Jump to content
  • Announcements

    • Allan

      Forum Rules   04/13/2018

      Hello folks! As some things are being fixed, we'll keep you updated. Per hour the Forum Rules don't have a dedicated "Tab", so here is the place that we have our Rules back. New Users Lounge > [READ] - InsanelyMac Forum Rules - The InsanelyMac Staff Team. 
rominator

Good news afoot in 10.8 Mountain Lion for Fermi cards

43 posts in this topic

Recommended Posts

Hey guys, wanted to be the bearer of good news.

 

Trying out 10.8 on a Mac Pro, I discovered that it is no longer necessary to add GTX560 id, my guess is this applies to all GTX5xx cards.

 

Also, no more need to fiddle with AGPM kext, cards throttling by themselves

Share this post


Link to post
Share on other sites

Damn good news!

Only pitty I just bought a Quadro4000.

 

Very curious if there are any SandyBridge-E references, would be nice!

Share this post


Link to post
Share on other sites

Is the 5xx card's HDMI working now ?

Because I found only one DVI port is working for MSI N550GTX-Ti Cyclone II in Mountain Lion DP1 without Natit.kext.

In 10.7.1/10.7.2 I can get two DVI ports working at the same time with Natit.kext.

In 10.8 still get 2 DVI ports working at most with Natit.kext.

How can we make GeForce 5xx card's HDMI working in Lion or Mountain Lion ?

Share this post


Link to post
Share on other sites

It's sort of a unified driver now.

 

<key>IOKitPersonalities</key>
<dict>
 <key>NVidiaRM</key>
 <dict>
  <key>CFBundleIdentifier</key>
  <string>com.apple.nvidia.nvGF100hal</string>
  <key>IOClass</key>
  <string>NVDAGF100HAL</string>
  <key>IOMatchCategory</key>
  <string>IOService</string>
  <key>IOPCIClassMatch</key>
  <string>0x03000000&0xff000000</string>
  <key>IOPCIMatch</key>
  <string>0x000010de&0x0000ffff</string>
  <key>IOProbeScore</key>
  <integer>60000</integer>
  <key>IOProviderClass</key>
  <string>IOPCIDevice</string>
 </dict>
</dict>

Share this post


Link to post
Share on other sites

no more need to fiddle with AGPM kext, cards throttling by themselves

I can confirm that on GT 425M, needed to patch AGPM on SL and Lion, now on ML it works OOB.

 

capturadetela20120217s1.png

Share this post


Link to post
Share on other sites

I can confirm that on GT 425M, needed to patch AGPM on SL and Lion, now on ML it works OOB.

 

capturadetela20120217s1.png

I can also confirm my 460 is working fine OOB, no OpenGL fix needed. However, Luxmark does not run properly, so I think I may need an OpenCL fix.

 

Proof: post-626596-0-49870900-1329501242_thumb.png

Share this post


Link to post
Share on other sites

YEAAAH! So 460 works according to Blah101! Any fermi freezes / KP's ? I'm downloading 10.8 right now and will post my results over weekend. I'm using Gigabyte version of GTX 460 768MB !

 

I bet there won't be freezes since it seems that AGPM is finally history acording to Rominator! YEA!

Share this post


Link to post
Share on other sites

YEAAAH! So 460 works according to Blah101! Any fermi freezes / KP's ? I'm downloading 10.8 right now and will post my results over weekend. I'm using Gigabyte version of GTX 460 768MB !

 

I bet there won't be freezes since it seems that AGPM is finally history acording to Rominator! YEA!

No freezes yet. Then again, I really haven't had frequent freezes since SL, I had maybe two or three in my whole time on Lion. Nothing on ML so far.

Share this post


Link to post
Share on other sites

I went ahead and replaced my Lion install with a fresh Mountain Lion. My GTX 460 worked OOB with my DSDT edit but how can I check that the power management is working without a AGPM edit?

Share this post


Link to post
Share on other sites

i finally got around to test this on my hackintosh (ml installer simply crashed or froze on my gtx 480, so i made an image of my mbp ml install, copied it and for whatever reason it just works ... dunno what happened there).

 

so, some additional observations to my earlier post:

  • opencl works ootb on fermi (as other people have already mentioned), but apple seems to have removed the sm_20 code and hard setting it to 1.2 or 1.3 won't work either (-> luxmark and all other programs that want to use more complex opencl stuff (like local memory atomics) will fail)
  • small upside: they finally managed to display the correct clock frequency (cache sizes are still missing though)
  • opengl 3.3/4.x and opengl es 2.0 are not supported (yet?) - context creation fails if i set NSOpenGLPFAOpenGLProfile to 0x3300 or higher
  • there is definitely some kepler support (try find /System/Library/Extensions -type f | xargs strings 2>/dev/null | grep -i -E "kepler|gk11|gk10|gk20" in terminal)
  • apple forgot to strip the symbols from libclh.dylib (this will make some peoples lives easier ;))

 

also, can anybody with access to a geforce 200 series / gtx 260+ / ptx 1.2/1.3 card check if luxmark is still working correctly?

