10.6.4 Really bad OpenGL performance with HD 5xx Solved

[sch8mid]

Looking for testers

 

Got a HD 5770 here in 10.6.4

 

tested

 

a. in 32 bit mode with netkas drivers and ATY-init

b. in 64 bit mode using 1.9 Aseres BL modified for 5xxxx

 

Resolution-Quartz-dual Monitors - HDMI -TV all work

 

OpenCL performance is very good with smalluxGPU benches (better than ATI 4870)

 

BUT :

 

Got severe degradation in "OpenGL Extensionviewer" benches

and I think I am not the only one..

 

you can download this very good app , that shows you a lot information about your GC here:

 

http://www.realtech-vr.com/glview/

 

 

In non benchmark mode my card hits 2800 fr .

My 3 year old ATI 3870 4500 fr.

 

 

And in Benchmark mode and activated Multisample and Anisotropy

 

the card gets completely down on its knees .

 

look:

 

 

Can someone please confirm this ?

 

please be sure that you hit the benchmark ,multisample,and anisotropy button !

 

 

Now the intersting part of the story :

 

Morpheus/ Netkas 5xxx card seems to behave much better :

 

 

 

And can someone of the more enlighted specialist shed a light on this ??

 

Best

 

armin

[The Finale of Seem]

It's a longshot, but what model is your hackie pretending to be? I only ask because I just found out through some trial and error that 10.6.4 does not like Nvidia cards + Mac Pro 4,1 in OSx86. I was getting a fraction of the real performance until I edited the SMBIOS.plist file to make it a Mac Pro 3,1 instead, which magically restored performance. Don't ask me why, but it's a wild stab in the dark that might just work out for you.

[sch8mid]

Hi FoS

 

Thats right and very interesting

and its true , I am on Macpro 4.1

and will try your suggestion

 

though in geekbench , measuring cpu and fpu performance

there is no gap between my former mac pro 3.1 under 10.5.8

 

Thanks a lot

[sch8mid]

sorry to say

edited smbios.plist

 

macpro 3.1 no positive effect

macpro 2.1 no positive effect

 

here is my current :

 

 

would be nice if one of the 5770 owners would add his personal test here

cause I really have no clue , why i got so bad results compared

to netkas tests.

[xX_SynT@x_Xx]

I made a test in 32bit. 64bit is not yet possible.

 

HIS ATi 5770

 

 

[sch8mid]

Thanx XX

 

so we see the absolute same behavior

 

Whats going on here ???? --and why does noone bother of the 5xxxx users ??

[M0rpheus]

The screenshot of the OpenGL extensions viewer from Netkas' site was originally from my Hackintosh, as I was the first to get the 5870 working. I did however have a newer OpenGL and OpenCL framework installed on my machine than the normal 10.6.4. I installed the 10.6.4 update that was meant for the Mac Mini and contains the driver kexts for the 5000 series. These newer frameworks could have resulted in higher benchmark results for me.

 

I recommend you download the Mac mini 10.6.4 specific update and extract the frameworks from that using Pacifist, and install them yourself.

[sch8mid]

thanx Morpheus...hat off

this is a phantastic reply

and saves me and others a lot of headache

I even installed 10.6.4 twice

 

____

 

But:

Whats the safest way to install the frameworks in question :

OpenCL

OpenGL

___

Quartz ?

QuartzCore?

 

Select it and install it with Pacifist ?

[xbb]

The screenshot of the OpenGL extensions viewer from Netkas' site was originally from my Hackintosh, as I was the first to get the 5870 working. I did however have a newer OpenGL and OpenCL framework installed on my machine than the normal 10.6.4. I installed the 10.6.4 update that was meant for the Mac Mini and contains the driver kexts for the 5000 series. These newer frameworks could have resulted in higher benchmark results for me.

 

I recommend you download the Mac mini 10.6.4 specific update and extract the frameworks from that using Pacifist, and install them yourself.

 

/System/Library/OpenCL.framework and /System/Library/OpenGL.framework

just these two?

