Jump to content

Nvidia Fermi GTX 4xx, GTX2xx (+ others) Users for Benchmark WANTED

mitch_de

By mitch_de,
in nVidia

 Share
62 posts in this topic

Recommended Posts

Tewan
Tewan
Posted
Posted

*** My CPU results are invalid ***

My Core i5 was running underclocked at 2ghz. I will test it in a minute.

To add, my GTX 280 results may be skewed as I have a defect card... maybe not though. (Runs under load at 105c)

 

Anyways, I'll post the results of the all the tests over with this ATI Radeon HD 4870 1gb and the correctly clocked Core i5-750.

 

Label: New (+/- diff old)

 

FILTER NONE: 1150.53M Samples/sec (+274.53)

FILTER PREVIEW: 223.94M Samples/sec (-48.74)

FILTER GAUSSIAN: 107.56M Samples/sec (-34.86)

 

CPU Only

Mid: 27.1 sec (-14.2)

High: 52.9 sec (-30.5)

Ultra: 52.9 sec (-21.5)

 

Hybrid

Mid: 8.4 sec (+1.4)

High: 18.8 sec (+4.1)

Ultra: 29.8 sec (-10.6)

 

GPU Only

Mid: 12.3 sec (+1.1)

High: 21.6 sec (+5.2)

Ultra: 53.7 sec (+14)

 

This is NOT with the flashed ATI bios... I have to boot into Windows as well.

Link to comment
Share on other sites
mitch_de
mitch_de
Posted
  • Author
Posted

Thanks ! In this case you see that the power of the OpenCL driver (better: OpenCL > native gpu code runtime compiler) has big effect how fast OpenCl works - beside the GigaFlops Power the hardware (shader units) can do.

The ATI 48xx cards are on HW and OpenCL normal code (smallluxGPU raytracing for example) equal than GTX 280, much faster than GTX 260.

But impelented the OpenCL pixeldevice (working with pics = OpenCL pixelbench) code wore than Nvidia. ATI starts to implement OpenCL pixeldevice code very late (on Win/OS X) : 10.6.3 Nvidia impelented that already in 10.6.0 and maybe already optimized a lot .

Link to comment
Share on other sites
  • 2 weeks later...
Hassan Rai
Hassan Rai
Posted
Posted

Hello Mitch... i have a question... i have been reading the benchmark reports over in both threads (nVidia and ATi) and although i appreciate the level of complexity you are adding to your utility... i have a burning question... which is better????

 

I'm a little confused and cannot figure this out right now....

 

But allow me to ask a simpler question: for someone considering to buy either 4870 or a GTX 260 right now, what is the recommendation based on the benchmarks? (because i dont understand the benchmarks as well as you do). If one does the typical things like using the hackintosh for photoshop, video and the few games that are available... which card is better?

 

Lastly, some people have just done the unthinkable and got their Fermi cards working in OSX!!!! Yes it's true...

http://www.insanelymac.com/forum/index.php...p;#entry1494563

 

I can't wait to see their benchmarks... though there are very few 4870 cards left in the market (where i live) and if i don't buy one soon, they will no longer be available. I think it will be tough to find a GTX260 now anyways... but please do try to answer for me in simple words.

 

Thanks!

Link to comment
Share on other sites
mitch_de
mitch_de
Posted
  • Author
Posted

Normally using Nvidia cards is much more easy than ATI - less or no patching needed / no need to use older drivers / ...!!!

That because the driver set of NVIDIA is much less complex than ATI driver set because: less driver parts / less "belongings" between that parts / less problems with the screen output (no dongles needed on DVI/VGA).

 

So the ATI vs NVIDIA is also an kind of easy vs nightmare decision beside the speed decision which to buy.

The GTX 2xxx run well also in OpenCL. Only the older gpus of 9xxx or 8xxx run much slower than ATI 4xxx.

Fermi 4xxx also should run much faster than 2xx or ATI 4xxx. If Fermi drivers really work sometimes i would buy Nvidia Fermi or older GTX 2xx card. The ATI problems will be neber easy get fixed because of their much more complex driver problems in case of using non Mac gpus.

