NVIDIA GeForce 5xx Graphics in Lion [TEST]

[TouchMyBox]

Is there a fix yet for people like me who can only get OpenGL to work by starting OSX up in verbose mode?

Kind of a pain if I have to wait 10 minutes for my OSX to load up.

 

[rdurty2]

Is there a fix yet for people like me who can only get OpenGL to work by starting OSX up in verbose mode?

Kind of a pain if I have to wait 10 minutes for my OSX to load up.

10 minutes to start up!? Damn, you need an ssd. Normal or verbose mode both start up in around 20 seconds.

 

[rdurty2]

Ok GTX580 owners, some big news!

First off, and this is a big one, MacPro3.1 does NOT load the agpm.kext So if you have been running MacPro3.1 system def the edits to that file from insanelymac do not even run on your system.

Now I don't know how this applies to others but here is what I found for my system:

MacPro3.1 - works just fine but does not load agpm.kext
MacPro4.1 - kernel panic upon bootup
MacPro5.1 - kernel panic upon bootup
iMac12.2 - omg it boots!!!!!!!!

Since I was able to boot with the iMac12.2 definition I ran some benchmarks. Cinebench gave me an enormous 11fps. Pitiful. Despite iMac12.2 loading agpm.kext it had no entry for the 580.

I edited one of the lines in agpm.kext as per advann and added 1080 (GTX580 device ID). Well that did absolutely nothing. Scores were still 11fps.

Then I opened the modified agpm.kext from insanely mac to see what was actually edited. Well the only edit was for the MacPro4.1 definition. So if you weren't running MacPro4.1 that edited kext file was doing nothing.

I then copied the modified entry from that and added to the iMac12.2 agpm.kext section and BAM!

Here are some screens:

Cinebench - I do NOT like Cinebench as a benchmark but I figured I would throw it up anyways. Why don't I like it? The results of the OpenGL test are affected by the CPU. So if someone (advann) has a lower GPU then me but a high CPU then he will bench the same or higher then me. So take Cinebench results with a grain of sale.

Despite this my Cinebench score went from 38fps (macpro3.1 def) to 42.03fps.

Now here is where it gets REALLY good

NovaBench

We should be comparing NovaBench scores instead of Cinebench

advann and his GTX560ti had a Score of 1123, and FPS of 1989. You can see my 580 destroyed the 560 score!

Attached is my modified agpm.kext for iMac12.2 system def ONLY

 

[Rick5]

These are scores with a 3,1 system definition.

 

[rdurty2]

Rick5 - interesting to see you have similar numbers using 3,1. Both benches were considerable lower using 3,1.

There really seems to be no consistency between setups.

 

[Rick5]

Rick5 - interesting to see you have similar numbers using 3,1. Both benches were considerable lower using 3,1.

There really seems to be no consistency between setups.

I believe that to be a point well taken. There is no consistency from any of the scores I've seen. All of these benchmark apps are flawed in one way or another and as many have stated, they "need to be taken with a grain of salt". As long as a person is happy with his or her set-up and the performance provided ... well ... that's all that really matters. We can tweak this and that till we are blue in the face, but there is no real way to quantify the results. Having a system that is satisfactory to the user seems like it should be the benchmark each should strive for. {just my opinion}

I like to play with certain aspects of the system just to see what happens, but all in all, I've gained very little from where I'm at right now. I'm pleased with the way things are working, so, unless someone comes up with something really mindblowingly phenomenal, I think I'm pretty much going to leave things as they are...until 10.7.1 or 10.7.2 when everything goes to hell in a handbasket again!!!

Gotta love those gray falling curtains ... and I still can't find out, ANYWHERE, what npci=0x2000 does. But I'd really like to know!!!!

just my 2 cents ...

Rick
Aurora, IL.

Even when the question was posed in the Lion forum no one responded ... not even the mods.

 

[Jordan2Delta]

@rdurty2

The AGPM.kext contains values I haven't seen before.
Somebody has put higher values in Threshold_Low compared to Threshold_High.
It doesn't make sense. Unless you wouldn't wanted G-state 1 and 2 executed. Looking at your benchmarks it seems to go all the way up correctly to G-state 0. I'd like to know if it still downclocks correctly to G-state 3 and the history of G-states. I am not going to put these values in my system. Someone other shall do it please.

The GTX 570 and 580 are very power demanding cards, so I don't want to either use the highest or lowest power mode. The lowest power mode is lagging and as a former Ati 5770 card owner it's just unbearable for me.

The following threshold values are chosen to stay most of the time in G-state 2.
It’s sufficient for most “everyday” usage.

With these settings I can save up to 50~110W
High
0::25
1::50
2::99
3::100

Low
0::0
1::26
2::51
3::99


I have also set IdleInterval to 400 and SensorSamplingRate to 25
By doing so, you get 92s delay between G-state changes. Luckily it goes up to high performance modes quickly enough. It takes about 3 seconds to get in a higher G-state.

CAUTION:
I have found out how they work myself by trying a few different values. I don't know if it's ok to alter them actually. I am no GPU developer.

== Cinebench's score mystery ==
When running it with above thresholds I get ~30fps just like I did with other settings before. I could see that cinebench isn't demanding enough to invoke G-state 0. It used to change between G-state 2 and 1 back and forth during the benchmark. GPU load is 70%!!! idle during benchmark at G-state 1.

Well, that has changed with the new values for IdleInterval and SensorSamplingRate.
Now G-state 1 is being activated all the time the bechmark is running. Annnnd, guess what? I still get only ~30 fps.


@Rick5
npci=0x2000 is used to get beyond the infamous [PCI config begin] issue at boot time. You can only see it when starting up using verbose mode. In reality it boots fine till the login screen without npci-0x2000. But as you can imagine, people think the machine has frozen at [PCI config begin] and are starting to reset/reboot their machines.

That's why

 

[rdurty2]

All the values that are in my agpm are from the AGPM.kext for 580's found on insanelymac. All I did was copy and paste.

Attached to the first post here: http://www.insanelymac.com/forum/index. ... pic=260074

This is the agpm.kext that pretty much every single 580 user is using. All I did was copy the values from MacPro4,1 to iMac12,2

 

[rdurty2]

Jordan - Can you please post up how to check the gpu states? I know you need opengl driver monitor, but I can't find it anywhere.

Also my 580 is all cool to the touch. Even the heatpipes are nice and cool to the touch.

 

[Jordan2Delta]

This is the agpm.kext that pretty much every single 580 user is using.

Really? To my eyes the values are faulty. It might work nonetheless, but...anyway this should be an interesting read! Thanks for the link.

Oh, just one word, while reading your previous post I was well aware you were using AGPM.kext from another person. Sorry, if was a bit unclear about that. I thought it would be enough when writing "Somebody has put higher values...."
Um, well, nevermind.

About checking GPU power state (G-states changes)......
Basically you need to change LogControl's value from 0 to 1.

With opengl driver monitor I only check GPU LOAD.
The threshold values inside the AGPM.kext are actually measured as % IDLE GPU
That still confuses me at times.