 

Thanks

[mitch_de]

Would be nice if someone with ATI 5xxx could test the new smalluxGPU(OpenCL) instances 130 Mill triangles Scene.

That huge scene may not run already (Nvdia does, ATI Mac unknown, ATI Win not).

Please post interactive mode instances scene gpu only - if it worked.

Speed (in the interactive mode) is shown as xyKile samples/Sec - my 8800GTX gets around 27k samples/sec. in gpu only.

 

 

Also benchmark times for sponza and luxball uktrahigh are welcome.

 

Here the prefinal (before macupdate upload with your reference times)

 

Now 1.7.1 PreFinal: http://rapidshare.com/files/409202831/smal...PU_171Final.zip

 

- removed OpenCL Code Type selector (0/1) from Benchmark. Always use now Type 1.

- added SPPM Method Type - special raytracing method in OpenCL. Glass / lights look mostly better, but more cpu usage

- added intances 130 Mill Scene (fixed res / special OpenCL Code = no res+no Type selector for that scene)

Instances 130 Million (triangles) may NOT work on ATI OS X GPUs (drivers will get update soon).

At least on Windows the ATI OpenCL drivers fail, Nvidia drivers can run that scene on Win/OS X

Also huge RAM usage, therefore 2,5GB+ RAM needed to perform fast. I would NOT run that GPU only on lowend GPUs.

- added again CPU cores count selector (2 or 4 cores) for interactive scenes

[Oxtie]

Sooo...I see this news about official drivers coming out for new Mac Pro (5870 + 5770) 

 

 

 

 

Question: Are these drivers going to work on 5970 (which as 2x58xx chip)?

[lenn4rd]

I started the benchmark incase you need another result for comparison. My system specs are listed below.

 

I tried the smallux benchmark as well, but the slg process crashes.

 

Update: I also ran the Luxball and Sponza Ultrahigh benchmarks and attached a screenshot. I got 33.8 and 38.2 sec, respectively.

 

Kind regards,

Lennard

 

[sch8mid]

I started the benchmark incase you need another result for comparison. My system specs are listed below.

 

I tried the smallux benchmark as well, but the slg process crashes.

 

Update: I also ran the Luxball and Sponza Ultrahigh benchmarks and attached a screenshot. I got 33.8 and 38.2 sec, respectively.

 

Kind regards,

Lennard

 

 

Thanx Lennard

 

Smalluxbenches OpenCL Pixelfilterbench is a no go for ati (appl. closes)

 

But your VaporX 5770 performed best so far

but way off Morpheus performance

it would be interesting to see his OpenCL benches

 

as a reference for you

 

Sponza bench: 44 sec.

[lenn4rd]

You're welcome! Glad I could be of help because I always enjoy a comparison and being able to judge a system correctly.

 

I'm wondering how much of a difference the newer Mac mini frameworks make. According to the screenshot posted they should improve performance by factor of 6 if only the graphics adaptor is measured (ran OpenGL Viewer for the first time). I'll try installing the frameworks over the next days—after backing up that is.

[sch8mid]

here are the results with win 7

 

though performance in non - benchmarkmode is slightly better with win 7 = 3200 fr

 

the

 

multiple cubes on win 7 = benchmark test on Mac

performs just the same

if not worse

 

 

Now this is over my head

where they got their 1000 fr in benchmarkmode

 

and just a mac- 5670 benchmark was released with opengl 3.0 in green

 

???

[mitch_de]

I started the benchmark incase you need another result for comparison. My system specs are listed below.

I tried the smallux benchmark as well, but the slg process crashes.

Update: I also ran the Luxball and Sponza Ultrahigh benchmarks and attached a screenshot. I got 33.8 and 38.2 sec, respectively.

 

Kind regards,

Lennard

 

Thanks - very fast benchmark times ! Will be used as reference in soon uploaded macupdate version. )

ATI OpenCL drivers (5xxx gpus only) has problems in OpenCL Pixelfilterbench (its like CoreImage usage). ATI 48xx or any NV can run that. Also unclear if ATI can run the new Instances scene.

