nVidia GTX 1080

[SoyCapitanSoyCapitan]

The ROM I used is the EVGA reference ROM posted on this forum.

The model number on the reference card I have is one of the compatible numbers listed in the 680 ROM thread on this forum.

The artefact and OpenCL bugs are on 7 and 9 series cards too, EFI ROM or not. I posted links to six forums where people using 6, 7 and 9 series cards all suffered from the same glitches.

Nvidia blogged that the current drivers only have basic beta support for Maxwell. Their driver support documentation doesn't mention Maxwell and states in black and white that they only support desktop cards up until GTX 680 and Quadro K5000 series, and beta support for Kepler GT 650/750/755.

Yet some people around here, who can't even produce evidence they are using Maxwell cards and pro apps (otherwise they would notice the bugs reported on so many forums), state with confidence and sometimes hostility that the drivers fully support Maxwell.

So beta support for mobile Kepler but full support for Maxwell? LMAO

When I tried to explain Nvidia's UDA last year those same people went ape shizzle on me for explaining how Nvidia's drivers have some forward compatibility built in.

Barefeats showed that the latest Nvidia driver optimisations put the 680 on par with the 980Ti in OpenGL performance. So what should we do? Hey let's keep posting links to the pre-optimisation benchmarks. LMAO. Why would you do that?

And we don't need to go into the hilarious people who suggested PCIE SSDs increase boot times, multitasking and app launching over a SATA 2/3 SSD, even after some of us posted hard results that can be reproduced by anyone.

Or the people who eagerly suggest buying an expensive 770 when a 680 will suffice.

But buy more stuff people, spend all your money! Trust me when I tell you that your cMP will rocket. Hmmm.

Or just save your money. The economy isn't exactly good enough to blow away your savings on stuff that might not make you any more productive than what you have already.

That's all I have to say until I give you all the full 10 series run down

 

[pastrychef]

New Nvidia web driver released.

 

[Asgorath]

The ROM I used is the EVGA reference ROM posted on this forum.

The model number on the reference card I have is one of the compatible numbers listed in the 680 ROM thread on this forum.

The artefact and OpenCL bugs are on 7 and 9 series cards too, EFI ROM or not. I posted links to six forums where people using 6, 7 and 9 series cards all suffered from the same glitches.

Nvidia blogged that the current drivers only have basic beta support for Maxwell. Their driver support documentation doesn't mention Maxwell and states in black and white that they only support desktop cards up until GTX 680 and Quadro K5000 series, and beta support for Kepler GT 650/750/755.

Yet some people around here, who can't even produce evidence they are using Maxwell cards and pro apps (otherwise they would notice the bugs reported on so many forums), state with confidence and sometimes hostility that the drivers fully support Maxwell.

So beta support for mobile Kepler but full support for Maxwell? LMAO

When I tried to explain Nvidia's UDA last year those same people went ape shizzle on me for explaining how Nvidia's drivers have some forward compatibility built in.

Barefeats showed that the latest Nvidia driver optimisations put the 680 on par with the 980Ti in OpenGL performance. So what should we do? Hey let's keep posting links to the pre-optimisation benchmarks. LMAO. Why would you do that?

And we don't need to go into the hilarious people who suggested PCIE SSDs increase boot times, multitasking and app launching over a SATA 2/3 SSD, even after some of us posted hard results that can be reproduced by anyone.

Or the people who eagerly suggest buying an expensive 770 when a 680 will suffice.

But buy more stuff people, spend all your money! Trust me when I tell you that your cMP will rocket. Hmmm.

Or just save your money. The economy isn't exactly good enough to blow away your savings on stuff that might not make you any more productive than what you have already.

That's all I have to say until I give you all the full 10 series run down

The blog post (https://blogs.nvidia.com/blog/2015/08/31/mac-driver/) says:

And, for the first time, our driver includes beta support for MacBook Pros and iMacs with Kepler GPUs, as well as beta support for those using Maxwell GPUs in older Mac Pro systems.

There is no use of the word "basic" in there at all. There is no mention of "Maxwell GPUs will run in backwards compatibility mode" at all. The fact that the products aren't listed under the "officially supported" GPUs means exactly that -- the cards are not officially supported, because there are no official Mac Edition versions of these cards. Why is that so hard for you to understand? NVIDIA has been unofficially enabling PC cards, including Maxwell cards, for years. This blog post was the first time NVIDIA has publicly mentioned that fact.

