New benchmark

[Humus]

Well, sort of new anyway. I ripped out the HSR test in my old GL_EXT_reme benchmark I wrote ages ago and put it into a new app. Then I added a fairly complex shader (basically ripped out the shader from the Mandelbrot demo) so you can measure the HyperZ efficiency in heavy shader situations too.

Download at the usual place.

 

[jvd]

athlon 64 3000+ 1.5 gigs of ram x800xt pe default clocks with 5.6 drivers

complex 8x overdraw front to back 430.84

complex 16x front to back 425.58

complex 8x back to front 56.49

complxe 16x back to front 28.32


complex random 8x 161.80


complex random 8x 132.64

Random with pre z complex 8x 406.54


random with pre z complex 8 396.32

random with pre z complex 64x 353.84

 

[Geo]

Being the suspicious type, might I inquire if you're thinking this kind of benchmark will have particular relevance in the near future? 8)

 

[Humus]

Ask DaveBaumann. He notified me that GL_EXT_reme didn't work on newer nVidia drivers, and that he still occasionally use it for overdraw tests, so since I can't debug it (though I think that sounds more like a driver bug of theirs) I thought I might just as well make a new version and bring it a bit up to date.

Heh, I understand what you're thinking though , but that's not it.

 

[Geo]

Darn! I like new benchmarks with particular relevance!

 

[trinibwoy]

So this doesn't work on NV40?

 

[Geo]

So this doesn't work on NV40?

Actually, I gathered that was the purpose of making a new one. . .with whatever was failing before (undetermined) left out. The implication being whatever the problem was, it wasn't in the core code of the test.

Either that, or Dave still doesn't have his tool.

 

[AlNom]

hm...

front to back, 64xOD 1280x1024 noAA complex shaders: 32.32
front to back, 64xOD 1280x1024 6xAA complex shaders: 34.15

 

[trinibwoy]

6800GT 400/1100 77.50

front to back, 64xOD 1280x1024 noAA complex shaders: 41.91
back to front, 64xOD 1280x1024 noAA complex shaders: 2 (exited before it completed but it was at 2fps for a long time)

Now anytime I run it I lose my mouse and can't access the menu...:?

 

[Geo]

My 6800GT opened a can of whup-ass on yours, Trini! Same settings BF, mine was 350% faster!!! 71.89 Official drivers.

[i.e. 7 fps 8) ]

The interesting part is that 6800GT was somewhat slower at 8x FtB than the XTPE upstream, at 398.30. . .and a touch faster at BtF, at 58.99.

 

[Humus]

hm...

front to back, 64xOD 1280x1024 noAA complex shaders: 32.32
front to back, 64xOD 1280x1024 6xAA complex shaders: 34.15

The difference between AA and no AA should be fairly small since this is just a fullscreen quad in both cases. Odd though that AA ran a bit faster.

Edit: The AA part in the Options tab doesn't take effect until you hit the switch button to go fullscreen or restart the app. There are still parts of the new framework that aren't quite 100% yet ...

 

[Rys]

GL_EXT_reme still works with the latest NVIDIA drivers for me. Just tested with 77.62, no SLI.

 

[fellix]

A fancy chart of mine:

Dont even got bother of testing B-t-F with 4xAA.

 

[DemoCoder]

250fps vs 30fps (front-to-back/back-to-front) complex shader, 8x OD, no AA on a 6600GT. At 1280x1024 and 64xOD I get 38fps front-to-back and <1fps back-to-front on 6600GT.

 

[Ailuros]

I'll burst if I don't say it....these tests make me want even more a high end TBDR to toy around with *sigh*

 

[AlNom]

what do most games use?

 

[Ailuros]

what do most games use?

Front to back, yet this still doesn't mean that the added benefit for back to front or random order wouldn't be welcome, even if the proportion in today's games is rather small. Pre-Z might eventually be also slightly better.

IMHO the most realistic (worst) case scenario for the foreseeable future would be 8x Random with pre-Z and 4xAA.

 

[trinibwoy]

My 6800GT opened a can of whup-ass on yours, Trini! Same settings BF, mine was 350% faster!!! 71.89 Official drivers.

Hmmmm, what'd you get on front to back?

 

[AlNom]

Front to back, yet this still doesn't mean that the added benefit for back to front or random order wouldn't be welcome, even if the proportion in today's games is rather small. Pre-Z might eventually be also slightly better.

IMHO the most realistic (worst) case scenario for the foreseeable future would be 8x Random with pre-Z and 4xAA.

Ah thanks.

Is there somewhere I can go to read more about these different methods? Pros and Cons?

 

[Ailuros]

Is there somewhere I can go to read more about these different methods? Pros and Cons?

Can't think of any relevant article right now; most games are optimized for front to back and we'll see early-Z optimized applications increasing (see Doom3-style rendering f.e.).

I said random order for a worst case scenario in order to cover the corner cases where back to front might actually be inevitable (like alpha blends f.e.) in a scene (think of a relatively small proportion though) and 4xAA, since I'm not personally interested in anything less these days. Early Z apparently helps increasing efficiency even more. As for the overdraw level I'm not sure where the average depth complexity factor would lie today, but I guess an average of 8.0 shouldn't be too far from reality, especially for the foreseeable future.