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

nVidia GTX 1080

i make movies said:
Did anyone manage to get their hands on on these bad boys yet and pop it in a Mac just to see what's what?
Click to expand...

If I can find a single 8 pin card this weekend I will test to see if the UDA part of the web drivers allows some forward compatability. If it works I will test:

Luxmark
A Photoshop OpenCL action I created today
Unigene Valley
Cinebench
All the tests in recent Barefeats articles

Anything you guys ask me to test if I have the software. I don't have FCPX sorry.

I will also bench the same tests in Windows on my 6700K.
[doublepost=1464379642][/doublepost]
tuxon86 said:
Actually, as JayzTwo said on it's video, you only need to up the voltage limitation and slightly raise the fan speed. By default the voltage limit is locked and it also cause the card to throtle.
Click to expand...
Yeah but you need Windows based utilities for that, which was the point I was trying to make earlier.
 
Here's some preliminary OpenCL Testing. I'm happy to say Nvidia has fixed some of the bugs in the latest web drivers for 10.11.5, but it seems they have not completed bug fixes with all GPUs.

The Photoshop action is attached. It basically creates gradient, noise and a pattern and then checks if the GPU can correctly calculate and render the position of the pattern after performing an OpenCL based image resize. If the result is a fail then more complex images would suffer.

Make sure all the Advanced options are enabled in Performance settings and relaunch the app. Also, make sure you have the latest web driver as there have been two releases for 10.11.5.

These are the results for two machines I tested so far. I no longer have a 9 series card so forum members who can test this would be greatly helping others.

iMac 2014 with Geforce GT 755M : Fail, see image attached. Artefacts and incorrect positions are seen on this GPU.
cMP (dual X5650) with Geforce GTX 680 : Pass, see image attached.

Until I get more results, it looks like beta supported GPUs (9 series and mobile GPUs listed in the support notes) are still prone to OpenCL errors. Results will vary from chip to chip.

Brushes, stamp tool and healing brush have been fixed, no artefacts were seen on either machine. On the cMP the action takes 5 seconds to complete on the GPU, 12 seconds to complete on the CPUs.
 

Attachments

  • OpenCLResizeTest.atn.zip
    1.8 KB · Views: 399
  • settings.png
    settings.png
    157.7 KB · Views: 200
  • 680.png
    680.png
    1,002.9 KB · Views: 195
  • 755.png
    755.png
    1.8 MB · Views: 205
Last edited:
Squuiid said:
Here's more info on the single 8-pin EVGA 1080 SC. Definitely the card to go for in a cMP.
Reference PCB but with custom cooler and dead silent at idle. Minor overclock is a bonus.

http://www.anandtech.com/show/10360/zotac-and-evga-reveal-custom-geforce-gtx-1080-designs
Click to expand...

Yes, this is definitely my opinion, too. Nice card. :)

Some questions @ MacVidCards:

- All these GTX 1080 cards are DisplayPort 1.4 'ready'. Is this meaning DP 1.4 is already active, just not yet officially confirmed because of lack of displays? Or the cards have to be upgraded with a new firmware? When executing this firmware upgrade will the card loose the possible future MVC-EFI-Firmware?

- This card has an enormous high over clocking potential even with only one 8 pin power connector (+ ~20%). When overclocking in Windows with the EVGA Precision OC tool, will the possible future MVC-EFI-Firmware be overwritten? Or will MVC not only create the EFI firmware, but also adjust the best stable overclocking for this card?

oc.png

 
Last edited:
  • Like
Reactions: m4v3r1ck
OC doesn't necessarily write to firmware. I have software OC control of CPU and GPU in Windows they don't write permanent changes to BIOS or firmware.

Those 1080 boosts and overlocking are software controlled features and to get the most out of them you need native software to control the settings, driver support for GPU Boost 3.0 and stable power delivery (that's why so many cards are coming out with 8+6 pin)

I checked all the main PC stores today. None have 1080 in stock yet.
 
Last edited:
  • Like
Reactions: Synchro3
@karsten , Can you leave a link to the finding?
AFAIK, DSDT injections are used to assign functions/ports through patches for hardware that already has drivers for it in OS X... I don't see how one can enable 1080 in OS X other than to cripple it to Kepler card for instance (or Maxwell best case scenario, but then i don't see that either will work)?
 
Last edited:
owbp said:
@karsten , Can you leave a link to the finding?
AFAIK, DSDT injections are used to assign functions/ports through patches for hardware that already has drivers for it in OS X... I don't see how one can enable 1080 in OS X other than to cripple it to Kepler card for instance (or Maxwell best case scenario, but then i don't see that either will work)?
Click to expand...

it's a closed group so i cant link it but heres a screen grab. i'm not well versed in the DSDT method but i think it just enables some sort of compatibility mode.

https://goo.gl/photos/fMs7SqoJgcJQwfJP6
 
  • Like
Reactions: Synchro3 and owbp
Yeah, I don't believe the claim that it's fully accelerated. Would be interesting to see some benchmark numbers.
 
He might have spoofed the device ID into Maxwell range, but I don't know why you'd want to do that with a DSDT, since that could be achieved in Clover in seconds.

I wouldn't expect much more than a kernel panic from doing that though.
 
  • Like
Reactions: TheStork and owbp
SInce gp104 is just a shrink of maxwell to 16nm ff + few minor features than gm100 kext might work, but it also means nvidia can update the driver for pascal support sooner than later.
[doublepost=1464468886][/doublepost]
Fl0r!an said:
He might have spoofed the device ID into Maxwell range, but I don't know why you'd want to do that with a DSDT, since that could be achieved in Clover in seconds.

I wouldn't expect much more than a kernel panic from doing that though.
Click to expand...
There is more things than just device-id. for example, in mmio bar there is a number to describe generation, (nve4, nvf0, nv114 sounds similar to you?), it's very hard to fake them (fakesmc is such example, needs a kext)
[doublepost=1464470401][/doublepost]That's great 1080 card, with proper power delivery and cooling systems, boosts over 2 ghz on stock speeds.

 
  • Like
Reactions: Synchro3, Fl0r!an and karsten
so worst-case scenario; nvidia doesn't release drivers for the 1000 series, we continue to use a supported card in slot 2 for os x and put the gtx 1080 in slot one for bootcamp?

i'm scared because they released the bsd and windows drivers already. bsd really?
 
Even if it were possible for the web drivers to include a kext for GP100 soon, at the moment Maxwell GPUs are still in beta since almost a year with no official released products and the performance in OSX is downright embarrassing compared to Windows. Not to mention the OpenCL bugs that need to be fixed.

http://barefeats.com/razer_core.html

It is going to be a loooong time before Pascal can show its true muscles in OSX, if ever. So have a realistic view of what you could expect from a Pascal under OSX. Good performance and accurate compute capabilities require a lot more than a kext.

The first review of the 1070 is out. Roughly the same as a Ti/Titan X. I'm going to SLI a pair of these in my Skylake build.

http://www.clubic.com/carte-graphiq...st-nvidia-geforce-gtx-1070-pascal-gamers.html
(Taken down for breaking NDA)
 
Last edited:
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