Jump to content

Unigine Heaven 4.0 Benchmark! Updated 14.02.2013 + BIG GPU test


mitch_de
 Share

128 posts in this topic

Recommended Posts

Hi.

i have great news about an updated (from 3.0) very good OpenGL Benchmark!

http://unigine.com/p...-benchmark-4.0/ Infos

 

Whats new (since last 3.0):

 

What's New in Heaven 4.0?

  • Benchmarking presets for convenient comparison of results
  • GPU temperature and clock monitoring
  • Drastic improvement of SSDO
  • Stars at nighttime
  • Improvement of lens flares
  • New, improved version of UNIGINE Engine under the hood
  • Detection of multiple GPUs
  • Anti-aliasing support on Mac OS X
  • Enhancement of automation scripts in Pro version
  • New Advanced edition (see details below)
  • Russian and Chinese localization

http://unigine.com/p...eaven/download/

 

 

This benchmark benches GPU good and long under 26 scenes (takes a few minutes)

My screenshoots shows 9600 GT, Preset BASIC (medium, 1280x720, windowed, AA*2)

 

 

Because the different possible settings makes it hard to compare already benched GPUs,

PLEASE

post at least also an Preset BASIC bench with

 

 

Preset BASIC (medium, 1280x720, windowed, AA*2)

to compare other users with smaller screens (not able to run 1600x1020 or 1900x1200 res).

Users with more screen res ( i am limited to 1440 :( ) and faster gpu can also post Preset Extreme result :)

 

 

EDIT Results from 4.0 will be NOT comparable to 3.0 results!.

 

 

RESULTS 4:

Sapphire Radeon HD 7970, BASIC (no AA) = 98,5 fps, BASIC = 91 ( but known artifacts)

MacPro, Nvidia GTX 680 , OS X 10.8.2 : BASIC=88 fps, EXTREME=57fps (rob from barefeats, EXTREME takes around 850 MB VRAM)

EVGA GTX 660 2GB, 10.8.2, BASIC = 65 fps, BASIC with Ultra Quality = 49 fps

Asus GTX275. BASIC = 34,5 fps

Nvidia 9600 GT, 10.8.3 D68: BASIC=21fps

GeForce GT 430, BASIC = 16 fps

 

 

EDIT:

I added 2 new and BIG gpu tests. COD + PREY in highres (1900x1080) tests over 100 gpus!

Good to take a look if you want to buy a "new" gpu.

Often new gpus are much worse (in fps) than oldies like 9600 GT / 8800 GT / 9800 GT or AMD 5770.

Example: GT 430 or GT 210/220/230/520 - good in power consumption but worse compared to above listed old+cheap (used) gpus.

Only avoid old 8800 GTX / GTS and 9800 GTX - fast but huge power consumption. Also AMD 4870/50 takes huge power consumption.

PCG (german computer games mag) march 2013 COD+PREY

detailed (cut out) screenshoot of one test - highlighted well known gpus

Bildschirmfoto 2013-02-17 um 12.52.54.jpg

Bildschirmfoto 2012-03-09 um 23.48.25.jpg

Bildschirmfoto 2012-03-09 um 23.55.24.jpg

PCG313_PREY_highres.pdf

PCG313_COD_highres.pdf

Link to comment
Share on other sites

Ani filtering = makes textures more clear/sharper = has less fps effect

Anti aliasing = less pixelation (steps) on lines = has much fps effect, much more gpu work + VRAM usage than OFF

Try to rerun with Anti aliasing = 2* and compare to Anti aliasing = OFF

Anithros Filtering 4*, 8*, 16* has speed effect but much less than Anti-Aliasing (multisampling, FSAA). The second takes much more VRAM+GPU time than off. The first not more VRAM and only a little more gpu time

ani filtering off vs 4* = near no fps lost BUT fssa off vs 4* at least 30% fps lost

To get more gpu related (than cpu related) results i would use Anti-Alisasing *2 , Shaders=high and an higher res than i used like 1900x1080+ .

Lower res in window mode , like 640x480 or 800x600 , Shader=medium, FSAA OFF Anitros OFF is good to test driver version diff speeds.

Happy benching - good that XBENCH & Co "benchmarks" are now complete outdated ;)

 

EDIT: I tested with 10.8 DP1 also. Used the 800x600 windowed, no FSAA mode - good for driver version compare because less gpu hw speed bound as FSAA*2 and high res.

Same (

Link to comment
Share on other sites

Great review.

But i am not sure if its really good to use such high FSAA *8. Normally no one uses that in games because it takes lot of max FPS.

I would use 4* FSAA as maximum - otherwise the different FSAA algorithm speeds of the gpus getting to much in focus and not the main OpenGL/Shader speed.

