some interesting mac pro old and new tidbits

[mattspace]

I don't know how many people here listen live to a certain tech podcast which has been receiving information from someone claiming to be at Apple, but a few things said in the live recording chat (which I didn't previously know) might be of interest.

• The nMP's graphics cards are entirely fabbed by AMD
• Nvidia's price to do the same thing was around $2000 more per machine (sounds like "go get stuffed" pricing).
• Thunderbolt 3 monitors which have a graphics card in the screen are in testing (though that's never been a guarantee of a product).
• the Nvidia web driver isn't being done as per the specs Apple wants driver makers to follow, installing in /system (replacing the factory driver?) rather than /library, and this is a pretty major sore point at Apple, described as being due to a lack of respect for the platform.

I've got no real knowledge of how drivers work for the newer Nvidia cards, but I'd be interested to see what people who are knowledgable about using / installing them think of this.

The TB3 monitors thing is interesting, and I have to wonder if eGPU support will be with TB3's release, and whether it will be restricted to Apple's display. I'd have to imagine that would either royally annoy all the third party display makers, or lead to a bonanza of hastened replacement cycles.

Of course, this (edit the "reveals" by the "Apple person") may all be an epic troll so appropriate skepticism is healthy.

 

[H2SO4]

Of course, this may all be an epic troll so appropriate skepticism is healthy.

It has the makings of a great troll post to be sure but interesting nonetheless.

 

[mattspace]

It has the makings of a great troll post to be sure but interesting nonetheless.

I'm referring more to the "revealing" of information that was happening in the chat as being a troll - but yeah, I am curious about how the web driver is doing its job. The "Apple" person seemed pretty dismissive of it.

 

[H2SO4]

I'm referring more to the "revealing" of information that was happening in the chat as being a troll - but yeah, I am curious about how the web driver is doing its job. The "Apple" person seemed pretty dismissive of it.

LOL, sorry. When I read your post that last bit kind of came across like a disclaimer.

 

[mattspace]

LOL, sorry. When I read your post that last bit kind of came across like a disclaimer.

edited for clarification

 

[lowendlinux]

Of course the drivers install in system i.e. root. You could push into user but that would require Apple and Apple's not going to do that. Divers are called kexts (kernel extensions) it's pretty hard to but video kernel extensions into user space without a pipe.

 

[mattspace]

Of course the drivers install in system i.e. root. You could push into user but that would require Apple and Apple's not going to do that.

could you expand on what that means? Do scanners or any other hardware devices require root drivers? What is it about graphics that require it?

The impression I got was that the attitude from Apple was that they'd be happy for them to do it properly - assuming this person is genuine and not just projecting their pet beliefs.

 

[lowendlinux]

could you expand on what that means? Do scanners or any other hardware devices require root drivers? What is it about graphics that require it?

The impression I got was that the attitude from Apple was that they'd be happy for them to do it properly - assuming this person is genuine and not just projecting their pet beliefs.

These would probably be good reads and they're nice an quick

https://developer.apple.com/library...r.html#//apple_ref/doc/uid/TP40001067-CH9-SW1

https://developer.apple.com/library...html#//apple_ref/doc/uid/TP40001067-CH273-SW1

 

[Fl0r!an]

Of course the drivers install in system i.e. root. You could push into user but that would require Apple and Apple's not going to do that. Divers are called kexts (kernel extensions) it's pretty hard to but video kernel extensions into user space without a pipe.

I don't think that was the point. The OP says, that, by Apples guidelines, a 3rd party kext has to be installed in /Library/Extensions instead of /System/Library/Extensions (not the folder in your user directory). From my knowledge you could totally do that with a GPU driver, the only problem is that at least one of the Nvidia kexts has to be overwritten to add the logic for switching between default driver and web driver. Apple is to blame here, Nvidia has no other choice as long as Apple neglects the latest Nvidia drivers.

 

[lowendlinux]

I don't think that was the point. The OP says, that, by Apples guidelines, a 3rd party kext has to be installed in /Library/Extensions instead of /System/Library/Extensions (not the folder in your user directory). From my knowledge you could totally do that with a GPU driver, the only problem is that at least one of the Nvidia kexts has to be overwritten to add the logic for switching between default driver and web driver. Apple is to blame here, Nvidia has no other choice as long as Apple neglects the latest Nvidia drivers.

/L/E is in user space S/L/E is root **never mind nix that both are root.

Which pushes up back to layers apple wants drivers in media and nvidia wants them in core OS. I'm going to stop typing in this thread and start reading maybe I'll learn something.

 

[mattspace]

the only problem is that at least one of the Nvidia kexts has to be overwritten to add the logic for switching between default driver and web driver.

right, so you're saying that the problem is that Apple's driver system doesn't feature the ability to pass off to a 3rd party graphics driver which overlaps a built-in one?

 

[lowendlinux]

NVM

 

[Fl0r!an]

right, so you're saying that the problem is that Apple's driver system doesn't feature the ability to pass off to a 3rd party graphics driver which overlaps a built-in one?

At least that's what I thought until a few minutes ago.
netkas just reported on his site that the latest Web Driver (El Capitan) doesn't overwrite NVDAStartup.kext any longer (not possible with rootless I guess) but still installs in S/L/E.

 

[ManuelGomes]

Where can you find the podcast?

 

[ActionableMango]

I don't know how many people here listen live to a certain tech podcast

Which one? Is it against the rules to name podcasts or something?

 

[Fangio]

Which one? Is it against the rules to name podcasts or something?

+1, I don't think it is.

