Conflict with macOS 10.12 & 10.13 Web Drivers and SETI@Home CUDA App?

I’ve been working on the SETI@Home Linux & MAC CUDA App for a couple of years and the latest version has some conflict with the current macOS nVidia drivers which doesn’t exist in Linux. Basically, the MAC build will miss Pulses depending on the Monitor state. Simply changing the display connection from one GPU to the other seems to solve the problem. Also, simply turning the monitor off seems to stop any further loss of data. The macOS App will work for weeks with the monitor turned off.

To use the machine while running the App it’s necessary to connect the monitor to an AMD GPU and leave the NV GPUs disconnected from any monitor, this way the App will also run for weeks without missing any data. The developer claims this is due to a Bug with the macOS display driver and cannot be fixed.

Any thoughts?

I’ve been working with T-Bar’s Mac CUDA App for awhile, now, on two machines. One is a cMP Mid-2010 5,1 Dual Xeon 2.4GHz System, and the other is a Hackintosh.

BOTH Systems exhibit the “Best Pulse = 0” Errors while performing CUDA Number Crunching Operations for the SETI@Home Project.

The cMP System has TWO GPUs, one ATI HD-5770, (monitor connected), and one EVGA GTX-1050 - NO monitor connected. On this System IF I turn the monitor on the ATI Card OFF the System crashes MULTIPLE Work Units with the “Best Pulse = 0” Errors.

The Hackintosh has one MacVidCards’ NVIDIA GTX-1070 8GB Card, and one EVGA GTX-1050 Card. On this System, IF the monitor is left ON, (connected to the 1070), the System exhibits the “Best Pulse = 0” Errors on MULTIPLE Work Units.

BOTH Systems are on High Sierra - (17G4015), with NVIDIA Web Driver 387. ANOTHER issue with the Web Driver is that on the Hackintosh IT REVERSES the PCI-e Video Card Slot Order. Thus, making PCI-e Slot 1 of the MOBO seen as Slot 2 by MacOS, and Slot 2 of the MOBO seen as Slot 1 by MacOS. This rendered OpenCL and OpenGL Support on the MacVidCards’ GTX-1070 GPU as DISABLED/Non-existent. When I found this, it was because Blizzard Games, (via the App), popped up a Dialogue Box stating “Missing OpenGL”, and “Please Install/Reinstall Web Driver.”

Thanks to T-Bar, I switched the physical Card Locations on the MOBO, and changed the PCI-e Video Boot Slot to PCI-e Slot 2. THEN, via BOINC for SETI@Home, OpenCL showed up for the 1070 Card, AND App was able to let me Launch StarCraft:Remastered - so OpenGL was also restored to the GTX-1070. HOWEVER; NOW, the GTX-1050 IS MISSING OpenCL and OpenGL Support!

It seems that Web Driver Support has FALLEN since MacOS Sierra. Though, El Capitan had some issues with Web Driver as well…

I wish that NVIDIA would put more effort and quality into supporting the Web Driver for MacOS. I sincerely HOPE and PRAY that Apple and NVIDIA come together for some BETTER Web Driver Support for Mojave.

[EDIT:] I forgot to mention that I’m on the Latest CUDA Driver, 410.130.



----> Bug Report Submitted: <----

Thank you for reporting a new issue. The bug ID is: 2484445

Bug Summary:
RE: MacOS NVIDIA Web Driver Issues.

Things that need to be addressed in MacOS NVIDIA Web Driver:

1.) In Sierra and High Sierra, the Web Driver REVERSES how MacOS sees the physical PCI-e Slot Locations.

PCI-e Slot 1 is seen by MacOS as PCI-e Slot 2, AND PCI-e Slot 2 is seen as PCI-e Slot 1.
Now, on cMP Systems, (4,1’s Flashed to be seen as 5,1’s and “Real” 5,1’s), I’m NOT entirely sure if this is still true; BUT, this is prevalent in Hackintosh Systems! So, the “cure” is to switch physical locations of Primary and Secondary GPUs in the System. Further, in Hackintoshes, this also means going into the BIOS and changing PCI-e Video Boot Slot Order to PCI-e Slot 2.
2.) In regards to #1, IF cards are NOT physically switched, this has the effect of STRIPPING OpenCL and OpenGL from, (what should be), the Primary GPU!

I witnessed this myself on a MacVidCards’ NVIDIA GTX-1070 8GB Card where BOINC, (Berkeley Open Infrastructure Network(ed) Computing), App clearly shows in MacOS Sierra and High Sierra that OpenCL is MISSING from the GPU!

Further, in Blizzard’s App, when Launching a game, (let’s say StarCraft:Remastered), App responds with an IMMEDIATE Popup stating “Missing OpenGL”, AND “Install/Reinstall Web Driver”!

Once GPUs have been switched so that, (in my case the GTX-1070 8GB), the Primary GPU is in physical location PCI-e Slot 2, (and the BIOS has been properly modified for this), BOINC now shows that the Primary GPU HAS OpenCL and NO LONGER gives the Popups and INSTEAD allows StarCraft:Remastered to Launch and Play.

HOWEVER; NOW, the Secondary and Tertiary GPUs, (if ANY in the System), NOW ARE STRIPPED of OpenCL and OpenGL - in my case, my Secondary EVGA GTX-1050 2GB Card.

3.) In addition to all of the above, with CUDA Driver 410.130 Installed in High Sierra, IF a System ONLY has NVIDIA Cards in the System, (in my case, my Hackintosh noted above), when performing Number Crunching Operations of CUDA Type in BOINC the attached Monitor to the Primary GPU MUST be turned OFF! IF NOT, this results in MASSIVE Crunching Errors of “Best Pulse = 0” Errors!

On a cMP with MIXED GPUs, (One NVIDIA Card and One ATI Card), and the monitor is attached to the ATI/Non-NVIDIA Card - the monitor MUST be ON to properly and successfully perform CUDA Number Crunching Operations of CUDA Type! IF NOT, MASSIVE Crunching Errors of “Best Pulse = 0” Errors occur.

SEE: Conflict with MacOS 10.12 and 10.13 Web Drivers and SETI@Home Number Crunching.

THESE ISSUES MUST be addressed and properly resolved BEFORE releasing MacOS NVIDIA Web Drivers for MacOS Mojave!

With the above in mind, I hope that SOON Mojave NVIDIA Drivers are released. I have signed the Petition.

This bug submission and its status can be viewed or updated here:

Thank you,
Developer Relations Team

----> Bug Status: Open. <----

I hope this is taken seriously.

Did you try 378. driver version?


Since I’m on High Sierra, AND working with the Number Crunching Application BOINC, it is NOT recommended to “Force” an older Driver to work in a newer OS. So, NO I have NOT used 378.

I’m sure T-Bar can provide more detail on why one should use the Recommended Driver for each OS Version, and each Security Updated Version of MacOS.

If you have data that shows that BOINC and the above noted issues do NOT occur in 378., we’d be grateful to see your data and supporting documentation and information.

I can tell you that even in El Capitan, after Upgrading to 10.11.4 from 10.11.3 that T-Bar and I witnessed my System having HIGH Numbers of Inconclusive and Invalid Results in BOINC-SETI@Home Work Units. To resolve this, T-Bar had me move from his CUDA75 to his OpenCL App for Crunching at SETI@Home while running El Capitan…

As explained in my prior posts and in T-Bar’s Post, Sierra and High Sierra Web Drivers ARE MUCH worse!