NV50 being optimized for UnrealEngine 3?

[3dilettante]

http://www.amdmb.com/article-display.php?ArticleID=267

"Here, the note mentioned that NVIDIA’s future NV50 GPU will have specific optimizations for the UnrealEngine 3. Without question, the two companies have a very close relationship which should translate into some excellent performance for gamers running NVIDIA graphics cards."

Am I reading this correctly: that the NV50 will have specific features only utilized when running that engine?

Am I just being paranoid in thinking that this level of hardware to software specialization may not be a good thing?

Now, instead of hardware cutting corners when it detects certain applications, we get hardware that hobbles itself if it doesn't detect certain software. The way it was stated, it seems to mean that instead of Unreal3 using Nvidia extensions, that Nvidia will have hardware designed around the particulars of a program.

I'm not sure I like this, it brings to mind a vision of a software market where there will be Nvidia applications and ATI applications, with neither side truly competing head to head. Worse, the success of hardware becomes tied to the success of a program, and it becomes in an IHV's interest to simply stop supporting the other half of the software market (or even trying to deliberately break such support to generate negative press for the competing software, and by extension harm the platform being formed by the competing hardware).

This reeks of a powerplay on the part of Epic as well, since the way this was worded hints that the hardware is not being optimized to a particular paradigm or method, but to an individual program.

Developers who want their games to not run like molasses on Nvidia hardware will have to license Epic's latest software just to get in the running. What if they don't want to license an engine just to get performance from one hardware vendor, and instead try to cater to a broader market that spans multiple IHVs? What happens should Nvidia see such multi-platform as a threat and decides to not support their software because it doesn't fall within the bounds of their deal with Epic?

I don't like how this seems at first glance. Hopefully there is something I missed in all this that provides a more positive light. I dislike the idea of a bifurcation of the software market into separate platforms based solely on video hardware.

Maybe I just need a new tin foil hat. :?

 

[AlphaWolf]

Well I can see how a developer could suggest certain things they would like to be available in the hardware that would make the gpu more friendly to the game.

If it's anything more than that though I would think it's a mistake designing hardware around a single title.

 

[nelg]

NVIDIA and Epic collaborate on next-generation hardware. NV50 will have UnrealEngine3 specific optimizations.

:?

 

[3dilettante]

I believe taking suggestions from developers is a good thing, which would lead to things like the enhanced depth culling for shadow rendering like in DOOM3.

However, in the case of Ultrashadow, it was a hardware optimization for a particular method of rendering, not to to the DOOM3 engine itself.

Theoretically, any other engine could use the same optimized operations, but that is definitely not how epic worded its presentation.

 

[Rolf N]

Strikes me what that may be. Larger texture cache perhaps?

It's pretty obvious that the statement has been through a fluff filter, ie severly misrepresented or even outright false, factually. Compare that to "UltraShadow" that's touted as a Doom 3 specific optimization, while in fact it's not. It's useful for Doom 3, but it's used today in Demos not written by id software. Other game engines may use it, too.

I don't see much point in discussing this marketing stuff. Even more fillrate could be touted as an Unreal specific optimization. Also note that even though they state "NV50 will contain" it doesn't mean that NV25 does not
Smoke and mirrors ...

*hands 3dilettante a tinfoil hat*

 

[3dilettante]

Strikes me what that may be. Larger texture cache perhaps?

It's pretty obvious that the statement has been through a fluff filter, ie severly misrepresented or even outright false, factually. Compare that to "UltraShadow" that's touted as a Doom 3 specific optimization, while in fact it's not. It's useful for Doom 3, but it's used today in Demos not written by id software. Other game engines may use it, too.

I don't see much point in discussing this marketing stuff. Even more fillrate could be touted as an Unreal specific optimization. Also note that even though they state "NV50 will contain" it doesn't mean that NV25 does not
Smoke and mirrors ...

*hands 3dilettante a tinfoil hat*

Perhaps there is nothing more to Epic's statement than fluff, it's just that I don't believe Nvidia has a corner on the market of shockingly obtuse and blatantly arrogant behavior, and with Nvidia actually being there as well...

