Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Webdriver for GTX 1080&1070

mavericks7913 said:
Did you install the driver first?
Click to expand...

Mini hijack, on that topic:

With the new driver installed before shutdown and card replacement.... there's no other trickery involved here, right?

I realise I don't get boot screen and so on, but without further hacks, the Mac Pro will boot to a functioning desktop now? Referring to let's say a 1080 Ti and latest Sierra on a Mac Pro 4.1/5.1.
 
  • Like
Reactions: MaSaLa
AndreeOnline said:
Mini hijack, on that topic:

With the new driver installed before shutdown and card replacement.... there's no other trickery involved here, right?

I realise I don't get boot screen and so on, but without further hacks, the Mac Pro will boot to a functioning desktop now? Referring to let's say a 1080 Ti and latest Sierra on a Mac Pro 4.1/5.1.
Click to expand...

Sorry i'm a little english language and Thank you for answer.
My mac pro is not original vga card because it crash.
First time I'm install osx sierra and webdriver nvidia with remote target disk from my MBP with firewire cable.
and finished to install all i shut down macpro install GTX 1060 power on wait a miniute macpro is shut down auto.

Mac Pro 3.1
2.8 GHz 8 Core
Ram 4 Gb
OSX Sierra

Thank you for answer. ^^
 
So whats the (new) deal?

- Do we get a boot screen with non EFI GPUs?
- Are we able to install OS X updates without the need of swapping cards?
- Does Cuda work?
- Do we have any performance and compatibility "test" with pro Apps (After Effects, Davinci Resolve etc.)

I would really appreciate if somebody could clarify these questions for all of us that don't have 10XX GPUs to test. Just to know where we're at.

Thanks in advance!!!
 
I'm going to do what I normally don't like so much myself, and give you a half-baked answer:

MaSaLa said:
So whats the (new) deal?

1. Do we get a boot screen with non EFI GPUs?
2. Are we able to install OS X updates without the need of swapping cards?
3. Does Cuda work?
4. Do we have any performance and compatibility "test" with pro Apps (After Effects, Davinci Resolve etc.)
Click to expand...

1: No
2: I'd say: make sure to not 'auto update' MacOS. You have have auto updates on for other apps. Then, after a new MacOS release, perform your install via Terminal without restart. Then install Nvidia WebDriver. Then reboot. Let's see if this pans out to be good advice. I have not used other Nvidia cards in my Mac Pro yet, but this is my best guess.
3. Initially no. Then I saw a link to a CUDA driver yesterday and one user said it worked—so I guess YES. But it's a separate CUDA install. Not included in driver at this point. Was it ever?
4. Barefeats.com is on it. Results shortly. There is also a YouTuber called Phil's Garage, or similar to that. I think he is also testing at this point.
 
Thanks Andree for clarifying!

So it's basically kind of the same deal like before with non EFI GPUs.

Greeting from Basel ;-)
 
  • Like
Reactions: AndreeOnline
ohmkumi said:
i'm install gtx 1060 on macpro 3.1 can't boot i don't know why? power on >> sound >> just minute is all off auto. thank you.
Click to expand...
From what I've read, the Pascal driver works only Sierra 10.12.4. Doesn't work on later versions of Sierra or OS X.
 
AndreeOnline said:
3. Initially no. Then I saw a link to a CUDA driver yesterday and one user said it worked—so I guess YES. But it's a separate CUDA install. Not included in driver at this point. Was it ever?
Click to expand...
I think there has always been a separate CUDA driver. The web driver (i.e. video driver) has been necessary since Nvidia's 900 series when Apple stopped using Nvidia in macs.
 
Please stop benchmarking Pascal cards at 1600x900 resolution. All you are doing is confusing people that the drivers need "more optimization work" or are running in Kepler compatibility mode or other crap. Modern GPUs, even mid-range ones like a 1060, will be completely CPU limited at that resolution. You should run at your native monitor resolution in fullscreen mode if at all possible.
 
  • Like
Reactions: Synchro3, LightBulbFun, TheStork and 2 others
Asgorath said:
Please stop benchmarking Pascal cards at 1600x900 resolution. All you are doing is confusing people that the drivers need "more optimization work" or are running in Kepler compatibility mode or other crap. Modern GPUs, even mid-range ones like a 1060, will be completely CPU limited at that resolution. You should run at your native monitor resolution in fullscreen mode if at all possible.
Click to expand...

Given that the frame rates average in the 80 range for the high end GPUs, its probably not CPU limited or at least not very much. The goal for PC/mac gaming is generally 60 FPS, but modern gaming monitors can hit 140+ Hz so we are within a reasonable FPS target.

The mac does need better benchmarking tools, and ones that utilize Metal such as games like the new world of warcraft and The Witness. Maybe we should all try running something like GFX Bench since it utilizes metal.
 
Stacc said:
Given that the frame rates average in the 80 range for the high end GPUs, its probably not CPU limited or at least not very much. The goal for PC/mac gaming is generally 60 FPS, but modern gaming monitors can hit 140+ Hz so we are within a reasonable FPS target.

The mac does need better benchmarking tools, and ones that utilize Metal such as games like the new world of warcraft and The Witness. Maybe we should all try running something like GFX Bench since it utilizes metal.
Click to expand...