There are reports of the Photoshop corruption bugs being fixed in the latest release. I guess you haven't been bothered to update and try the latest version, so you'll just continue to spread incorrect information.

BareFeats ran a wide variety of tests. Some tests are GPU limited, and showed Maxwell already beating Kepler with the old drivers. Some tests were CPU limited, and showed large improvements with the newer web drivers. I don't understand how you can take these facts and jump to the conclusion that Maxwell is running in compatibility mode. Maybe BareFeats didn't include the old results in the new run, because the results hadn't changed with the new web driver? Again, all signs point to NVIDIA spending a large amount of effort to tune the CPU performance of their drivers, which they have been widely criticized about in the past. BareFeats selected tests that highlighted these improvements.

So, when the 1080 gets released and NVIDIA releases a driver that enables the card (which still won't show up as an officially supported product on OS X, because there won't be a Mac Edition) and GPU-limited tests show large improvements while CPU-limited cases do not, are you going to stick with your position that Maxwell and Pascal are running in backwards compatibility mode with no color compression?

 

[SoyCapitanSoyCapitan]

There are reports of the Photoshop corruption bugs being fixed in the latest release. I guess you haven't been bothered to update and try the latest version, so you'll just continue to spread incorrect information.

We also have a benchmark from ActionableMango who states that the 780M is beating the 980 in FCPX.

I'll ignore the rest of your word soup, because that's all you posted. It's a tired old forum debate tactic to keep readers confused when a subject is settled. You ignored that simple fact that mobile Kepler support isn't even complete let alone any newer architectures. Enough said.


But I'm quoting the above for reference. You have posted this accusation that I'm posting incorrect information yet you are acknowledging that the bugs existed. Where were you for several months to warn people about these bugs when you were recommending people to buy these GPUs and install these web drivers? That's highly irresponsible of you if you did know the bugs existed. And if you didn't know the bugs existed then it's highly arrogant of you to act like the forum expert on Nvidia drivers without knowing about these issues.

You have no pro Adobe apps to see these things? How come so many people here recommending these GPU upgrades don't have pro apps and can't warn about bugs experienced by pro users?

And you also have not produced any citation that the bugs have been fixed. I'm going to test this very thoroughly and if the bugs are still present then your quote above will be remembered by all. I do hope the bugs are fixed though.

 

[pastrychef]

It's like talking to a wall.

 

[Asgorath]

We also have a benchmark from ActionableMango who states that the 780M is beating the 980 in FCPX.

Literally one guy posted very questionable numbers for a heavily-modified 2011 iMac that as far as I've seen nobody else can reproduce. Even ActionableMango was questioning those results. His score of 27 seconds is pretty good for a single-GPU system, certainly comparable with AMD. I posted my TITAN X numbers for the BruceX test and my number (14.62 seconds) beats every system that I've seen results for, single or dual GPU.

But no, the only conclusion we can draw is that Maxwell is running in backwards compatibility mode with color compression disabled. Right?

You ignored that simple fact that mobile Kepler support isn't even complete let alone any newer architectures. Enough said.

You seem to be confused about what beta means. In the past, NVIDIA has released officially-supported drivers for their officially-supported add-in cards for the Mac Pro, such as the GeForce GTX 680 Mac Edition from EVGA. The web driver was explicitly not allowed to be installed on any system that didn't have a PCIe slot, such as the MacBook Pro and iMac.

Then, last year, NVIDIA removed those restrictions and let folks with laptops and iMacs install their new web driver that featured a large increase in performance over the stock Apple drivers. People had figured out ways to work around this installer restriction, but NVIDIA decided to just unlock the drivers and let them be installed anywhere.

The driver for the GTX 680 Mac Edition is exactly the same one that will run on a Kepler iMac, since the GK104 GPU is essentially the same in both cases. Beta in this context means "not officially supported". Beta in this context does not mean "running in backwards compatibility mode" or "driver is not finished". It just sets an expectation for the level of support you should expect. For example, if you email NVIDIA customer service and tell them that the web driver is causing problems on your MacBook Pro, their answer is likely going to be "just switch back to the stock Apple drivers".

 

[DearthnVader]

The artefact and OpenCL bugs are on 7 and 9 series cards too, EFI ROM or not. I posted links to six forums where people using 6, 7 and 9 series cards all suffered from the same glitches.

Only links I saw were 3 from 970 users, no hardware specs, and some hackintoch people complaining about something.

What I don't see is is Retail Mac Edition users, or even MVC customers complaining about this so called bug.

