21 replies to this topic

[khe91]

I don't do any games, 3D modeling or CUDA. So I don't care about 3D performance.

What I'm looking for is the fastest 2D card. Basically I want all my window changes drawn as fast as possible, same with moving windows.
Still I got a GTX670 because I thought with it's PCIe 3.0 and 256bit GDDR5 memory interface it should be very fast at 2D too.

Didn't find any OSX 2D benchmark software.

For Windows there is 2Dbench.exe from TomsHardware.com.
So I run it under Fusion 5 and got really surprised.

The HD4000 got more then double the 2D performance to the expensive GTX670.
Also tested a GT640, it's close to the GTX670 and but also half the 2D performance of the HD4000.
All test run fullscreen 2560x1600, OSX 10.8.2., Fusion 5.0.1, on a Dell 3011, 30" screen connected with DP.

HD4000 - 1030
GT640 - 510
GTX670 - 473
(yes higher numbers are faster ...)

This results say that any PCIe graphic card is much slower then the integrated HD4000 in 2D.
And all the normal daily stuff, like browsing the web, email, movies etc. is 2D.

Is there any OSX 2D benchmark software to confirm this ?

Attached Files

•  GTX670-2560x1600.tiff   44.81K   1 downloads
•  GT640-2560x1600.tiff   44.16K   0 downloads
•  HD4000-2560x1600.tiff   42.72K   1 downloads

[mitch_de]

You can try my 2D Window Benchmark which is OS X native (Intel only).
It produces 333 windows (with 2D text content) - the more 2D speed the less time it needs.

My 9600GT needs 17.5 seconds. CPU speed doesnt matter much.

In general: doing some low end OpenGL tasks (OS X uses OpenGL for window/text= 2D) an good cpu internal gpu like 4000 can handle that very fast. It has very fast access to cpu / cpu cache for small/less complex tasks like 2D. That can be an advantagte over high end gpus which must use "longer way" over PCI-E slot.

EDIT: updates 2D Benchmark with an more nice window content. Result/Time should be same as first version.

Attached Files

•  Bildschirmfoto 2012-10-03 um 18.15.51.jpg   97.27K   1 downloads
•  2D Benchmark.zip   1.98MB   16 downloads

[robertx]

using the benchmark i get 18.5 sec.

[khe91]

mitch_de, on 03 October 2012 - 03:00 PM, said:

You can try my 2D Window Benchmark which is OS X native (Intel only).
It produces 333 windows (with 2D text content) - the more 2D speed the less time it needs.

My 9600GT needs 17.5 seconds. CPU speed doesnt matter much.

In general: doing some low end OpenGL tasks (OS X uses OpenGL for window/text= 2D) an good cpu internal gpu like 4000 can handle that very fast. It has very fast access to cpu / cpu cache for small/less complex tasks like 2D. That can be an advantagte over high end gpus which must use "longer way" over PCI-E slot.

Thanks for the app.

HD4000  12.6 seconds
GT640 21.7 seconds
GTX670  21.4 seconds

Same result, so no need for any PCIe card.
Now I just need to find a way to use 3 screens on the HD4000 in OSX.

[mitch_de]

Thanks for results!
As i said, because the tasks of 2D (by OS X, OpenGL) are very less complex AND use very less memory, internal new+fast gpu can do that also very fast or even faster than any pci-e gpu card.
Only if some more gpu memory (for textures) needed or more comples tasks are needed, such internal gpu getting weaker than even lowend pci-e gpu cards with own VRAM >=512 MB).

EDIT: updated the 2D Benchmark with more nice window content. Results should be same as before (less than 3% diff).
I also checked gpu load when the 2D benchmark is running. Has very low gpu % load - so speed of gpu doesnt matter.
Most speed diff comes by cpu<>gpu data/memory transfer speed , which is on newest on-cpu gpus very fast.
If that window would have an added OpenGL task running the results Intel HD 4000 vs real gpu card would be different = real gpu faster.

[mitch_de]

Hi, i made an new version which does 2D (OS X uses OpenGL for QE) as before together with some small 3D task showing an textured OpenGL cube within the window.
For me time is same = gpu can handle that without lost of time.
Please report your results compared to the 2D only bench.

Attached Files

•  Bildschirmfoto 2012-10-04 um 10.40.19.jpg   112.76K   0 downloads
•  2D_small3D Benchmark.zip   2.02MB   10 downloads

[khe91]

mitch_de, on 04 October 2012 - 08:48 AM, said:

Hi, i made an new version which does 2D (OS X uses OpenGL for QE) as before together with some small 3D task showing an textured OpenGL cube within the window.
For me time is same = gpu can handle that without lost of time.
Please report your results compared to the 2D only bench.

