Unigine Heaven 4.0 Benchmark! Updated 14.02.2013 + BIG GPU test

[mitch_de]

Yep, at least OpenGLViewer (Test/Benchmark) results are critical for VSYNC :

Its not DP only - also in 10.6 times , as i changed Monitor 120 Hz to TFT 60 Hz i got , as same resolution (fullscreen) + same cpu/gpu much lower FPS.

Its because OGLViewer will not switch off VSYNC and so the lower 60 Hz limits FPS against 120 Hz or (even better) windowed mode FPS.

As i said, only in windowed mode its garantied that VSYNC is OFF (because not used in windowed mode).

 

Thanks for testing Unigine fullscreen vs windowed mode = same FPS. SO we can be sure that Unigine didnt use VSYC.

 

EDIT: I tested OGLViewer 4.x (after long pause) again with 10.7.4 dev. Funny, now windowed mode (even a bit less res) is much slower than fullscreen.

 

And i seen that you can test 3.1, 3.2 OpenGL bench if you use CORE and not COMPATIBILITY (in the TEST Menue selectable).

Seems that default mode checks if all functions of 3.x / 4.0 are implemented / available by HW and if not (even only 1 of 10) isnt it disables the newer 3.1...4.2 Tests.

I now got running 3.1 and 3.2 Tests using CORE with my 9600 GT. Seems that for that tests the partly implemented 3.x features are enough to bench at least 3.1 + 3.2 OpenGL.

[oSxFr33k]

@mitch_de,

 

And to compound the issue I noticed if I move the window around in OpenGLViewer my scores go way up from the 100's to several thousand, Try this see if you get the same result. Just move the window around immediately when the program starts and watch what happens.

[mitch_de]

Does this, window moving rise FPS from hundrets to tousands also needed if you use Multisampling *8 ? (produce much more GPU load )

For sure the FPS are lower but maybe fast at max. possible without moving window.

If Multisampling *8 (OpenGLviewer TEST windowed) has NOT same effect : moving window rises FPS much (many times) than i guess the AGPM works but needs too much GPU load to rise the gpu MHz.

Also you can try to use KING insted of CUBE, also produce some more gpu load.

CUBE is very easy - perhaps to easy for fast gpus

 

EDIT: If i move window my FPS LOWERS not rises, what should be normal.

[oSxFr33k]

Does this, window moving rise FPS from hundrets to tousands also needed if you use Multisampling *8 ? (produce much more GPU load )

For sure the FPS are lower but maybe fast at max. possible without moving window.

If Multisampling *8 (OpenGLviewer TEST windowed) has NOT same effect : moving window rises FPS much (many times) than i guess the AGPM works but needs too much GPU load to rise the gpu MHz.

Also you can try to use KING insted of CUBE, also produce some more gpu load.

CUBE is very easy - perhaps to easy for fast gpus

 

EDIT: If i move window my FPS LOWERS not rises, what should be normal.

 

Interesting it should rise I beleive. I went to a Mac Store and tried this on the latest Macbook Pro sandybridge with the latest AMD graphics card with the same results as my Graphics card. When I started it I was around 600-700FPS and moved it over a bit and all of a sudden went to 2500-3500FPS.

 

I have not changed any settings in OpenGLViewer. I have to try some of the settings you suggested to see when mooving the window if this actually lowers the FPS count with my system.

[mitch_de]

I would temporary remove the AGPM (System/Library/Extensions/AppleGraphicsPowerManagement.kext).

Copy it to deskop, delete whole system cache /System/Library/Caches/com.apple.kext.caches (will be rebuild automatic)

Reboot and try again if that 600 > 3000 also happens or not. If also, its not an AGPM problem.

You can use kextwizard to easy install AppleGraphicsPowerManagement.kext again + reboot.

[oSxFr33k]

I would temporary remove the AGPM (System/Library/Extensions/AppleGraphicsPowerManagement.kext).

Copy it to deskop, delete whole system cache /System/Library/Caches/com.apple.kext.caches (will be rebuild automatic)

Reboot and try again if that 600 > 3000 also happens or not. If also, its not an AGPM problem.

You can use kextwizard to easy install AppleGraphicsPowerManagement.kext again + reboot.

 

Just got home and tried without AGPM no difference at all. I don't think you have to actually only move the window but clicking in the window increases the FPS count as well.

 

WIth King around 2000 FPS with CUBE around 2500 FPS. Windowed or FUll screen minimal difference. Changing Multisampling from default 2 to 8 no difference maybe a couple hundred FPS. If I do not move the window or click in the window the FPS average around 600 FPS.

 

Must be a bug in the OpenGLViewer Software maybe?

 

If I change Compatibility mode to CORE the program just stops after the first couple tests. I tried the default SGI and Nvidia no difference in scores

[mitch_de]

Yep, OpenGLViewer may have an bug handling some tests and maybe its only for some driver / gpu combinations. Dont worry, simple remember that problem and dont try this TEST with your GPU if other tests are more valide.

The CORE Test Mode , which tries to use uncomplete (Drivers for gpu type hasnt 100% impelemnted of that) OpenGL 3.1 features may depend also on gpu type / drivers.