The actual framerate has nothing to do with how CPU limited a given application is. Some (well-written) apps might be CPU limited at 300 FPS. Other poorly-written apps might be CPU limited at 50 FPS. When you don't see any improvement going from a GTX 680 to a GTX 980 to a GTX 1080, all signs point to the application being CPU limited at the given resolution/settings level, no matter what score you get. This does not mean the NVIDIA drivers are terrible or poorly optimized -- the simple fact of the matter is the GPU is so fast that the CPU cannot keep up with it and give it enough work. If you were running at 4K resolution with 8xAA and everything maxed out, then you'd be much more likely to be GPU limited and see the difference between a 680 and a 1080.

TL;DR - if you want to benchmark high-end modern GPUs, make sure you're using an appropriate resolution. You'll notice how basically every tech review website uses 4K to measure GPUs these days, and only uses lower resolutions like 1920x1080 for the low-end (or maybe even mid-range) GPUs. There's simply no point in running a GTX 1080 at 1920x1080 (or lower, in the case of Unigine's defaults).

Edit: But yes, GeekBench apparently has a Metal benchmark now, so that would be a better app to use. GFXBench is a mobile benchmark and doesn't use modern graphics techniques, so while it does have a Metal version, it's using rendering that can work on phones (which isn't really a great test for a high-end GPU like a GTX 1080).
 
Stacc said:
Given that the frame rates average in the 80 range for the high end GPUs, its probably not CPU limited or at least not very much. The goal for PC/mac gaming is generally 60 FPS, but modern gaming monitors can hit 140+ Hz so we are within a reasonable FPS target.

The mac does need better benchmarking tools, and ones that utilize Metal such as games like the new world of warcraft and The Witness. Maybe we should all try running something like GFX Bench since it utilizes metal.
Click to expand...

For gaming in 1080 the cpu will be the bottleneck. That is the problem with Ryzen and 1080p. Ryzen single core performance is not as good as intel so gaming at 1080p is bottlenectk for Ryzen. It you truly want to measure gpu then do higher resolution to max out gpu. I'm not a gamer so all this doesn't really matter to me but to others...
 
Asgorath said:
The actual framerate has nothing to do with how CPU limited a given application is. Some (well-written) apps might be CPU limited at 300 FPS. Other poorly-written apps might be CPU limited at 50 FPS. When you don't see any improvement going from a GTX 680 to a GTX 980 to a GTX 1080, all signs point to the application being CPU limited at the given resolution/settings level, no matter what score you get. This does not mean the NVIDIA drivers are terrible or poorly optimized -- the simple fact of the matter is the GPU is so fast that the CPU cannot keep up with it and give it enough work. If you were running at 4K resolution with 8xAA and everything maxed out, then you'd be much more likely to be GPU limited and see the difference between a 680 and a 1080.

TL;DR - if you want to benchmark high-end modern GPUs, make sure you're using an appropriate resolution. You'll notice how basically every tech review website uses 4K to measure GPUs these days, and only uses lower resolutions like 1920x1080 for the low-end (or maybe even mid-range) GPUs. There's simply no point in running a GTX 1080 at 1920x1080 (or lower, in the case of Unigine's defaults).

Edit: But yes, GeekBench apparently has a Metal benchmark now, so that would be a better app to use. GFXBench is a mobile benchmark and doesn't use modern graphics techniques, so while it does have a Metal version, it's using rendering that can work on phones (which isn't really a great test for a high-end GPU like a GTX 1080).
Click to expand...

A low resolution alone doesn't mean you are CPU bottlenecked. You could be running an incredibly complex scene at low resolutions and still be limited by the GPU. For instance that Heaven benchmark may get 10 FPS@4k, 30@1080p and 80@1600x900, and be GPU limited in all 3 cases.

We don't really have enough data to determine if this benchmark is being CPU limited. Barefeats has already shown that there are still GPU compute benefits that can be gained with pascal cards in the classic mac pro. Maybe they can do some gaming/graphics oriented benchmarks to settle this issue.
 
Stacc said:
A low resolution alone doesn't mean you are CPU bottlenecked. You could be running an incredibly complex scene at low resolutions and still be limited by the GPU. For instance that Heaven benchmark may get 10 FPS@4k, 30@1080p and 80@1600x900, and be GPU limited in all 3 cases.

We don't really have enough data to determine if this benchmark is being CPU limited. Barefeats has already shown that there are still GPU compute benefits that can be gained with pascal cards in the classic mac pro. Maybe they can do some gaming/graphics oriented benchmarks to settle this issue.
Click to expand...

Except I've already run OpenGL Driver Monitor with this application on my Maxwell GPU in the past and confirmed it's solidly CPU limited. Again, the key point here is that a 980 and a 1080 perform the same (and I think even a 680 performs the same, at least in a cMP). Most people jump to one of the following conclusions:
  • The NVIDIA drivers are terrible.
  • The NVIDIA drivers are running in Kepler compatibility mode and haven't enabled color compression or other hardware features.
The simplest and most often correct conclusion is that you are CPU limited, and it's relatively easy to test this. I think even iStat Menus tracks GPU performance now.
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back