forced FP16 on HL2 fixes FX performance problem???

Wow. . . When people were saying that the GeforceFX still wasn't competitive even with FP16, I didn't realize they meant it was getting totally trounced. So, at only 31 fps. . . why is this an issue? How many times do you hear people saying that 60 fps is the playable-minimum for shooters?

 

<offtopic>
Try http://www.pricenetwork.ca/. It functions sort of like Pricewatch.
</offtopic>

 

OT, sorry, so I'll be quick.

Canadian customers have very many options:
www.pricenetwork.ca (lists all the mom and pop stores' prices)
www.computers-canada.com (lists all the mom and pop stores' prices)
www.redflagdeals.com (lists all of the deals at stores)
All prices listed in Canadian dollars.
So an American quarter will buy you a 6800 Ultra up here.

And the forums at each have used stuff for sale.




Oh, and for everyone who's complaining about this HL2 thing, I think that it needs to be reiterated that Valve is a businesss and not some modern day computer deity that owes us anything. They put out a product. If it's not to your liking, tell them the best way you can--with your wallet.

Either that or fix the bloody game yerself


EDIT:

Partially beaten by mondoterrifico.

 

Thanks to Demirug for the pics and Beren_PCE for the benches! I was going to ask Demi for framerates, as the visuals look very close in his pics, but Beren's numbers don't appear to be worth making a fuss over. Sure, you gain 30-50%, but you're still running at basically half ATi's DX9 speed and nV's DX8 speed, and only 31fps at most at 10x7. Doesn't seem worth the bother for the initial launch and benchmarking festivites, tho perhaps focusing on a water demo paints an unrealistically unflattering pic of the 5900--it might pull closer in other scenes.

I think Valve should at least make the _pp effort for FX owners, perhaps as an unsupported beta patch.

 

Good. The only question now is how much will GF6x00 get from using FP16 instead of FP32. What if it's gonna get that kind of speed-up too?

 

But you would probably still be able to play at 800*600. And perhaps AA is free if the game is completely shader limited. There might be a lot of FX owners who would prefer that instead of running DX8.

And as you say, at least a unsupported beta patch would be a nice effort for the FX owners.

Will be very interesting to see what happens with the GF6x00 performance. Especially in those levels where f.e the 6800 GT performs close to the 9800 XT.

 

Yes, but you have to do some heavy config work prior to that. Ever tried to use projects written in VS6 in VS.NET? Even the preconfigured conversion mostly fails. I admit I haven't bothered trying much, but it doesn't work "out-of-the-box".

 

Maybe there is some more speed if we convert some of the 2.0 shaders back to 1.X. A quick test show that ~8% of all 2.0 shaders in the engine should work well as 1.X version. But I am not sure how many times HL2 use this shaders during the game.

Unfortunately Valve make my life harder. All my timedemos are dead after the new patch.

 

If you start a new projekt you only have to select the right type to get an unmanaged application.

Converting an old VS6 app to VS.Net can be a pain. Most times we have trouble with MFC or ATL.

 

Thanks for the info!

// looking for VS.NET CD's...

 

You can just guess why i recorded my own.


Update:

As for ForceWare 67.02 (supposedly nVidias "HL2" drivers) - same performance as 66.93.

 

Maybe 31 fps is not much in comparison to ATI-s 70 or so, but it's more or less playable (unlike 19,46 fps on FP32).

 

You have to remember frame rate is HIGHLY subjective. 31 fps maybe enough for one, but it may NOT be enough for another. Given that the DX8 is still x2 fast (ok its not always x2 faster but close enough) it still backs up Valves claim on shader last year.

 

IMHO

For a user who wants to play HL2 on his FX5900 with DX9 eyecandy it's highly acceptable. Even if it's not fast enough for HL2, it's still fast enough for CS Source.

 

A 31 fps average implies too many low-dips for me that would be unacceptable. (Not trying to flame, honestly pointing out something.)

 

benchmarking is not real gameplay. usually the fps in real gameplay is better than those average benchmark numbers suggest.

 

Ah, thanks. I thought it was the other way around, that benchmarking was a little higher than actual gameplay since no AI was involved.

I have GOT to get busy benchmarking. :?

 

Well...i kind of recorded a real playing session. So posted numbers are very close to the real thing.

 

Depends on the game and the benchmarking. In a lot of games as soon as you go off the benchmark, you have to contend with AI, player input, sound, physics calculation, fast changes of graphics/textures/models, etc - all things which usually drop the framerate compared to the carefully controlled benchmark mode.

 

Generally speaking I would say you are closer to correct with your second statement Digi, there is usually a lot less going on with a benchmark (the benchmark itself, the hardware environment and software environment on a user PC) than actually playing a game. Exactly how much is going on with a benchmark is entirely dependant on the benchmark itself, and what it does, but I would always look at them as "best case" scenarios.