Nvidia Mac likes much more hackintosh than ATI Mac :)

 

smallluxgpu updated to V170 !

http://www.macupdate.com/info.php/id/33632/smallluxgpu

Link to comment
Share on other sites
mitch_de
mitch_de
Posted
  • Author
Posted

I would ask in the main nvidia thread / or ask in similar questions already exits which card can be the best buy for you.

Also your main doings may be helpfull: Gaming, OpenCL/CUDA, normal QE/ CoreImage

Link to comment
Share on other sites
meroy
meroy
Posted
Posted

Core i7 870 @ 3.68Ghz + EVGA GTX 295 Co-op Edition (Single PCB) @ 684/1512/1180 MHz

 

-- Utilizing both GPUs

GPU Only Midrange: 5.3sec

GPU Only Highend: 6.5sec

GPU Only Ultra: 15.7sec

 

-- Only one GPU is used for the Pixelfilter test

FILTER NONE: 855.93M

FILTER PREVIEW: 292.16M

FILTER GAUSSIAN: 177.43M

 

-- Utilizing both GPUs (Luxball glass Scene)

GPU1: Rays/sec 5714K

GPU2: Rays/sec 5715K

Avg. Rays/sec 11428K

post-217574-1277654763_thumb.png

post-217574-1277654786_thumb.png

post-217574-1277654799_thumb.png

post-217574-1277654812_thumb.png

post-217574-1277654826_thumb.png

Link to comment
Share on other sites
mitch_de
mitch_de
Posted
  • Author
Posted

WOW - extreme FAST with 2 GPUs !

Single ATI 48xx cant follow :D

 

I see (in the help screen) that i build the 1.7.0 with slg 1.5Beta4 - will be update to 1.7.1 as an fix .

sppm (switch Path/sppm mode in interactive scene to see the diff) and some other optimisations , enabled in 1.7.0 cant run .

I uploaded fixed 1.7.0 (V2) to macupdate.com !

Link to comment
Share on other sites
  • 3 weeks later...
mitch_de
mitch_de
Posted
  • Author
Posted

Soon there will be an update to V171, which has slg V1.62Dev inside and also some fixes.

I removed High gpu only bench and added UltraHigh Sponza gpu only.

Would be fine if you check new Sponza Benchmark.

I uploaded the V171 to RS for getting some reference benchmark times. No major changes in known Luxball UltraHigh benchmark .

http://rapidshare.com/files/407420780/smallluxGPU_V171.zip

Bildschirmfoto_2010_07_17_um_10.59.14.jpg

Link to comment
Share on other sites
  • 2 weeks later...
mitch_de
mitch_de
Posted
  • Author
Posted

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.

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

Also huge RAM usage, therefore 3GB 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

Bildschirmfoto_2010_07_26_um_09.18.28.jpg

Link to comment
Share on other sites
quattrofx
quattrofx
Posted
Posted

Test Part II:

 

GALAXIES 32K_V2

NVIDIA INNO3D GeForce GTX 285 1GB

 

_________SIM:___________________ GIGAFLOPS:

Vector Single Core CPU:_________________12

Vector Multi Core CPU:__________________46

GeForce GTX 285:_____________________ 329

Hybrid Multi Core CPU+GPU:______________80

 

OpenGL 3.1:

post-292451-1280391616_thumb.png

post-292451-1280391623_thumb.jpg

 

For comparison:

ATI Sapphire Radeon HD4870 1GB

 

OpenGL:

post-292451-1280392313_thumb.png

 

OpenCL:

post-292451-1280392337_thumb.jpg

post-292451-1280392350_thumb.jpg

Link to comment
Share on other sites
mitch_de
mitch_de
Posted
  • Author
Posted

Please try the more new V171(V2). Sponza Ultrahigh and Luxball Ultrahigh - but should perform near same as with that older V170 version.

Also some GUI cleaning (removed OpenCL Type setting 0/1 - now alwys uses Type 1 which run normally faster)

Even more interesting is how fast the GTX 285 perform the extreme huge scene instances. 130 million triangles !!

I get around 27 kSamples/sec (gpu only, 8800GTX ). instances scene can be selected (only) in the interactive mode and shows the informations (CPU/GPU performance) in the bottom line as kSamples/sec.