Show me a link to a 680 Retail Mac Edition using the Web Driver with this bug?

 

[Asgorath]

Only links I saw were 3 from 970 users, no hardware specs, and some hackintoch people complaining about something.

What I don't see is is Retail Mac Edition users, or even MVC customers complaining about this so called bug.

Show me a link to a 680 Retail Mac Edition using the Web Driver with this bug?

And this, ladies and gentlemen, is the exact difference between officially supported and not officially supported (or beta support).

 

[Draeconis]

Can we stop the personal attacks please, it's getting quite dull and it's frustrating having to dig through it all to actually see any information on topic.

@SoyCaptainSoyCaptain If you'd like to discuss your 680 woes at length please do so in another thread, or create your own.

Can we try and keep this thread focused on discussion around the 1080 please?

 

[ActionableMango]

We also have a benchmark from ActionableMango who states that the 780M is beating the 980 in FCPX.

I don't have enough knowledge to wade into this debate on either side.

But on this one specific benchmark I will point out that the 780M that beat my 980 also beat a 770 by a whopping 30%. This cannot be explained away with lack of Maxwell driver optimization since both the 780M and 770 are Kepler cards, and the 770 should be the faster of the two Kepler cards.

 

[MacVidCards]

You ignored that simple fact that mobile Kepler support isn't even complete let alone any newer architectures.

Hmm, didn't want to give away my next EFI quest, but after that gibbertastic remark I had too.

Here we see a Kepler and a Maxwell MOBILE chip working in X-Serve 3,1.

In a 2 lane slot, so nothing stunning in the speed department, but should be full speed once in an iMac.

I have also tested a 630 (actually Fermi) and tonight I will toss in the 780M, tomorrow I will have a 660M.

These cards are working in 10.10.5 using Web Driver. Soon I will try with 10.11.5 as well. All this before I have added EFI to make them work better. Pretty amazing considering you just stated they didn't work.

How's that shoe leather tasting?

 

[Naijaguy]

As of now, will the GTX 1080 and GTX 1070 work on Mac OS X?

 

[Asgorath]

As of now, will the GTX 1080 and GTX 1070 work on Mac OS X?

Nobody knows yet. There are no drivers that have the Pascal device IDs in them, but then again, the cards aren't even available right now. Personally I'd be very surprised if we don't see Pascal-enabled web drivers fairly soon, it took them a little while to release Maxwell drivers a couple of years ago but NVIDIA has been doing a pretty good job of enabling all the new GPUs when they come out.

 

[MacVidCards]

Going by past, yes, they will have drivers soon.

 

[flowrider]

Well, Nvidia just issued new web drivers (346.03.10f02) Available here:

http://us.download.nvidia.com/Mac/Quadro_Certified/346.03.10f02/WebDriver-346.03.10f02.pkg

Has anybody looked at this to see if, maybe, Pascal support is in there:

Lou

 

[karsten]

Well, Nvidia just issued new web drivers (346.03.10f02) Available here:

http://us.download.nvidia.com/Mac/Quadro_Certified/346.03.10f02/WebDriver-346.03.10f02.pkg

Has anybody looked at this to see if, maybe, Pascal support is in there:

Lou

this looks promising:

<upgrade-bundle>

<bundle id="com.nvidia.web.GeForceTeslaGLDriverWeb"/>

<bundle id="com.nvidia.web.NVDAResmanWeb"/>

<bundle id="com.nvidia.web.NVDAGK100HalWeb"/>

<bundle id="com.nvidia.web.GeForceGLDriverWeb"/>

<bundle id="com.nvidia.web.GeForceMTLDriverWeb"/>

<bundle id="com.nvidia.web.NVDAGM100HalWeb"/>

<bundle id="com.nvidia.NVDAStartupWeb"/>

<bundle id="com.nvidia.web.GeForceTeslaVADriverWeb"/>

<bundle id="com.nvidia.web.GeForceTeslaGAWeb"/>

<bundle id="com.nvidia.web.GeForceVADriverWeb"/>

<bundle id="com.nvidia.web.NVDAGF100HalWeb"/>

<bundle id="com.nvidia.web.GeForceAIRPluginWeb"/>

<bundle id="com.nvidia.web.GeForceWeb"/>

<bundle id="com.nvidia.web.GeForceTeslaWeb"/>

<bundle id="com.nvidia.web.NVDANV50HalTeslaWeb"/>

<bundle id="com.nvidia.web.NVDAResmanTeslaWeb"/>

</upgrade-bundle>

 

[MacVidCards]

Actually, you would want to see "NVDAGP100HalWeb" to get excited.

