Of course XNA can be used to make a game 'on that level'. It's not the API that makes the game; it's the content.

I second the side-scroller beat'em'up.

I can only be helpful if it's a 3d side-scroller

A multiplayer side scrolling beat-em up would be cool.

As long as it has nothing to do with retarded memes, anime, or the internet.

Yeah I agree with that last statement

lets make a game where you can play wow and have sex at the same time

You can probably still play WoW with just your right hand.

some of us jerk it with our right hand

Ambidextrous people have it so easy...

So would this end up being an Xbox live arcade thing?

That's true, but he still has a valid question. I haven't used XNA much so I wouldn't know, but maybe a better question to ask is, does the API become a hindrance at a certain level? If I wanted to make Far Cry 3, at what point (if any) would XNA cease to be useful?

Floating point arithmetic performance is roughly 20% that of native code.

Agreed. For my part, I'd rather take part of a serious project. By serious I mean a good and honest effort at a decent game, not necessarily with a super serious plot. I found Castle Crashers to be a fun and interesting game. Something along those lines could be successful and amusing both to design and play.

Make that 3 people!

I think he was talking about Ruthy.

XNA gives you high-level abstractions to play with. The advantage : You don't have to **** around with that low-level crap. The disadvantage : If you need to **** around with that low-level crap, you can't.
I don't know where exactly XNA falls in that spectrum because I've never used it, but I don't think we will need to **** around with that low-level crap anyway. A side-scroller does not require a sophisticated super-efficient engine.

No. The person I'm talking about knows. If you have to ask, you aren't him.

JM, the only thing I can think of that you can do with a devkit but can't do with XNA is make your vector class use ALTIVEC.

How do you define "low level crap"?

Do you have any benchmarks for that? All I can find is this:

http://www.tommti-systems.de/go.html?http://www.tommti-systems.de/main-Dateien/reviews/languages/benchmarks.html

And an excerpt from an IEEE article (full text not available to me) that says:

Figures 2 and 3 show C# floating point performance to be virtually identical to C, but .NET memory management requires further tuning to reach C efficiency...

Also, it was my understanding that XNA isn't really that high level. To write a game engine with XNA you still need to know how to write a game engine. It provides APIs for sound, video, input etc., but doesn't give you much in the way of gamestate management or scripting.

In this context, vertex buffers, texture memory, things down there. I was under the impression that XNA abstracted most of that away. Load a model, render it, two function calls; it'll go ahead and use VBOs and load all the textures and what not.

which gave C# a score of 46421 and MSVC++ a score of 8750 for a matrix multiply (18.8%).

An infinite loop with some floating point operations is meaningless. We're talking about a JIT'ed language that was created by the exact same people that made the C++ compiler we're comparing it against: if the exact same loop in both languages isn't converted into exactly the same set of instructions there is something seriously wrong. What we're seeing is the performance difference of a loop in both environments.

And they're absolutely right. The code gets JIT'ed and runs on the exact same hardware as native code: the performance degradation comes from other sources as well as an inherent handicap in the platform neutrality of the runtime.

Assuming two vectors have already been loaded into memory, it will take a minimum of 7 instructions to calculate a dot product compared to 2 if you write it with native code.

Pretty much. If you can write a game with XNA you can write one with C++.

Scripting is an interesting topic. XNA on the 360 uses the .NET Embedded Framework which doesn't support codedom or loading assemblies. .NET makes a pretty interesting target for a game in Windows just because of the kind of scripting support you can get: within an order of magnitude of native code, easily, compared to 20x for some of the fastest object-oriented scripting languages.

If you implement a scripting language in an XNA game you're going to see a ~200x performance penalty in execution over native code just because you can't make a jump table.

Actually OpenGL is higher-level than XNA is.

Someday, I'll learn to read. Actually though, that benchmark is .NET 1.1. What about the 2.0 runtime?

What performance critical areas would you need matrix multiplication done in C#? I thought DirectX provided APIs for doing really math-heavy things in the hardware?

This really isn't my area (the last renderer I wrote was in 11th grade with OpenGL 1.1). All I remember is people going "managed DirectX is 98% as fast as native!" I would think that a game loop would still spend most of its time waiting on the graphics hardware.

How about, a 3d side-scroller, where you are a tank! That has rockets on the back and gives it the ability to jump high. Multiple guns attached, blow **** up

There's slimDx, which is awesome, but I do not think it works with XNA.

Same **** different namespaces.

Transforming a graph node into world space, pre-multiplying matrices for shaders, things like that. Matrix multiplication is the tip of the iceberg though. Do four dot products and it's the same as doing a matrix multiply. Care to take any guesses on how many millions of times Havok computes a dot product in a single frame of Oblivion?

