Interesting F.E.A.R. benchmark anomally with X1800

Have you asked Monolith about this as well?

 

I smell a hotfix for FEAR released soon after the CATALYST 5.11 drivers are released...

 

My guess to why this is occuring would be this:

There was some kind of rendering anomalies with FEAR with Cats so ATI fixed it by having the drivers detect if FEAR.exe is running and fixing the problem (the problem may not be apparent in the timedemo alone). Obviously fixing the problem could cause a performance hit as the drivers would have to re-write shaders or whatever for one particular instance. Renaming the .exe would stop this fix occuring and performance would increase.

Just a guess, but makes the most sense!

 

Well, how do we determine getting the "exact same (render) output"?

 

Several people have posted that FEAR running normally feels "smoothest" whereas the re-named EXE, though faster, also stutters more.

So this adds a whole new dimension to benchmarking if ATI is actually optimising for smoothness, not highest average.

Damn, the questions this topic has generated are just too much fun.

Jawed

 

sized

It’s a "bug" in that we simply got an IF statement backwards Thanks for pointing this out to us, you have just helped us get a big performance gain. Of course there is no difference in the rendering, it’s just a CATALYST AI game specific optimization that was good for the demo version, but backfired in the final version. We will get it sorted out in a future Catalyst (not 5.11 which is being posted tomorrow by the way).

Any other help anyone else can provide in bringing up our performance, make sure you send it my way.

Special thanks to PC Format magazine in the UK (and especially Jeremy) for getting this to our attention.

 

*dispatches two crates of Red Bull*

New Cats tmrw, cool.

 

Ya, and if you havent heard of or used the 5.10a beta's youre in for a pleasant suprise

 

Hopefully you weren't too mean to the poor guy who coded that part!

 

Not at all, I actually pointed him to this thread and told him he made the daily forum happenings. His 15 minutes of shame in the forums you can say.

 

Well, I guess that the performance will be even better than the currently renamed fear executable... all there needs now is a hotfix...

I expect some rebenchmarking soon... see how much of a change this confirmed info makes (probably good to wait til they release the fixed version, unless people really want to see the performance differences of the renamed executable.

 

I hope you guys don't wait to fix it untill Cat 5.12 arrives, this issue is something which needs a quick hotfix right now!

 

Hopefully there won't be as much choppiness when the fix comes around, so the higher average frame rate can be put to use more effectively in the real world and not just showing higher average frame rates in reviews.

 

First of all I hope that claims of choppiness will be verified by objectively measuring and reporting.

 

Does this bug also applies to Driverheaven's recent comparsion vs the 6800 GS ?

http://www.driverheaven.net/reviews/X16_GS/fear.htm

 

I'm sure the bug was present on pretty much every benchmark until now (including the dh one).


EDIT: Just rerun the test and the X16 outscores the 68GS in fear at 1280 2aa 8af now... will update the article accordingly.

 

What?

 

Could someone with a high end ATI card, X1800 preferrably, run this renaming test while making absolutely sure that no AA or AF is enabled (this means making sure there is no AA override in the CCC either). Otherwise the settings should not matter. Run it once as FEAR.exe and once as whatever you care to name it and post your results here, pretty please.

 

@Veridian3

Thx it really was strange to see those results, it feelt like a major setback for ATI i knew it couldn't be right, btw do you know when ATI is gonna release the next infos about Hardware Video Transcoding like for example:

- Is H.264 Transcoding implemented now ?
- Is it hurting the Quality much compared to CPU encoding and if it does about how much quality degredation we talk here (MSE,PSNR) ?
- Is it be able todo High Profile (H.264) on any core ?
- Is it working on every core and how much will be the Speed difference ?
- Is it be able todo 1 Slice on the CPU and 1 on the GPU ?

Maybe Terry could answer this even better (it's little oftop i know)