 

[Squuiid]

Actually, you would want to see "NVDAGP100HalWeb" to get excited.

Here are what we're looking for, in full!

http://wccftech.com/nvidia-pascal-device-id-gp-102-gpu-confirmed/

 

[netkas]

Owners of 980ti/titanx now have to think twice, get 1080 or wait for gp102. hard choice.

 

[karsten]

Here are what we're looking for, in full!

http://wccftech.com/nvidia-pascal-device-id-gp-102-gpu-confirmed/

do you know what files to look in in the web driver? i poked around a bit but i'm not familiar with the nvidia file layout

 

[Squuiid]

Owners of 980ti/titanx now have to think twice, get 1080 or wait for gp102. hard choice.

I think the 1080 is a better bet. 1080Ti power reqs rumoured to be 8pin+6pin, 250W draw.
1080 is a single 8pin (essentially 2x6pin).
Unless you want to start looking at dual PSUs I think the 1080 is a better fit for cMPs.

However, I see why you qualified your statement to 980Ti and Titan X as they too need 250W

 

[Fl0r!an]

Nvidia blogged that the current drivers only have basic beta support for Maxwell.

Beside the fact that they didn't call it basic, it's actually contradicting to your theory.
The most common definition of the term beta is:

Beta phase generally begins when the software is feature complete but likely to contain a number of known or unknown bugs.

If you were right with your "incomplete compatibility-mode driver"-theory and Nvidia was aware about the common terminology (which we can assume as Nvidia is not exactly a small backyard workshop), they'd call the Maxwell driver state something like alpha or experimental instead of beta.

I'm not denying that the Web Drivers may contain one or the other bug, but it's certainly way better than what Apple has provided on the AMD side.
Just a week ago, I was trying to help a Hackintosh guy out, who was experiencing some serious issues when connecting his R9 290X to a 144Hz display (a setup which most likely no one has used before in OS X ).
I gave him some hints on framebuffer patching and VBIOS modifications, but he didn't want to get his hands dirty. Instead he explained to may that Apple is obliged to fix this (since the presence of their pre-alpha drivers clearly shows that they fully support the Hawaii chipset ), so he decided to contact the Apple support. Well, good look with that.

This clearly shows why Nvidia doesn't call any GPU supported which they don't have to support officially. They obviously do not want to help every single hackintosher out there getting their boot loader configured properly...

 

[MacVidCards]

do you know what files to look in in the web driver? i poked around a bit but i'm not familiar with the nvidia file layout

I already told you.

Look in those driver names. "K" is Kepler, "F" is Fermi and "M" is Maxwell.

Now see if you can guess why I told you the driver needs a "P" in the name

 

[SoyCapitanSoyCapitan]

Latest web driver, latest El Cap, latest Photoshop. Before I test on the cMP with a full fat GPU, I tested on iMac 2014 with GT 755M. OpenCL enabled. You can see from this simple test that artefacts and error calculations still persist.

I created a gradient, drew some lines, added some noise, and then resized and enlarged the image from a 5000x5000 square to an 8000x6000 rectangle.

The web driver incorrectly rendered the lines I drew. Disabling the web driver made Photoshop calculate correctly again.

I will test the same and similar processes again on the GTX 680 and the 10 series (if it installs) on the weekend.

Resizing uses OpenCL (when enabled). Rendering correctly is really critical to professionals who are working with patterns and gradients. This is a real let down and Nvidia hasn't been able to fix a number of OpenCL and display bugs for many many months. I have been tracking these bugs since 2014 with the 980, 680 and this 755.

People who claimed the bugs have been fixed in the web drivers or that these web drivers fully support Maxwell when they still don't work properly with Kepler really need to apologise to readers and customers, especially those customers who couldn't fall back to the built in OSX drivers.

 

[Asgorath]

Latest web driver, latest El Cap, latest Photoshop. Before I test on the cMP with a full fat GPU, I tested on iMac 2014 with GT 755M. OpenCL enabled. You can see from this simple test that artefacts and error calculations still persist.

So 346.03.10f02 on 10.11.5? All the other links you posted before talked about corruption from the healing brush or masks, that's the first I've heard of corruption from resizing. Do you see any problems with the brushes/masks? I only saw reports that the brush corruption was fixed in 346.03.10f01.