn't about me, I don't have an Optimus laptop. It's about properly understanding the issue at hand.

Right so again this topic could be useful for others looking to buy the latest NVidia mobile hardware and informing themselves on a proper purchasing decision.

EDIT: I should add that it is critical to be aware of proper support for DMA-BUF as well. Because obviously the first place I checked was the Nvidia driver release notes. Which could be misleading (when they do support later 6xx series) if many new laptops will not have the GPU connected to the internal display.

[DanielVS]

Hi all,

I am looking for a new laptop with updated cpu and nvidia gpu for CUDA and image processing - some designs based on Clevo's became very interesting (nice specs, price, etc..).

Which brought me here, looking for the situation of optimus support under linux for the sake of using the nvidia gpu - I really _don't_ care about the integrated gpu and don't plan to ever use it. Which turns out to be a situation, imho, very similar to geophph's.

Considering that nor manufacturers, nor nvidia provide details about what's really going on under the hood of any laptop, the _only_ reliable way to get the precious info (if the gpu will or not work under linux) is with people's reports.

So, thank you very much geophph for opening this thread, I will not buy a notebook with IVB+Nvidia combo as I don't intend to pay as much for an useless brick.

For Gusar: understanding properly the issue is indeed very important. If I didn't get wrong, the very issue of this thread is to provide information to others about the current state of _nvidia_ support in linux, and for now, it seems to be null for IVB+nvidia (anything with intel and nvidia w/o optimus these days??).

So you totally missed the point by pushing the DMA-BUF issue here. To attract new developers to solve the problem, I believe the lmkl is a better place to start.. But thanks anyway for your resume on optimus mojo.

If Nvidia can't deal with linux kernel people (because nvidia doesn't want to gpl their code) and if kernel people also can't (because they can't let go non gpl stuff there), it is sure not the topic. To that matter, you should consider that it is because of gpl that linux is where it is in the first place, so, personally, I support the kernel plp's decisions and believe that blames should go all to nvidia.

[Gusar]

Originally Posted by DanielVS

the very issue of this thread is to provide information to others about the current state of _nvidia_ support in linux, and for now, it seems to be null for IVB+nvidia (anything with intel and nvidia w/o optimus these days??).

No, the state is not null. It's the same as it always was - to get Optimus running, you need to use Bumblebee, which is a hack created by the community. Why are people behaving as if there's suddenly a new situation now?

Originally Posted by DanielVS

I really _don't_ care about the integrated gpu and don't plan to ever use it.

Then a laptop isn't for you. Laptops with the igpu deactivated are pretty much if not completely extinct. All of them are Optimus nowadays. And the point of Optimus is to run as much as possible on the igpu and only activate the Nvidia card for specific heavy-duty tasks (games, 3d modeling, and such).

[DanielVS]

Originally Posted by Gusar

No, the state is not null. It's the same as it always was - to get Optimus running, you need to use Bumblebee, which is a hack created by the community. Why are people behaving as if there's suddenly a new situation now?

Ok Gusar, I've found this post (_http://www.pavanky.com/archlinux-cuda-on-ivy-bridge-kepler/) which reports a working IVB+nvidia under linux with bumblebee. But is it a *guarantee* that any other IVB+nvidia will work? There are certain scenarios where plp want more hardware than that Asus nb.

Originally Posted by Gusar

Then a laptop isn't for you. Laptops with the igpu deactivated are pretty much if not completely extinct. All of them are Optimus nowadays. And the point of Optimus is to run as much as possible on the igpu and only activate the Nvidia card for specific heavy-duty tasks (games, 3d modeling, and such).

Talking about scenarios, mine is just like that, I don't care about the igpu at all, even more if it doesn't do any good for me besides boring me with hacking other stuff than work. The fact that nvidia doesn't _mandate_ a bios/efi switch is enough excuse to blame them. There are *some* nb models with said switching scheme, but they're not everywhere.

Well, back to the topic, yes there is IVB+nvidia working, but maybe not what serves you best...

[DanielVS]

Originally Posted by Gusar

Ah, the "let me make insults and engage in ad-hominems because I have no real arguments after being exposed that I have no clue what I'm talking about" game. Even more fun.

I'm used to see this in philosophy debates, which is precisely the "question" behind the usage of DMA-BUF by proprietary drivers: to gpl or not to gpl

[Gusar]

Originally Posted by DanielVS

Ok Gusar, I've found this post (_http://www.pavanky.com/archlinux-cuda-on-ivy-bridge-kepler/) which reports a working IVB+nvidia under linux with bumblebee. But is it a *guarantee* that any other IVB+nvidia will work?

There is nothing special about IVB, that was the point of my previous post. So *all* IVB+nvidia will work with Bumblebee. That the 675M currently doesn't is a separate issue from IVB or Optimus, it doesn't work because it's too new, but like I said, for sure the next driver will add support.

Though, can anyone actually confirm it doesn't work? I assumed it doesn't because it's not on the list of either the stable or beta drivers. But the 650M at that link you gave isn't on those lists either. So maybe the 675M already works, the lists are just incomplete currently.

Originally Posted by DanielVS

The fact that nvidia doesn't _mandate_ a bios/efi switch is enough excuse to blame them.

Well, this is how I see it: Why would they mandate it, when the point is to operate without it. And it simplifies the hardware design when you can just statically wire up the chips and that's it.

[geophph]

Originally Posted by Gusar

There is nothing special about IVB, that was the point of my previous post. So *all* IVB+nvidia will work with Bumblebee. That the 675M currently doesn't is a separate issue from IVB or Optimus, it doesn't work because it's too new, but like I said, for sure the next driver will add support.

Though, can anyone actually confirm it doesn't work? I assumed it doesn't because it's not on the list of either the stable or beta drivers. But the 650M at that link you gave isn't on those lists either. So maybe the 675M already works, the lists are just incomplete currently.

I thought there is something different going on with IVB+nvidia. Perhaps not, maybe all it comes down to is the current NVidia drivers do not support the later 6xxMs yet. I was told by Power Notebooks that there is something different with the architecture and basically you need to have Optimus working in order utilize the NVidia GPU. I would be happy if things would function like my GTX 420M laptop.

Section "Device"
   Identifier     "Device0"
   Driver         "nvidia"
EndSection

Just use the NVidia GPU all the time and ignore Optimus.

If it's just a matter of waiting for NVidia to update their drivers then that's one thing. The concern is if this requires proper kernel support as well.