Have you tried to run instances scene (select instances 130 Mill scene in the interactivemode).

[M0rpheus]

/System/Library/OpenCL.framework and /System/Library/OpenGL.framework

just these two?

 

Thanks

 

Yeah those 2 frameworks in the /System/Library/Frameworks folder. You can either extract them with Pacifist and let it install them (possibly get errors due to them being in use, you might have to boot in safe mode to get just VESA and avoid conflicts) of you could try manually putting in the updated files and repairing permissions.

 

I would say that now the 5870 and 5770 etc... will be officially supported in the new Mac Pro and iMac updates, that newer/updated drivers will come with their special custom 10.6.4 builds or in the 10.6.5 beta seeds that will be starting soon.

[lenn4rd]

Thanks - very fast benchmark times ! Will be used as reference in soon uploaded macupdate version. )

 

Cool!

 

Have you tried to run instances scene (select instances 130 Mill scene in the interactivemode).

 

Yes, as written above, the process crashes. The console log within smalllux shows command run:

 

bash: line 1:   385 Segmentation fault	  ./slg -o scenes/instances/render-fast.cfg -D opencl.nativethread.count 0 -D accelerator.type 3 -D opencl.renderthread.count 4 -D image.width 768 -D image.height 768

 

Let me know if you need to know anything else.

[mitch_de]

OK, that slg crash comes from the ATI OpenCL driver (excat OpenCL framework). Also ATI Windows driver crash in that scene. Nvidia seems to be a bit more further in this case of OpenCL.

 

I see that Apple updated the OpenCL Framework in their iMac Mid 2010 update (AMDxxx.dlib). Perhaps soon available for all.

[sch8mid]

OK, that slg crash comes from the ATI OpenCL driver (excat OpenCL framework). Also ATI Windows driver crash in that scene. Nvidia seems to be a bit more further in this case of OpenCL.

 

I see that Apple updated the OpenCL Framework in their iMac Mid 2010 update (AMDxxx.dlib). Perhaps soon available for all.

 

it is available for dl

[mitch_de]

it is available for dl

Yep, but be careful using pacifist to try OpenCL Framework (is updated) or other ATI drivers.

At least an backup (bootable) for such tests is very usefull. Also ATI 4xxxx drivers and OpenGL Framework got updated.

[sch8mid]

someone tried installing the mac mini frameworks ?

here is un upload from netkas :

 

http://forums.macrumors.com/showpost.php?p...p;postcount=152

 

I would do it but am in the middle of an aperture project for three days

and am a lttle bit afraid

though with a timemachine backup this should be no problem at all ...

 

Andreas , is the reported 725 Mhz by smallux the correct speed of the 5770 ?

cause I flashed a custom bios (20% fan speed and 300/800 for H.264 acceleration)

 

B.t.w. in the new mini frameworks is an H.264 decoder init

Could not believe that apple will sort it out after 3 years lethargie

to get GPU acceleration of H.264 ....

 

though for now Apple H.264 QuickTime trailers in 1080p need 50 % CPU on a QuadCore

 

(Win 7 has 3 % CPU load for years (decoding H.264 completely on modern GPUS) b.t.w...since the time of a HD 2600 ....brrrr....)

 

Apple the multimedia company

 

best

armin

 

 

best

m

[le jean]

I just installed the framework extracted from the mac mini Update, yet I can not see any progress in Glview scores.

[sch8mid]

Yepp

same here

did a Timemachine backup

...installed opengl framework with Pacifist -as well as the new framebuffers and the quartz frameworks)

though do not know if all went well ..but had no KP

restarted without any problems

but

just a little lower as your score

 

under 200 fr.

 

my 3850 scored 1400 in Benchmarkmode 4 years ago

 

Thinks we have to wait for the champs Netkas and Morpheus

 

best

armin

[Mauley]

Hey guys, i would like to contribute as well. Here are my benchmark results.