Share this post


Link to post
Share on other sites

It looked to me like that my 460 was running at full speed without a LegacyAGPM kext so I put it back and it's switching power states fine again (going by the console log anyway). With logging enabled in the vanilla AGPM kext (and no Legacy one installed) I didn't see any state switching in the console so I don't think the card was throttling automatically. Perhaps it only works on certain cards/configs.

 

Also, the AGPM kext has changed slightly. The model identifiers are under a "Machines" subdirectory now.

Share this post


Link to post
Share on other sites

Thanks cmf about the detailed info why OpenCL cant work if normal (complex) OpenCL code is used and onyl such easy things like OCLinfo works.

Share this post


Link to post
Share on other sites

okay, i take back what i said earlier. i did some more digging and sm_20/fermi is fully supported (as is sm_30+, btw.).

the reason why it currently fails for some opencl programs is because of bugs in apples/nvidias code (e.g. using a mem_fence, read_mem_fence or write_mem_fence inside an opencl kernel will result in a build error).

since this is only the first preview and they changed lots of code, i guess this is to be expected.

 

i'm not sure where the problem with luxmark is, but it's probably a similar issue.

 

 

on a more or less related note: there is some interesting source code /System/Library/PrivateFrameworks/GPUSupport.framework/Versions/A/PrivateSources/ (tessellation shader anyone? ;))

Share this post


Link to post
Share on other sites

Good news, I hope that my EVGA 450 GTS now don't has any freezes anymore.. :)

 

Seems to be likely an Opengl issue, that they now, from what I read here, finally support. :)

Share this post


Link to post
Share on other sites

Is there any way to confirm that the cards no longer need an AGPM edit for power management? I know my 460 is probably running full speed when I benchmark it, but how can I tell if it's throttling down when I'm just idling at the desktop?

Share this post


Link to post
Share on other sites

@Riley Freeman you could always use iStats to check on the temps of your card during the bench and when it's done and just idling at the desktop and see if there's a difference between the temps there.

Share this post


Link to post
Share on other sites

Start OpenGL Extensions viewer and run the test in Full Screen. If it starts out with FPS in the hundreds and then suddenly jumps to 4000+ FPS then throttling is working.

 

Good idea!

 

I tried this a couple of times. First it started out around 2500 and quickly ramped up to around 3200. Then I left it for a few minutes, tried again and it seemed to start around 1000 or so. So it looks like throttling is working!

Share this post


Link to post
Share on other sites

That doesn't sound right to me - my 570 starts in the 100's and a second or so later jumps up to 5,900 ish on the first run - may depend what you have running in the background though.

Share this post


Link to post
Share on other sites

I've nothing running in the background. It's just idling. I doubt the model identifier matters if the throttling is being done in the nVidia drivers but I have mine set as iMac9,1.

 

I'll try it a few more times with different pauses in between to make sure everything is nice and idle but it certainly didn't start off as low as 100s. And it made it up to the full 3200-ish within a second or two. It's kinda hard to see exactly what it starts at.

 

Would speedstepping also be a factor? I have it disabled here as I have my C2D overclocked. So the CPU is always at full speed.

Share this post


Link to post
Share on other sites

I'm seeing the exact same behavior as Andy (low hundreds and then boom) except 5900 FPS is insane!

 

I'm getting around 4300 here at 1280x1024 - GTX 460, 10.7.3 with the new nvidia drivers, MacPro3,1 identifier - which means no AGPM.kext loaded.

 

AGPM loads on iMac9,1 so you might have a conflict of interests there? Remember that the iMac9,1 is built almost like a laptop, it makes sense for its AGPM configuration to employ agressive throttling.

 

Even if you have customized it yourself, try temporarily removing AGPM.kext and see what happens then. I believe it's safe to remove it!

Make sure to do sudo touch /System/Library/Extensions after moving it out so that it doesn't linger in the cache.

 

3200 FPS sounds kind of low which I guess confirms my theory, unless you're running at a really high resolution?

Even then I don't think your score should be a whole 1000 FPS lower than mine, that's a bit much.

Share this post


Link to post
Share on other sites

I stopped using the legacy AGPM kext a while back so I'm not sure the AGPM kext is doing much as it isn't going to match the 460's vendor/device id without the legacy kext override. When I had it installed I was getting scores around 2400 as I had it geared for power management rather than raw power. So just removing the legacy AGPM has given a 33% performance boost.

 

The benchmarks are running at 1680x1050 as that's my native screen res. Also, the values I'm throwing out are just for the first test (OpenGL 1.1).

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now


  • Recently Browsing   0 members

    No registered users viewing this page.

×