GTX 970 is performing poorly, what gives?

[witherman]

The short version:
I installed a new GTX 970 and it's performing at about the same level as the 660ti it replaced. I'm using the current NVIDIA web drivers and the card appears to be recognized by the system.

The long version:
I installed a new EVGA GTX 970 SC, replacing a GTX 660ti, mainly for gaming. I dual boot into windows and the performance there is great. In 10.10.4 the performance is about the same as the 660ti. I installed the current NVIDIA web drivers (for 10.10.4) before swapping the cards and injecting the NVIDIA web drivers.

The GTX 970 is recognized in the System Profiler with 4095 MB of memory. Inside the NVIDIA Driver Manager control panel the web driver is selected but under the ECC section the card isn't listed under a (PCI?) slot. Slots 1-4 are shown as empty. The GTX 970 is listed as slot -. Firing up the Galaxies app the GTX 970 gets about 60fps. The 660ti would get around 45fps. Testing out Cinebench (OpenGL), the 970 clocked in at about 45fps, slightly worse than the 51fps the 660ti would do. In Windows it gets around 100fps.

I did some research and installed an updated version of the AppleGraphicsControl.kext and saw some improvement. Cinebench testing now yields 54fps.

I've played around with different boot flags and couldn't find anything that made a difference. When GraphicsEnabler=Yes the 970 would properly be shown in slot 1 within the ECC section of the NVIDIA Driver Manager. Performance was unaffected.

Watching the HWManager stats during the Cinebench test showed the GPU Core hitting 1.05 Ghz and the GPU memory peaked at 7.1 GHz. The upper clock limit of the 970 GPU Core should be around 1.3 GHz. It seems like something is throttling the GPU.

I installed the current NVIDIA CUDA drivers and didn't see any difference.

Aside from waiting for new drivers from NVIDIA, is there anything I can do? I don't mind booting into Windows for gaming but I'd like the card to perform better than the 660ti it replaced.

My 10.10.4 install uses the current MultiBeast boot loader (Chimera v4.1.0) and I'm using the correct DSDT for my board (GA-x58a-UD3R REV. 2 BIOS FH). Core i7 970 (3.3GHz, normally OC'd to 4.2GHz. 6 GB (3x2GB) 1600MHz RAM. Boot disk is a Samsung EVO 850 (250GB) SSD.

Here's my org.chameleon.Boot.plist:

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
<key>EthernetBuiltIn</key>
<string>Yes</string>
<key>GenerateCStates</key>
<string>Yes</string>
<key>GeneratePStates</key>
<string>Yes</string>
<key>Graphics Mode</key>
<string>"1920x1200x32"</string>
<key>HDAEnabler</key>
<string>Yes</string>
<key>HDEFLayoutID</key>
<string>01000000</string>
<key>Kernel Flags</key>
<string>kext-dev-mode=1 npci=0x2000 PCIRootUID=1 nvda_drv=1 </string>
<key>Legacy Logo</key>
<string>Yes</string>
<key>Timeout</key>
<string>2</string>
<key>UseKernelCache</key>
<string>No</string>
</dict>
</plist>

and my smbios.plist:

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
<key>SMfamily</key>
<string>Mac Pro</string>
<key>SMproductname</key>
<string>MacPro3,1</string>
<key>SMboardproduct</key>
<string>Mac-F42C88C8</string>
<key>SMserial</key>
<string>###############</string>(note: my serial removed)
<key>SMbiosversion</key>
<string>MultiBeast.tonymacx86.com</string>
</dict>
</plist>

Thanks for the help!

 

[Fl0r!an]

You migth want to do some more meaningful benchmarks, Cinebench is just crap.
It will report the exact same performance (variance 2-3 fps) for all of my cards, ranging from HD 5770 over 7770/5870 up to R9 280. I think it's running extremely CPU-limited in OS X.
In real life situations (or better benchmarks like Unigine Valley) I can see the expected improvements in OS X (R9 beeing 2-3 times faster than the slowest of my cards).

 

[kanttii]

My 970 is not working well either. I suspect it's because it's not boosting like at all in normal usage like web browsing, so it stays at 135Mhz core and 648mhz memory according to HWMonitor... I started looking into this because the scrolling performance is horribly bad on just about every site. With sites with only a little of text it scrolls better, but still lags. I have the newest drivers 346.02.02f03 and they helped a bit.. but it's still really bad. I modded my vBIOS so that it should use the highest clocks all the time, but that works only in Windows. On OSX it still stays at very low clocks. How can I fix that? And the max it boosts to is 1.22Ghz when in Windows it properly goes to 1506Mhz as my BIOS mod tells it to, when gaming. I removed the baked in clocks after realizing it didn't help OSX at all...

CPU is switching clocks nicely, but this GPU thing is driving me nuts. On Windows and same version of Chromium it's butter smooth. OSX and Chromium and Safari, horrible.

Is there any way to force it to run at higher speeds or edit the steps it takes, like min clocks etc? On Windows I've set the default to be 1114Mhz and it works properly there too. OSX, not so. Apple drivers didn't let me even boot so I guess I gotta be using the NVIDIA ones...

TIA!

 

[witherman]

Thank you for the feedback. I ran the suggested Unigine benchmark, results below.

Unigine Valley Benchmark 1.0

FPS: 43.5
Score: 1820
Min FPS: 22.0
Max FPS: 81.2

System
Platform: Darwin 14.4.0 x86_64
CPU model: Intel(R) Core(TM) i7 CPU 970 @ 3.20GHz (3197MHz) x12
GPU model: NVIDIA GeForce GTX 970 (4095MB) x1

Settings
Render: OpenGL
Mode: 1920x1200 8xAA fullscreen
Preset: Custom
Quality: Ultra
Powered by UNIGINE Engine
Unigine Corp. © 2005-2013