Half-Life 2 will not feature MRM Technology

Originally posted by fifty7var
That was my point, the automatic LOD routines, including dynamic LOD routines often produce subpar results when compared to those manually created by the artists.

Adam
Click to expand...

No, no. Lol. It depends on who made the LOD. If it was a moron, then yeah. :p

Put it this way... you can't get more real than real.
 
Originally posted by fifty7var
That was my point, the automatic LOD routines, including dynamic LOD routines often produce subpar results when compared to those manually created by the artists.

Adam
Click to expand...

Please read Intel's MRM technology page before saying such a statement.
 
Originally posted by $niper
Please read Intel's MRM technology page before saying such a statement.
Click to expand...
Its been a while since the tf2 days that i've read anything on it, but i'll give it another read.

I think i'm thinking of this too much from a 3d animation standpoint. The meshes it produces are ugly, ie. no preservation of edgeloops (very useful for proper facial animation), or detail around joints (for good deformation), etc. Intel's MRM was a very good implementation if i remember, but i don't see the lack of MRM as a problem. Humans are still the best poly reducing algorithm around.

Adam
 
fifty... dynamic LODs are made by humans. Lol.
 
Originally posted by Lifthz
fifty... dynamic LODs are made by humans. Lol.
Click to expand...
Dynamic LODs are made by algorithms ; and compared to humans, pretty dumb algorithms.
 
$niper, no matter how 'smart' MRM tech is, it is very possible for it make little mistakes that, while minor, can be jarring to view. The same goes for programs that generate lower-poly models; the program doesn't know what you are trying to draw and so it cannot possibly do as good a job as a human.

I don't know how many of you all program, but performance has to be paramount...you will all bitch a lot more about crappy framerates than you will about static vs. dynamic LOD.

Also, what is this vs. stuff? This is not a black and white issue. "Dynamic LOD is just better". Ok, well not if I make 50+ models because I want it perfect and MRM wouldn't do as good a job. Or if it does it in smart ways like only switching objects at the edge of screen or only when the view is moving, since it's harder to notice switches then.

And don't tell me that since MRM didn't slow down Battlefield 2 or whatever shows it isn't a performance hit. If I hack MRM into Quake and show that I still get 500 fps, that doesn't mean anything. There's an old expression, but your money where your mouth is: buy everyone who buys HL2 a dual CPU professional graphics level workstation, and then start complaining about features that were cut for performance.
 
Last post before i go to work. It turns out that Max's Multi-res Modifier is Intel's MRM...

MRM demonstration

I made this 5 mins ago. The original model was around 10k, and it scales down to 1k faces over 100 frames. no, it doesn't play this animation smoothly in max (p4 1.6 geforce2 256 ram). If i have time tomorrow, i will make an animation with 8 LOD meshes and compare these two zooming out.

Adam
 
Neither method of implementing LOD is better across the board. There are times when dynamic is better and there times when static is better. In Valve's case, using Intel's MRM technology actually decreased performance, so as an LOD system it was a failure and was clearly not the better choice over their own in-house static system.

You can pontificate and compare white papers all you want, but it's real world performance that matters, not which system is theoretically better.
 
Static LOD is inferior to dynamic just as Static lighting is inferior to dynamic. I know it's two different stories.
Click to expand...

You're a big fan of buzzwords, but hype does not a good technology make. No form of LOD is realistic: it's always a hack in order to lessen the load on poly renderers. What matters is how good you can get things to look at the best performance possible, not what buzzwords the tech sheet says. In this case, dynamic LOD is not necessarily more advanced than static at all. Dynamic lighting is better than static because it's closer to the "right" way to light objects, and because many in game events alter lighting in an area. But dynamic and static LODs are just two ways of trying to achieve the same basic effects.

Or maybe we're back to the days of using "all of" DX9. Look: I baked you cake, and I used "all" the ingredients! Including porkchops and onions and orange juice!
 
In theory, MRM or dynamic LOD systems would be superior, as it would be great to make 1 model with a million polygons and have the engine dynamically scale it down to the point where your computer can handle it. Of course the quality of this scaling is questionable. An algorithm may lower polygon counts uniformly or by some other algorithm, while an artist would keep higher detail where it matters.

However I think it's fairly obvious that this would take a lot of processing power, and in a game that is so processing intensive, it's just not a feasible feature to impliment. The obvious choice is to use a static system.

If you do happen to have enough computer power left to do something like MRM, I'm sure it would have been taken into consideration. Maybe that's why it was originally planned for TF2; It was being built off a modified HL1 engine and most computers would be able to handle it and features like MRM would be considered.

