HL2 30% faster on ATI Hardware.

As a newbie I would appreciate the valve comments more if they included a statement as to why the results were so heavily in ATI's favor.

Looking at benchmarks across the board, it's obvious that ATI's DX performance is excellent and I would not be remotely surprised to see HF2 run faster on their hardware. What I am surprised at is that it runs 30% faster vs. current generation hardware. My limited understanding is that there is a technical reason for why the NVIDIA cards have a lead in doom3 dealing with colorless pixels and with the way they deal with overdraw. Given that the optimization of the two companies would tend to produce some polar specializations over time, is there a similar advantage on the ATI side of things that will help them make great gains in a game like HL2? Are there new technologies that HL2 is introducing that work better on ATI cards? The main reason I ask is that there isn’t another DX benchmark where ATI is beating current generation NVIDIA hardware by such a large amount. If we are dealing with a technology issue, then by all means I want to hear about it. If we are dealing with an optimization issue (as some suggest is the case with D3) then we all loose. This trend is rather disturbing. If the two companies continue to funnel money into game developers like they are marking their territory we will end up having to purchase a second computer with a different type of GPU if we intend to enjoy all the popular games in any given year.

I can stomach the idea of purchasing a new GPU every few years. I can even stomach the idea of purchasing 2 GPUs to support a technology like NVIDIA’s SLI since the second purchase adds value to the first. However if I have to purchase two different GPUs just to enjoy all the games available in a given year I’ll buy a next generation Xbox or something and forget about PC games all together.

 

While I do not know the exact benefits of membership in TWIMTBP program, I believe that it's commonly held that Carmack's wishes have influence at nVidia and that nVidia (clearly, as evidenced by Carmack's own statements) optimisizes for his engines (specifically DOOM3). I didn't post on this because I didn't want to be involved in the flame war.

Additionally, it is reasonable to assume that Carmacks "gold standard" statements were made with the same intent as Doug's benchmark statements. Carmack made a statement that he believed to be true, based on his assement of the available drivers. Doug made a statement based on benchmarks run internally at valve. EACH made those statements in an atmosphere from which they stood to gain, regardless of the statement's validity. They both presented the "truth" (although admittedly, Carmack's 'truth' is far less quantifiable than Lombardi's) but both had conflicts of interest in which the gaming public may have lost. This is a major problem IMHO. Developers and IHVs on both sides of the divide are equally at fault, and we get to deal with the mixed messages. Sorry if i seemed biased by only posting about this with reference to ATi, because it is truly an industry wide problem.

As you pointed out jvd, there is a clear distinction between optimising in order to make a particular IHV look better, and just getting the hardware to work, which, as you pointed out many NV3x "optimisations" were done for compatibility/playability's sake (just to get marginal framerates from the occursed hardware) this was done at both valve and id.

Does id's relationship with nVidia cast their "optimisations" in a more dubious light? Absolutely. Is valve's relationship with ATi (and botched voucher deal) a possible impetus for R3x0 and R420 performance in Source? Perhaps, but given R420 PS2 performance in other apps, probably not. The same could be said for nVidia OGL performance. Please do not mistake this for an assertion that valve is doing in-engine optimisation, as their is no evidence of that, but the conflict of interests still exists, which is the real problem. As I stated above, it is an industry wide issue.

** edit **
I think the DOOM3 took so long because of content creation.
--peh--

 

You really are amusing you know that. This topic was directed toward the ATI/Valve issue until you brought this in -

 

trinibwoy you should go read the first post in this topic as that person brought up doom3 and the similar ways this is going down

 

Touché !!

Still shouldn't use the behavior/practices of one company to excuse that of another.

 

and people shouldn't dismiss another companys actions to make another companys look worse

 

Good to know I can go away a few days and nothing changes.

 

You were gone?

 

Huh? What are you talking about?

Digi, where did you go?

 

read back a few pages

 

What about these benchs?

http://www.xbitlabs.com/articles/video/display/graphics-cards-2004.html

http://www.xbitlabs.com/articles/video/display/graphics-cards-2004_26.html




This is Half-Life 2 beta...

Using "normal rendering mode" for GeForce cards (rendering in 32-bit FPP).

VALVe/Gabe will cut down 30% of the performance in GeForce cards?

For me, this shows the influence of ATi in development of the game. This is a shame. nVidia don't do that with iD and Doom 3.

 

We have no way of knowing what modifications, additions, subtractions, and optimizations Valve has made to the engine and game since the beta code was written. Also remember, at the time that code was written, Valve had put much more effort into optimizing for nVidia's FX then they had spent optimizing for r3x0.

Because of all these things, benchmarks from the stolen and incomplete beta code are useless. Worse then that, it can easily be misleading.

 

They look pretty much spot on - previous generation of ATi's cards slaughtering Nvidias due to nv30s weak ps units, but Nv almsot catching up with the new cards because of the nv40's better shader units(which is pretty much the case for all DX9 titles).
I dont expect the final results will be too far from what was shown there.

 

we shall see . But remember at some ponts 30% may only be a few fps more .

ALso this game will have support for sm2.0b and 3Dc right out of the box.

We don't need to wait for it like in other games .

 

Well, this is not very hard to explain. The X800XT PE has a 30% higher fillrate over the 6800 Ultra. Unfortunately for ATI, in most games they have more bandwidth limitations than NVidia since both cards have the same bandwidth, so the X800 can't take advantage of it's high clockspeed. HL2 has extensive shader usage, and the X800XT doesn't have to worry about it. Much like FarCry:
http://www.firingsquad.com/hardware/far_cry_ps2.0b/page14.asp

I think it's perfectly reasonable. Only when NV40's architectural advantages come into play or R420's clock speed advantage gets masked will NV40 win (though I'm stating the obvious here...)

 

Remember this?
Beta
Final

Things change. Even his 30% number could change drastically. ATI now has 3Dc too, which should help them since it wasn't in the beta.

Don't assume he's going to "cut down 30% of the performance in GeForce cards". That's a dumbass statement with no validity and you know it.

 

Remember that the beta also showed the X800XT taking the lead with af/fsaa.

The beta frankly looked system bottlenecked up to 1280x1024 4xfsaa 8xaf :?

Anyway I think it was digit-lifes benchmarks or maybe ixbt (same thing really) that showed at 1600x1200 4xfsaa 8xaf the X800XT with a 10% lead. If the 6800 runs a more full precision path in the final version and with the addition of 3DC I can see a 20% lead, but 30% is a stretch too far IMO.

 

I don't know . the x800s will have the advantage of sm2.0b and 3Dc . I'm also sure valve developed closely with ati to get good performance outo f the cards .

the 6800ultras however might be fored to use the x800's path. Much like doom3 where the x800's are forced to use the 6800 path.


I frankly don't see why we can't see a 30% diffrence in speed.

Remember the x800xt pe is clocked at 520mhz on the core vs the what 425 mhz of the 6800ultra ? It has almost a 100mhz more on the core clock.

 

Well those scores from the beta are only going to get better for both cards (I too dont believe Gabe/Valve would be so dumb as to stifle 6800's performace, and given the leak was available to many it wont be hard to see if they have) so all is good regardless of wether you own a x800 or 6800.

 

3DC is about quality, not speed, AFAIK. Except if Valve is not planing to use the DXT5 swizzle the red and alpha trick.