 

XFX 5870 1920x1200x32 shaders medium, AA off, OpenGL, Full Screen, Mac OS X 10.6.8 Score: 1298

XFX 5870 1920x1200x32 shaders medium, AA off, DX11, Full Screen, Windows 7 Score: 1528

Thanks.

Would be interesting to compare Win OpenGL vs OS X OpenGL too (even more interesting).

For my knowledge AMD OpenGL Win drivers arent very good. Some GTX 570 got little better (

With AMD i guess the Win7 OpenGL result will be less than OS X OpenGL.

Link to comment
Share on other sites

Thanks. As i said before i would use FSSA * 4 (not *8) as max. Because otherwise you measure how fast the gpu can do the FSAA / how fast is the algorithm of FSAA and less the real Game / OpenGL / Shader performance.

Only if the diff between *4 and *8 is less (less than 10%) the gpu isnt limited by FSAA and the *8 result make sense.

So check the diff between *4 and *8 and if >> 10% please post also *4.

 

EDIT: I compared VRAM usage without / with FSAA 1280x1024 (ca. 1400x900):

Without FSAA - all other things high, my 512 MB is used 70-75%

With 2* FSAA 77 -85%, With 4* 88 -93%

Because VRAM never runs out of mem by swapping VRAM PCI 2.0 Bus RAM at least at 90%+ used the FPS break down by that very slow

VRAM swappings which happen frequently.

So beside the gpu loads by FSAA 4* and more it may happen that your VRAM gets swapped and then the measurement also gets bad.

Only if you have 768MB+ VRAM you will not get / very less VRAM swappings - at least not if you use

Link to comment
Share on other sites

Thanks. I would use shaders= high because often+massive used in games >= 2009.

Also interesting is FSAA no vs FSAA*2 (or *4). The faster the gpu the less is the slowdown by using FSAA against FSAA no.

Link to comment
Share on other sites

Great review.

But i am not sure if its really good to use such high FSAA *8. Normally no one uses that in games because it takes lot of max FPS.

I would use 4* FSAA as maximum - otherwise the different FSAA algorithm speeds of the gpus getting to much in focus and not the main OpenGL/Shader speed.

 

 

Thanks.

Would be interesting to compare Win OpenGL vs OS X OpenGL too (even more interesting).

For my knowledge AMD OpenGL Win drivers arent very good. Some GTX 570 got little better (<= 3%) OpenGL speed in OS X against Win with that bench.

With AMD i guess the Win7 OpenGL result will be less than OS X OpenGL.

 

Earlier I made a wrong description. Here is the corrected descriptions and results:

 

XFX 5870 + i7-870 1920x1200x32, shaders medium, Anisotropy x4, OpenGL, Full Screen, Mac OS X 10.6.8 x64, Score: 1298

XFX 5870 + i7-870 1920x1200x32, shaders medium, Anisotropy x4, DX11, Full Screen, Windows 7 x64, Score: 1528

XFX 5870 + i7-870 1920x1200x32, shaders medium, Anisotropy x4, OpenGL, Full Screen, Windows 7 x64, Score: 1237

Link to comment
Share on other sites

As my thinking: Only DirectX11 is faster than OpenGL OS X , not the OpenGL Win.

Second guess: I believe that the diff in OpenGL speed OS X vs WIN rises for OS X winning even more , if you use an bench mode with much less gpu load, so OpenGL driver speed differences can be seen much better. Like 800x600 windowed, no fsaa

Link to comment
Share on other sites

I am running 10.7.2 with a GTX 460 GLH ver. 1024MB RAM too but my results are low even with res. 800x600 something is wrong with injector

 

 

146uiu.png

 

 

Although Win 7 X64 performed as should be with this card

 

flw88.jpg

 

 

Maybe somebody suggests something to workout my problem with Lion+GTX460 GLH 1GB Card

Link to comment
Share on other sites

GTX 460 OS X results indeed much to low. I think AGPM (Graphicspowermanagement) problem - gpu does always run in idle/low MHz stepping.

In Nvidia GPU thread there is already an AGPM thread (in first threads pinned)

  • Like 1
Link to comment
Share on other sites

Aniso Filter 4* > 8* or 16* hasnt much negative speed effect on faster gpus. So diff between 4* and 8* / 16* is little. (Ani makes textures more clear / sharp)

The FSAA X* has much, much more effect on speed and i would not use > 4* to bench the gpu, otherwise you bench more FSAA algorithm speed than overall OpenGL speed.

But FSAA * 2 ( at least on res

Link to comment
Share on other sites

 Share

×
×
  • Create New...