Most of the physics assemblies available for .NET are just wrappers of C libraries. ODE.net, for example. Marshalling is actually faster than implementing physics in C#! It's hilariously tragic.

Managed DirectX is just a thin wrapper over the native API. Why wouldn't it be almost as fast? The only information you can get out of the claim is that COM interop has a 2% performance penalty.

idea

side scrolling shooter (think contra, metal slug, etc)

main character: Carl Thibault... a werewolf with an AK47

But the actual CLR runtime is different. I don't know if this matters for floating point, though. You tell me.


I dropped out of linear algebra. :saddowns:


Yeah, I wasn't going to suggest implementing (complex) physics in managed code.


To be fair, there isn't much overhead when marshaling blittable types. But why is .NET so slow in this respect? The benchmark I posted showed Java having nearly a two-fold increase in speed. Is it a problem with managed code or a problem with .NET?

i dont understand the conversation, but i cant seem to stop reading it...

My vote goes to "SPOOKY TACO - THE GAME."

Guess I'm a screwball.

Anyway, I'll be willing to help out if I think something will actually come out of the project. I can do writing (any, not just "story writing," including writing a game design document for an idea that's *gasp* not my own), and whatever art is needed. Should the team think I should do any sort of game design, I of course will help with that too (because I'm screwball enough to think I'd be useful with that as well.)

It shouldn't. The main way of optimizing floating point math is vectorization and as far as I know even .NET 3.0 doesn't do it. They do use SSE for some casts but that's about it I think.

I'm honestly not 100% on x86 code generation but as I understand it the main way of doing automatic vectorization is with some pattern matching tomfoolery. The CLR can't even inline functions or unroll loops so it would be utterly impossible to use SIMD on floating point calculations performed in a loop like that matrix example. The author of the test would have gained ~2% performance relative to C++ if he had manually unrolled the loops first, incidentally - check out the XNA matrix classes in Reflector for a good example of how you're supposed to do things.

But it's so easy.

The physics for a game has to come from somewhere and if you're developing for XNA it's going to have to come from C#. At least 2D physics has roughly ~1/3rd the complexity of 3D physics, and you can mitigate most of the performance consequences of managed code by properly using value types.

A little bit of both.

Almost everybody thinks they can design a game but almost none of them can actually do it. Don't take it so personally.

I was mostly making fun of the fact that you seem to qualify "good game designer" with "someone who's going to school for it" which I personally find laughable since (at least when I was applying for colleges as I admit it's slowly changing now), learning "game design" means learning how to program and make 3D art and working with a team, not specifically game design. I'm not claiming to be the exception to the rule, but I am competent enough to take a stab at it more than most. Just because I haven't been modding videogames doesn't mean I'm not a competent game designer.

Also, Ooo, ow, you've pierced me in the heart! Woe is me, for Jon`C has indirectly insulted me with his words of wisdom!

As far as I know, the runtime has been virtually unchanged since version 2.


Learning linear algebra and taking a class in it are different. I was going to minor in mathematics until I realized that linear algebra II and abstract algebra would be for me, and none of the other math classes are easier


I thought you just said using a native wrapper would be faster? Or are you talking about the performance hits that result even from that?


I'm experimenting with writing a game engine right now, using C#. I'm using a simple task kernel (like that Enginuity article), hierarchical game states, and scripting through assembly loading (with fast calls from dynamically generated delegates using Reflection.Emit, actually similar to what C# 4.0 does for dynamic dispatch).

I'm nearing the point where I need to think about some kind of renderer. Writing my own is outside my experience level and outside my desired area of learning. Mogre, the Ogre3D wrapper, made me want to shoot myself. Irrlicht is a C++ renderer that looks nice. I figured I would use C++/CLI. Write an interface in C# for a renderer plugin and implement it in C++/CLI. I don't care that much about performance but I'm still going to design it to be as fast as I reasonably can. Do you have any advice on this?

YES

No, I mean if you want your game to work on the Xbox 360 you need to write your physics code in C#.


Use XNA for the renderer. It's really not that hard.

Ohhhhh


I've thought about it, and I'll consider it. Does it provide much of an abstraction vs managed DX? Thing is, I'm sure I could write it, but I don't really want to.

bored

the idea of a sidescrolling vehicle based shooter with a vehicle that jumps reminds me of something....


not sure what though...




ok i'm sold on the idea

How about a steam punk adventure/role playing game with time travel themes with scenes like a zeppelin and a bar???

How about we do that but with just the zeppelin and make a Hindenburg disaster simulator.