I wouldn't be worried about popping all over the place. It's up to the designers what parts of the game should have multiple models and when they want to sub them in. It would be safe to assume the models wont change unless you change settings or only if the model is so far away you won't even notice anyways.

Point is this: Valve knows what they are doing, they won't be doing anything that will pull you away from the experience of the game.
 
Flynn, you're pushing it a little here. I suggest you calm down and think before you post. Remember that the HL2.net guys have to pay for all this bandwidth. :)
 
lol sorry, i bought a case of bawls and am shaking really bad, but the goodside is im typing very fast, lol
 
It's just impossible to discuss anything in a healthy and rational manner in this forum. This place is just crowded with nubs that have nothing clever to say, and just post stupid things that fill up the thread pages and cluster the visuals, making it pretty hard to read!

:pUKE:

d_P
 
Question : Can the Maya/SoftImage/whatever automatically generate lower poly models, given a high poly original? I would have guessed this is automated by now, so static LOD (on models, anyway) isn't quite so unpleasant.
Click to expand...

Auto LOD'ing on a model (like 3DMax's 'Optimise' modifier) is never going to be as good as a hand-reduced model. The Optimise modifier, for example does not know what parts of a model are most important.

However - an auto optimised mesh will also generally fix your UV co-ordinates also, where a manually optimsed model will need the UV hand-altering also.
 
Originally posted by Apos
Or maybe we're back to the days of using "all of" DX9. Look: I baked you cake, and I used "all" the ingredients! Including porkchops and onions and orange juice!
Click to expand...

You referring to S.T.A.L.K.E.R out of nowhere like that?
 
Originally posted by Apos
Look: I baked you cake, and I used "all" the ingredients! Including porkchops and onions and orange juice!
Click to expand...

Hey! you stole my idea!

Anyway, I agree, it is like saying "handguns are inferior to rocket launchers". Until you get stuck in a cupboard and need to shoot the lock out.

I code proffesionally and almost every decision I have to make is based on this type of thing. If some method is better in one situation then it is gonna be worse in a different one. In the end you look for the one that suits your requirements best (or some kind of compromise).
 
Just pointing out the perchance people have for getting a little too excited about PR fluff that doesn't actually mean anything. Just because you slap "dynamic" on something doesn't make the end result any more advanced or realistic.
 
Here's a better answer I received from Valve on the matter:

"The CPU overhead for MRM cost too much. It was originally designed for an environment where per triangle costs versus tesselation were much higher than they turn out to be if you organize them in a hardware-friendly format. Without MRM we render higher-resolution models faster. Since lower resolution models are contained in the higher resolution meshes, there isn't a memory cost.

The simple answer is graphics hardware got fast way sooner than anybody expected." -Brian Jacobson @ Valve
 
Originally posted by Apos
Just pointing out the perchance people have for getting a little too excited about PR fluff that doesn't actually mean anything. Just because you slap "dynamic" on something doesn't make the end result any more advanced or realistic.
Click to expand...

The reason why Intel's MRM tech would have been a godsend to modders is not because it looks better (though it doesn't look too bad), but because it would save modellers a lot of work. Not only that, but it would save memory and disk space for more important things you want to include in your huge install file.
 
Since lower resolution models are contained in the higher resolution meshes, there isn't a memory cost.
Click to expand...
Ok, this seems to hint that the subset of verts in the lower res model are in identical positions as they are in the higher-res model (the same must be true for the UV information too).
 
Sure, I agree that letting the engine do it is much easier on modders. Though, with most modern modeling programs, there are already pretty simple functions that you can use to create simpler meshes at the touch of a button once you've created your high-res model. And it doesn't sound like the memory overhead is much of an issue at this point.
 
Originally posted by $niper
- Increased workload. Instead of just needing to create one model with INTEL's tech, modellers will need to create - according to Valve - 3 to 4 LOD models.
Click to expand...

3 to 4 models? That seems like a huge waste of time! Making just 1 relatively low poly model for HL1 took forever, now I have to make 4?!?You'd think it wouldn't be that hard for them to scale poly counts down or up on the fly.

By the way $niper. Where does this "according to Valve" come from? Is it through your work on Sven Coop?
 
If you were expecting them to scale models on the fly in the game in real time, how hard can it be to to have a modeling program do it in a few clicks and even a chance at some fine-tuning?
 