In addition, I feel it is kind of hard to state a general hardware improvement like increased fillrate can be called a specific optimization, since anything can benefit from that, and if Nvidia's engineers are so prescient as to have specific optimizations for Unreal3 in hardware that existed before Unreal 2, then why doesn't the GeforceFX have a flux capacitor?

We're dealing with a master of smoke and mirrors, and for some reason he seems to be letting others in on the game. :?

 

[Bjorn]

Am I reading this correctly: that the NV50 will have specific features only utilized when running that engine?

Am I just being paranoid in thinking that this level of hardware to software specialization may not be a good thing?

I think you're being a bit paranoid

If the upcoming Unreal 3 engine can use these features, so can other engines, just like the Ultrashadow thing.

And who's to say that Ati's R500 won't have these optmizations either ? After all, it's going into the next X-Box and i would think that Epic would be interested in making their upcoming engine running as good as possible on that machine.

 

[Rolf N]

Okay, I see

What's specific, anyway? Usually it just means that a given title is the first to make heavy use of a feature, not that it's the only one.
Just like Doom 3 would be the first production renderer that really uses the stencil buffer, thus any stencil optimization is automatically a Doom 3 specific optimization as long as no other games w a similar performance profile ship. Alright with that?

In case of NV50 and Unreal 3, I must admit that fill was stupid and I hereby apologize
Maybe it's float cube maps? That would be nice. Would qualify as an exclusive, too, by the above definition because obviously no major title uses them now (not supported on current NV hardware).

What's really disturbing here is that they talk about NV50 at all after just having released NV38. I didn't really notice that on first look, thought it was about NV40. Resembles a kid making a lot of noise to garner attention
Also resembles the good ole Intel strategy of building up myths about distant future products, just in case the imminent new product doesn't satisfy expectations. That's worrying because, if true, would infer that NV40 will be an okay product but not more.

 

[Vince]

Wait, why is this an inheriently bad thing?

Other than the fact that the company doing it is... <gulp> nVidia, what's so wrong with it fundimentally? With the ever increasing investment in a game that developers must expend to keep up with the fast advancement and progress made in usable computational power and storage, we'll only see the number of propietary technology used diminish even more. So, if a large percentage of games played by consumers utilize engine U - why not optimize for it?

IBM's Fast Path technology that's slated to be used in Power5/Power5+ and Power6 does the exact same thing for common applications.

Hell, it's a great idea when you think objectivly about the percentage of time that these ICs will be spent running common applications.

 

[THe_KELRaTH]

HL2 has specific optimizations for the FX but as we know why it means nothing, and as mentioned by Bjorn it's far more likely that Epic would add feature enhancements as opposed to just optimizations to the R500 for Xbox 2 which might then be seen on the PC version.

It's all just the usual NV PR....

 

[Ante P]

UltraShadow is largely a Doom 3 optimization.
I'd expect this to be something similar and I can't see anything bad about it if the same info is available to ATi (ie if ATi has the chance to do a similar optimization).
Devs playing favourites sucks so I surely hope these optimizations doesn't stem from info only available to nVidia through their TWIMTBP.

 

[jb]

Epic in the past as done things like this. We all remember the Glide path of UT/Unreal. And how much better Unreal looked at first in Glide Ahh watching the Nali Castle flyby the first tim in Glide..oh my!

So I can only hope that the gaming comunity can move away from IHV spefic features.

 

[OpenGL guy]

Epic in the past as done things like this. We all remember the Glide path of UT/Unreal. And how much better Unreal looked at first in Glide Ahh watching the Nali Castle flyby the first tim in Glide..oh my!

So I can only hope that the gaming comunity can move away from IHV spefic features.

There was a MeTaL path as well for S3 cards.

 

[skoprowski]

Epic in the past as done things like this. We all remember the Glide path of UT/Unreal. And how much better Unreal looked at first in Glide Ahh watching the Nali Castle flyby the first tim in Glide..oh my!

