2D Graphics Performance

[RonS]

I'm writing software that requires good 2D performance from a video card. My program relies on BitBlt and PlgBlt calls for real-time displays, so 3D performance is irrelevant. The test results should be *fairly* independent of the host platform -- this mostly just pushes the video chip.

I've written a test app that benchmarks a video card for these calls. Could someone with either of these cards run my program on their system and report back the results?

NVidia 6800
NVidia 5700
ATI X800
ATI 9800
ATI 9600
(any variations of the above cards)

Please PM me with your e-mail and I'll send the program to you (it's 53K in size). If you're worried about a virus, I can also send you source to the test app so you can compile it yourself. I can guarantee that this program is clean.

Thanks!

 

[Zebo]

pm

 

[302efi]

pm

 

[FlasHBurN]

pm

 

[kmmatney]

My guess would be the cpu would play as large a role (if not larger) than the video card...

 

[zephyrprime]

Performance for bitblt is gonna be heavily dependent on memory bandwidth.

 

[xtknight]

I sent you a PM, but I have a 9500 PRO (very, VERY, VERRY close to 9600 PRO). Last time I checked only Direct3D and OpenGL were accelerated by the video card, am I wrong? Think of it as "software rendering mode". That's why all games are programmed in the accelerated APIs instead of GDI. The video card still processes the CPU-decoded video data though (which I'm sure a Voodoo 3 can handle). Also, when I get my new system together I will try it as well if you want. See my sig for details (Computer 2).

 

[RonS]

The first test, bitmap rotation via PlgBlt, will depend somewhat on memory bandwidth. It's a relatively small bitmap. I think it depends on the driver whether the actual rotation happens in video memory or host memory.

The second test, BitBlt, performs a SRCCOPY bitblit from video memory to video memory. If the driver is written correctly, it should not cause any graphic data to be moved on the bus; the operation should be performed entirely within local video memory. It gives a real-world measure of video memory bandwidth.

So far seven guys have responded... Thanks!

 

[302efi]

I'm waiting for the email

I have a 6800 UEE, so I'll see what that can do

 

[xtknight]

Just out of curiosity, what program are you making? If I'm right that GDI isn't accelerated and you want extremely high-performance, you may want to use OpenGL instead. But since I don't know the application I don't know if that's applicable or feasible to implement easily, or just plain overkill for drawing small 2D shapes.

 

[dug777]

What sort of differences b/w the kards are pple seeing?

I can't really imagine the day u'd need even close 2 a 9800 2 push your 2D gfx

EDIT: Tho when avalon rolls out it might be a different story...

 

[RonS]

I've got eight participants so far... waiting for the first results.

 

[zakee00]

LOL yeah its called Longhorn. i find it funny that Mac OS 10.3 looks almost as good, and dosnt need DX9 spec (or, should i say opengl equiv.) GPU hardware

 

[dug777]

heh- thats not the point it will be a new toy 2 play with in a windows environment & i will feel my 9800 pro is actually doing something on the desktop...

 

[StrangerGuy]

check you PM!

 

[Pete]

P3-700@866, Radeon 9100 64MB (250/250MHz), 1280x1024@75Hz:

RotateBitMap took 460.9 seconds.
ScrollTest saw 487MPixels/sec, 1948 MB/sec memory bandwidth.

Now that I've benched this vintage system for laughs, I'll update with scores from a more recent vintage ASAP.

 

[McArra]

PM

 

[Maxil223]

pm

 

[Squally Leonharty]

I'll be happy to help you out with your program. I own an X800 XT (flashed to XT PE), so it'll be useful. I'll be awaiting your PM.

 

[FlasHBurN]

YGM. Sent you some results Ron.

 

[RonS]

Thanks to all who have participated so far. I've learned that

1) 2D Bitmap rotation is almost completely dependent on host CPU speed. This tells me that the rotation is not happening within the video card.

2) Early result point to video memory bandwidth varies widely in BitBlt performance. Remember that these are BitBlits that move memory completely within the video card, and thus are not dependant on host CPU speed.

So far, it looks like:
X850XT PE << fastest
9700Pro
9500Pro
Matrox Parhelia
5650/9000/9100 << slowest

I'll post exact numbers when I get more results.

 

[302efi]

First test: 98.2

RotateBitMap took: seconds: 98.2

ScrollTest saw: 1520 MPixels/sec... 6080 MB/sec memory bandwidth.

Are these pretty good scores ?

 

[imported_Noob]

Originally posted by: kmmatney
My guess would be the cpu would play as large a role (if not larger) than the video card...

That it true. But now because the X800/6800 cards are so powerful, the CPU doesn't play as much of a role as it used to.

 

[Zebo]

In sig

Rotate bitmap = 63.2 sec

ScrollTest saw: 1706 MPixels/sec... 6844 MB/sec memory bandwidth.

 

[TStep]

pm2u