Would also like to see this statement in public: "Nvidia's price to do the same thing was around $2000 more per machine"

 

[deconstruct60]

/L/E is in user space S/L/E is root **never mind nix that both are root.

Those are two directories; not user/kernel space. Where the stuff in those directories are loaded into RAM/Virtual memory is a "user" vs. "kernel" space distinction.

What Apple is enforcing is that the files in /System belong to Apple only. That doesn't 'nuke' any 3rd party drivers that are hooking into the interfaces ( APIs, kernel hooks , etc ) that they are suppose to hook into. It does stop yahoos who think installing their driver means clobbering/destroying/deleting/mutation code that is not theirs (i.e., Apple's). Proper extensions do just that "extent"; not screw with other people's code.

Which pushes up back to layers apple wants drivers in media and nvidia wants them in core OS. I'm going to stop typing in this thread and start reading maybe I'll learn something.

That's an odd characterization of the issue. Apple wants people to put their drivers in a different directly/folder than Apple puts their "extensions". The "CoreOS" essentially an Apple code release zone. As long as Apple has mechanisms to load the signed kernel extensions there is major difference in locating them in directory A or directory B . Those aren't where they actually run anyway.

Apple has been a bit sloppy and inconsistent in the past. Some stuff that they have put into /System that probably should have been segregated out. Also with root esculated privledges it was also easy for 3rd party installers to just ignore the rules and just put things where ever they wanted mutating permissions, directories, files of other folks stuff along the way. That ends with SIP.

 

[deconstruct60]

I don't know how many people here listen live to a certain tech podcast which has been receiving information from someone claiming to be at Apple, but a few things said in the live recording chat (which I didn't previously know) might be of interest.

• The nMP's graphics cards are entirely fabbed by AMD
• Nvidia's price to do the same thing was around $2000 more per machine (sounds like "go get stuffed" pricing).

Not sure what the notion of "fabbed" is when AMD doesn't particularly make circuit boards/cards themselves. I suppose this could mean that AMD weaves these under the same contract manufacturing deal they have for their upscale FirePro cards, so that Apple and AMD get joint scale pricing on those aggregated volumes.

Nvidia's price is more along the lines of "I like my Pro card mark ups just where they are" as opposed "go away". Materially I'm not sure there is a big pragmatic difference but Apple puts a target profit mark-up on their systems and devices that they don't waver from. This is Nvidia doing the same thing for that "Pro" market targeted cards. [ It is quite likely Apple was asking for the same "Pro" brand sharing for Mac Pro cards from Nvidia as well as AMD. ]

• Thunderbolt 3 monitors which have a graphics card in the screen are in testing (though that's never been a guarantee of a product).

Sounds like putting more iMac components in a Display Docking station more so than what would generally excite a Mac Pro ( or Mini ) customer base. The GPU in the same "relatively thin" monitor case is going to run into very similar constraints as the iMac. I'd be extremely surprised if this was better performance than the Mac Pro GPUs.
For a MacBook ( with CoreM GPU ) or a MBA ( non Intel Iris ) GPU this might prove a value add, but in the Mac Pro space it seems like just a bigger mismatch better the need for a external monitor versus a deep seat need for a docking station.

This product is a better docking station than a monitor. Maybe they will release something like that but sounds like an internal R&D project to support work that they may need to do anyway to more generally support eGPU if that is a required part of TB certification process in the future.

• the Nvidia web driver isn't being done as per the specs Apple wants driver makers to follow, installing in /system (replacing the factory driver?) rather than /library, and this is a pretty major sore point at Apple, described as being due to a lack of respect for the platform.

Signed 3rd party kernel extensions are suppose to go in /Library. The graphics stack "work" is split between Apple and the GPU vendors. I think some of the tension is also folks on different sides of the shared interface wanting to release updates at different speeds and scopes. The 'blame Apple' tend to pop up even when it is Nvidia that is out of sync with what they share with Apple. Could Apple incrementally improve the APIs so there was slightly less coupling? Perhaps. but this release and then cheer lead your fan boy base into blaming Apple isn't likely to bring about change.

 

[mattspace]

Not sure what the notion of "fabbed" is when AMD doesn't particularly make circuit boards/cards themselves.

The impression I had was that they're supplied to Apple by AMD as a complete unit, rather than being something Apple has manufactured through their contractors, based on technology they licence fro AMD. A question that was posed was to look at the games consoles, which are all AMD now - my guess is the implication is that Nvidia are oriented towards standard desktop graphics, and embedded / tablet.

Sounds like putting more iMac components in a Display Docking station more so than what would generally excite a Mac Pro ( or Mini ) customer base. The GPU in the same "relatively thin" monitor case is going to run into very similar constraints as the iMac. I'd be extremely surprised if this was better performance than the Mac Pro GPUs.

Basically an iMac without the computer. What I was wondering last night is if the function of the graphics card that's responsible for putting stuff on the screen has to be in the same hardware as the part that does all the 3d calculations. If not, there might be something to the idea that a screen with a graphics card that only has to drive that one screen might not become obsolete in the same way that the 3D and OpenCL oriented hardware function stuff does

This product is a better docking station than a monitor. Maybe they will release something like that but sounds like an internal R&D project to support work that they may need to do anyway to more generally support eGPU if that is a required part of TB certification process in the future.

I have a hard time thinking that they could do a monitor that you have to ditch in order to get faster graphics, AND prevent external PCI boxes with retail graphics cards from doing the same job when plugged into normal dumb monitors.

For people asking, this conversation wasn't on a podcast, I just saw it happening in a chatroom while the show was being recorded - but I'd rather not say where.