They had no choice but to use Glide as the main API- Direct X was just in it's infancy when Unreal was finally released. I remember when Unreal was supposed to be one of the first games to use MMX extensions.

 

[Arun]

You guys sure have a short memory

http://www.beyond3d.com/interviews/ps_precision/

Long-term (looking out 12+ months), everything's got to be 32-bit IEEE floating point. With the third generation Unreal technology, we expect to require 32-bit IEEE everywhere, and any hardware that doesn't support that will either suffer major quality loss or won't work at all.

We can live with this in the 2003-2004 timeframe, but after that, if you don't do full 32-bit IEEE floating point everywhere, your hardware is toast.

Obvious implication: No more mixed precision in the NV50. Not that this should be a surprise to anyone considering its ILDP nature and the obvious sharing it does between VS, PS and PPP.

There got to be more, but little we can guess based on public info I fear...


Uttar

 

[3dilettante]

Not having mixed precisions doesn't seem to be an example of hardware being optimized for a single engine.

I can't think of anything that could really be called an engine-specific hardware optimization that couldn't be used for pretty much anything else, unless the hardware purposely checks for the name of the software.

Hate to see a situation where hardware functions at full capability only if you rename all of your executables Unreal3.

edit: There is another avenue available to Epic and Nvidia that still conforms to the idea of software specific hardware. If internal data types and software calls are optimized for in NV50, it can be possible for code calls for NV50 to be considered integral parts of the UnrealEngine 3. In that case, any such calls made in other software developed by those who did not license epic's engine would be violating software copyrights.
Suddenly, epic gets a defacto game standard due to one of the primary hardware designers supporting it, and Nvidia gets a huge boost by garnering an entire subsection of the game market.

Then again, the argument for epic's statement being pure BS is very strong, it's just not as fun to speculate on.

 

[PaulS]

It's worth mentionning that just because he said the NV50 was getting some Unreal 3 optimisations, it doesn't mean they're actually that specific, or that ATi aren't putting similar optimisations into their next-next generation hardware. I doubt it'll be that much more specific than something like Ultrashadow.

Wait, why is this an inheriently bad thing?

Other than the fact that the company doing it is... <gulp> nVidia...

That's all it seems to take on some forums, which is a really sad state of affairs.

 

[Reverend]

UnrealEngine3 is a year+ away so we maybe "worrying" prematurely about thhis thread's subject matter. Although the engine will not work on anything less than DX9 hardware, a lot can happen in one year.

 

[3dilettante]

UnrealEngine3 is a year+ away so we maybe "worrying" prematurely about thhis thread's subject matter.

My question is with the hardware and software being so far in the future. How could Epic possibly claim optimizations to software they haven't even gotten running on hardware that doesn't even exist yet?

That wouldn't work unless they have come to an agreement of some sort that would provide some form of a guarantee that things will turn out that way.

It seems kind of unwise to tie either an engine and an architecture so closely, since a delay to either will be a detriment to the momentum of the other.

In addition, if the collaboration is kept closed, any optimizations and feature calls may not become public until after Epic's engine is released, leaving the rest of the software world to figure out how to use these features while Epic has a head start.

It would be no good for Epic to devote a lot of effort into this collaboration and to have Nvidia touting these same optimizations to the competition, but to not do so would hurt the acceptance of these features.

There are a lot of uncertainties, and I fear that the first impulse on the part of the corporations will be to minimize uncertainties by keeping the system closed. It would be a decent power play, but not too great in the long run.

 

[Bjorn]

My question is with the hardware and software being so far in the future. How could Epic possibly claim optimizations to software they haven't even gotten running on hardware that doesn't even exist yet?

Carmack showed a version of the Doom 3 engine on the 2/21/01 John Carmack showed off the Doom 3 engine for the first time at the MacWorld expo.

And i'm guessing that he had the basic engine up and running quite some time before that. I would also think he at that time already had some pretty good ideas about what future hardware had to improve for the Doom3 engine.