 

DL :

http://rapidshare.com/files/409790525/smallluxGPU_V171_2.zip

Link to comment
Share on other sites
quattrofx
quattrofx
Posted
Posted

post-292451-1280439335_thumb.png post-292451-1280439358_thumb.png

Sponza Hybride 4cpu + gpu-------------Sponza GPU

post-292451-1280439416_thumb.png post-292451-1280439440_thumb.png

Luxball Chrome Hybride--------------Luxball Chrome GPU

post-292451-1280439472_thumb.png post-292451-1280439496_thumb.png

130 Mille Hybride----------------------130 Mile GPU

post-292451-1280439511_thumb.png--------post-292451-1280439520_thumb.png

 

Have a nice day! :(

____QuattroFX____

Link to comment
Share on other sites
mitch_de
mitch_de
Posted
  • Author
Posted

Great ! You get much more speed in instances : 146 vs 27 k Samples/sec - me, 8800 GTX !

Diff to 8800GTX is much bigger than in Sponza or Luxball benchmark (30 vs 80 sec) . I will add that instances also as benchmark (until ATI OpenCL bug fixed with ATI information that it may fail).

Also the other Benchspeeds (Sponza, Luxball) are very good - at least ATI 5770 cant bet them.

Link to comment
Share on other sites
quattrofx
quattrofx
Posted
Posted
Great ! You get much more speed in instances : 146 vs 27 k Samples/sec - me, 8800 GTX !

Diff to 8800GTX is much bigger than in Sponza or Luxball benchmark (30 vs 80 sec) . I will add that instances also as benchmark (until ATI OpenCL bug fixed with ATI information that it may fail).

Also the other Benchspeeds (Sponza, Luxball) are very good - at least ATI 5770 cant bet them.

 

Hi Mitch_de,

 

I will be on holiday next 3 weeks, so cannot make some more tests.

 

I am interested in to see some GTX 470 - 480 tests for comparison to 285, because they still use drivers from 200,

if they can come to similar results? Do you know something about it? Do they support OpenCL at the moment?

 

Grüße

QuattroFX

Link to comment
Share on other sites
Marucins
Marucins
Posted
Posted

 

10.6.0 <---> 10.6.4

01_13.jpg <---> 001_11.jpg

 

 

10.6.0 <---> 10.6.4

02_13.jpg <---> 002_13.jpg

 

 

10.6.0 <---> 10.6.4

03_28.jpg <---> 003_6.jpg

 

10.6.0:

NVDAResman.kext, NVDANV40Hal.kext, NVDANV50Hal.kext, GeForceVADriver.bundle, GeForce.kext, GeForce8xxxGLDriver.bundle, GeForce7xxxGLDriver.bundle, GeForceGA.plugin

http://dl.dropbox.com/u/6372548/OSX86/nVidia%2010.6.0.zip

Link to comment
Share on other sites
mitch_de
mitch_de
Posted
  • Author
Posted

"I am interested in to see some GTX 470 - 480 tests for comparison to 285, because they still use drivers from 200,

if they can come to similar results? Do you know something about it? Do they support OpenCL at the moment? "

 

No, Fermi GPUs fail in OpenCL - missing Apple Support. Only Opengl works but not as fast as they could do.

ATI 5xxx - OpenCL works , but instances scene fails - maybe the new iMac update will help / 10.6.5 will help.

But as i said before - OpenCL is today (and next 6 months) not really needed until OpenCL is supported by Apps.

Only Starfire uses OpenCL (beside CoreImage) for some speed enhancement on filtering, scaling and denoising/sharpening.

 

DL 55 MB: 171V3 with benchmark instances scene (8800GTX = 150 sec)

http://rapidshare.com/files/410151278/smallluxGPU171_V3.zip

Link to comment
Share on other sites
  • 3 weeks later...
sch8mid
sch8mid
Posted
Posted

10.6.0 <---> 10.6.4

01_13.jpg <---> 001_11.jpg

 

 

10.6.0 <---> 10.6.4

<--->

 

 

 

 

 

 

post-253749-1282203226_thumb.jpg

 

 

 

OpenGl implementation in win 7 is much better

this is a stock ati 5770

with a 2,6 downclocked old Q 9550

 

same config in 10.6.4 with newest drivers available (graphicsupdate 1.0) = 27, 1 fr.

 

people with GTx 285 and i7 usually get > 60 fr . in 11.5 (win7)

Link to comment
Share on other sites
 Share
Go to topic listing
×
×
  • Create New...