Same results.
But the app always jumps to the main screen, that makes it a little complicated to test the different screens connected to different graphic cards.
Can you please make it run on the screen it's open and not move it to the main screen if "Start Bench" is pressed.
Thanks und viele Gruesse aus dem Erzgebirge nach Stuttgart ;-)

[mitch_de]

Ok, try that new version (should stay on the display were it was started and not use main/standard display)

Edit: If it also didnt work (windows generated not on the display where the app was started) please run + report
Test_screen results. (Tell me if result main screen size is  really main screen and not the other one - for that it would be helpfull if both screens have different size (pixels).

Attached Files

•  2D_small3D Benchmark2.zip   2.02MB   4 downloads
•  Test_screens.zip   1.16MB   1 downloads

[khe91]

mitch_de, on 04 October 2012 - 11:24 AM, said:

Ok, try that new version (should stay on the display were it was started and not use main/standard display)

Edit: If it also didnt work (windows generated not on the display where the app was started) please run + report
Test_screen results. (Tell me if result main screen size is  really main screen and not the other one - for that it would be helpfull if both screens have different size (pixels).

Don't works, always jumps back to main display.
Test_screens says: You got 4 screens ! (thats correct) Your main screen is 2560x1600. Your second screen is 1600x1200. (what about 3rd and 4th ?)
I have two 2560x1600 and two 1600x1200 screens at the moment connected.

[mitch_de]

Thanks - i only showed first 2 displays. I will make another build soon.

PS: Have you started the bench from the desktop of the display/gpu you want to test?
Perhaps it cant work if you start the bench from other desktop/gpu and move only window to the test screen.

[khe91]

mitch_de, on 04 October 2012 - 02:26 PM, said:

PS: Have you started the bench from the desktop of the display/gpu you want to test?
Perhaps it cant work if you start the bench from other desktop/gpu and move only window to the test screen.

I moved the app to the screen I want to test.
If I press "Start bench" then the app window jumps to the main screen and the benchmark runs always there.
So I have to move the main screen, the menu in the Displays Preferences to the screen I wanna test to start the app there.

[mitch_de]

OK, try next one - be sure to put/move Bench on the desktop of the screen you want to check.
PS: I removed OpenGL part - doesnt matter in speed.

Attached Files

•  2D Benchmark.zip   1.9MB   4 downloads

[khe91]

mitch_de, on 04 October 2012 - 06:36 PM, said:

OK, try next one - be sure to put/move Bench on the desktop of the screen you want to check.
PS: I removed OpenGL part - doesnt matter in speed.

No changes.
I open the app and move the app window to the screen I wanna test.
Press "Start Bench" and it jumps back to main screen.
Also on the main screen it always jumps to the same pixel position close to upper left corner.

[mitch_de]

Ok, i made an new test app which let you select screen on which the window should open.
Starts up on main screen and has radio buttons to select up to 4 screens , listed with each sizes. Press start to open the window in the middle of the selected screen.
PLease check if all displays are listed as radio buttons and after start if the window really opens on the selected screen.

Attached Files

•  Test_screens.zip   1.18MB   2 downloads

[khe91]

mitch_de, on 06 October 2012 - 10:22 AM, said:

PLease check if all displays are listed as radio buttons and after start if the window really opens on the selected screen.

Yes works, just the screen sizes of the 2nd and 3rd are cut (see attached pic).

Attached Files

•  screens.tiff   30.89K   3 downloads

[mitch_de]

Thanks!
I made new 2D Benchmark (based on the radio button display results). It now uses more bigger window sizes (if display is > 1440 width).

Attached Files

•  2D Benchmark 2.zip   1.91MB   5 downloads

[khe91]

mitch_de, on 06 October 2012 - 12:01 PM, said:

I made new 2D Benchmark (based on the radio button display results).

Yes, works now on the different screens.

mitch_de, on 06 October 2012 - 12:01 PM, said:

It now uses more bigger window sizes (if display is > 1440 width).

They are to big, some of the windows are visible on two screens, so the result depends on different graphic cards if the two screens come from different cards.

[mitch_de]

Thanks! I will reduce window sizes back.

[STLVNUB]

On Gt640 22 seconds using PCIE-4 slot. as card uses 2 slots and pcie-1 slot for Airport card is under PCIE-16
Will test hd4000, at least that will run at native speed

[khe91]

mitch_de, on 06 October 2012 - 05:55 PM, said:

Thanks! I will reduce window sizes back.

With the bigger windows the GTX670 (24 sec) wins over the HD4000 (29 sec).