Re: Re: Half-Life 2 will not feature MRM Technology

Originally posted by DarkStar
3 to 4 models? That seems like a huge waste of time! Making just 1 relatively low poly model for HL1 took forever, now I have to make 4?!?You'd think it wouldn't be that hard for them to scale poly counts down or up on the fly.
Click to expand...
No, it's not that hard, but it is resource intensive and they were able to realize better performance with a static LOD system. The trade-off is more work on the front end, but better perfomance and arguably better results. As for the number of models, I thought only two were necessary, a high-poly model and a low-poly one. I could be wrong, though.
 
Technically the number of models can be anywhere between 1-??? The idea with having more than just a high and low, is that the transition is smoother.

Some things i've found....

1.) With High-poly models (>10000) MRM is very slow.
2.) Around 5000, its decent at scaleing
3.) less than 5000 performs pretty well.

my conclusion, As 3 years passed between TF2 and HL2, the bottleneck was no longer video rendering perfomance, but cpu proccessing. With added things like Havok, and more complex AI, it didn't make since to waste cpu power on MRM since it no longer spead up the pipeline.

Working on a comparison video between static and dynamic LOD IQ...

Adam
 
So what's the average number of models for each element in Half-Life 2?
 
Ok, lets play a little game. Watch this video of two identical models side by side as they move into the distance. One features Static-LOD. The other features Intel's Dynamic LOD. They both start out at full detail, and are at 10% detail by the time they reach the end.

The Static LOD uses 8 meshes with detail evenly spaced between 100% and 10% (ie: 100%, 87.14%.... 10%)

The MRM mesh has a smooth transition from 100%-10% over 200 frames.

But the question is: Which model uses which? The left, or the right?
If i didn't make it, I wouldn't really be able to tell. Slap on textures, and animate these suckers, and there's no way you could.

MRM vs 8-x Static-LOD (271 KB, DIVX 5.03)

Adam
 
Originally posted by fifty7var
Ok, lets play a little game. Watch this video of two identical models side by side as they move into the distance. One features Static-LOD. The other features Intel's Dynamic LOD. They both start out at full detail, and are at 10% detail by the time they reach the end.

The Static LOD uses 8 meshes with detail evenly spaced between 100% and 10% (ie: 100%, 87.14%.... 10%)

The MRM mesh has a smooth transition from 100%-10% over 200 frames.

But the question is: Which model uses which? The left, or the right?
If i didn't make it, I wouldn't really be able to tell. Slap on textures, and animate these suckers, and there's no way you could.

MRM vs 8-x Static-LOD (271 KB, DIVX 5.03)

Adam
Click to expand...

They look the same :|
 
And to further illustrate my point, this version has the static-lod (in the same position as the previous video) only using 4 LOD meshes instead of 8. if anything, the Static looks better, because it retains its detail for a longer period then the MRM version.

We shouldn't have any LOD problems really in HL2.

Adam

PS. Static LOD is on the right, MRM is on the left.

{edit} helps if i post the link.
4x static lod vs MRM
 
Originally posted by fifty7var
And to further illustrate my point, this version has the static-lod (in the same position as the previous video) only using 4 LOD meshes instead of 8. if anything, the Static looks better, because it retains its detail for a longer period then the MRM version.

We shouldn't have any LOD problems really in HL2.

Adam

PS. Static LOD is on the right, MRM is on the left.

{edit} helps if i post the link.
4x static lod vs MRM
Click to expand...

Yeah I see what you mean. Looks perfectly good to me.
 
oh. it does look fine. i was worried before because my idea of static LoD is what was going on with the architecture in serious sam 2, i.e. god-awful popping.
 
Done badly, any LOD system looks shit.
Of course, Valve have a habit of doing things well....
 
static LOD looks good, but with all the zombies and flying things shooting at you i dont think you'll have the time to study where the models change (if vavle did a good job, which they probably did :p), like someone sad, more poeple will bitch about the framrates :D

i know i wont.
 
Excellent thread, gentlemen...

This was a good conversation to follow; thanks for the post, $niper.

Cheers,

EnochLight
 
You must log in or register to reply here.

Similar threads

B
Valve artist Drew Wolf's new website reveals Team Fortress 2's female cast and more
Replies
0
Views
60K
Barnz
B
B
Half-Life: C.A.G.E.D. is now available
Replies
0
Views
15K
Barnz
B
B
Half-Life: C.A.G.E.D. press release
Replies
0
Views
17K
Barnz
B
B
Barking Dog Studios' work for Valve
Replies
0
Views
12K
Barnz
B
B
From Warren Spector to Arkane Studios: Half-Life 2's cancelled Ravenholm episode
Replies
2
Views
10K
Clan Wolf 2
Clan Wolf 2
Back
Top