For me CORE (insted of default compatibility) with NV 9600 GT runs at least OpenGL 3.1 and 3.2 without errors. Result around 410 FPS KING.

PS: OGLV is Version 4.0.5, maybe you check for update. For me same results with .05 as with .02.

 

Screenshoot shows CORE Test OpenGL 3.2

benching 3.1+3.2 then it stops , because in CORE Mode OpenGL 3.3+ havent all functions implemented for my GPU (less than 100%)

[dMopp]

Hmm, my 470 GTX has only OpenGL 2.1 Compatibility ? WTF ?

[nyolc8]

Sapphire Radeon HD 4830 512Mb > 800x600 | 2XAA | window | Shader:High | Ani:4X > 1237

Sapphire Radeon HD 4830 512Mb > 1920x1080 | 2XAA | fullscreen | Shader:High | Ani:4X > 464

 

 

Not the best, but it does the job for me (for now)

[mitch_de]

Yep - and for gaming (not benching) in that high res (1900x1080) i think you would get near playable min FPS if you would use NOFSAA with that engine.

[Wayang-NT]

HD6870 - DP3 (AMDRadeonAccelerator removed)

 

[mitch_de]

Great . Also an good idea to bench with 4*FSAA on very fast gpus (even more less cpu speed dependent results than with 2*FSAA).

But please add also the 2*FSSA results to compare with others - lowend-midrange gpus (in conclusion with 512 MB VRAM or less) cant handle 4*FSAA good in higher res and so 2*FSAA is an common bench environment for all.

[Wayang-NT]

800x600 vs 1920x1080 2xFSAA :

 

 

 

800x600 no FSAA :

 

[mitch_de]

Thanks. Beside resolution, FSAA is the parameter which most FPS effect.

Interesting is the big diff in spped windowed vs fullscreen at same resolution with your AMD card + ML DP3.

I get same bench results in 800x600 windowed vs 800x600 fullscreen in 10.7.4 E52 (Nvidia).

Would be fine if others with AMD/NVIDIA check (again) windowed vs fullscreen 800x600 (no fssa for more diff) also.

I bet that only AMD drivers have this huge diff windowed vs fullscreen. Perhaps some AMD driver problems in windowed mode.

Or an ML DP2/DP3 problem. For my knowledge Nvidia cards cant run Unique in windowed mode in DP3 - freeze.

[RobertX]

just because...on 10.7.4 now

[npwski]

OS 10.7.4 with Nvidia retail driver v.270.00.00f06 ("Quadro driver"),

GTX580 MSI "Lightning" @920Mhz

1920x1080, FSAA=4, all other at max

 

And no, I don't agree with you, Mitch: I think that "00f06 Quadro driver" is best seen driver for Fermi cards. Because no need for AGPM edit, because of very good performance in games, because of stability et ct.

[mitch_de]

I said ( in thread 10.7.4 out, news) that the Nvidia drivers compared to Apple drivers may only help for Quadro (=Fermi) users. Non fermi users better use Apple drivers!

[RobertX]

I said ( in thread 10.7.4 out, news) that the Nvidia drivers compared to Apple drivers may only help for Quadro (=Fermi) users. Non fermi users better use Apple drivers!

 

...well...apple 10.7.4 drivers work best for me...GT520(a fermi...but a bit of a {censored} i think)

[npwski]

I have a "little {censored}" too = Geforce GT430, it based on GF108 chip with 96 cores and overclocked @822Mhz with passive cooling.

 

OS 10.7.4 with Nvidia retail driver v.270.00.00f06 ("Quadro driver")

800x600 windowed, FSAA=2, Shaders=High, Aniso=4

1920x1080 fullscreen, FSAA=4, Shaders=High, Aniso=16

 

IMHO "Quadro" driver demonstrate good performance with this fermi card.

[badaxe2]

[Wayang-NT]

HD 6870 - OS X Mountain Lion build 12A269 .....

 

 

 

 

800*600 windowed:

 

 

 

800*600 fullscreen:

 

 

 

 

1920*1080 fullscreen :

 

[Squidderlee]

[Colask8]

Felt like sharing... By comparing with other GTS450 seems like fine... Still gonna compare on windows. It really makes me sad that i cant play wow on full settings with opengl, and gll not working on 10.8... And Dota 2 is freezing like hell 20fps max 2 fps min...

[mitch_de]

NEW 4.0 Version Results:

 

Preset BASIC (medium, 1280x720, windowed, AA*2) Nvidia 9600 GT, 10.8.3 D65

 

I see that AA*2 and medium already needs 90% of the 512 MB VRAM. Higher quality + Higher res + more AA will reduce speed (short fps downs) with 512 MB only.

 

 

 

 

Same but HIGH:

[k3nny]

Version 4.0 produces following images - first one with AA, second without AA. This is not the same scene.

Currently running 12D65.

 

EDIT:

 

AA was not fixed with 12D68 and it seems tessellation doesn't work either.

My benchmark results: Left Basic w/o AA, right Ultra w/o AA.

 

EDIT2:

 

Benchmark result on Basic w/o AA in Windows 8 x64 with OpenGL:

FPS: 132.6

Score: 3340

Min FPS: 9.9

Max FPS: 255.0

Edited February 15, 2013 by k3nny