Multiple CPU Core Utilization

[pjknibbs]

[phoenix-it]

You guys forget that developers in big studios tend to setup their content for DirectX and XNA at the same time. XNA doesn't support DX10, at least not on the 360. That's probably one of the biggest reasons you're not seeing it more wide spread. Another reason being it raises some machines to the level of requiring a discrete video card to play the game.

TBH, I'm sick of the road apples dropping out of the big studios these days. Nothing has any replay value, not to mention being a near verbatim copy of most of the games in it's genre.

I feel sorry for kids today... Their have few choices outside of FPS X or FPS Y and the occasional RTS they throw in for "flavor". All of course not counting WoW clone #17 that hasn't failed yet.

[notaterran]

My computer, a Q6600 with a 4850 and 3GB RAM has been around since 2007. Once XR is released I will upgrade, but not sooner. No one needs to upgrade every year. Right now a 3960, a 680 and 16GB should last you a good 5 years until they feel obsolete.

[wwdragon]

*nods* My last major upgrade was a couple years ago. I think this configuration of my machine will last for another 5 or so.

[Bobucles]

What's easy to upgrade?

- Video Card
- Ram (up to motherboard cap)

What's hard to upgrade?
- CPU ( especially if it takes a new Motherboard)

If the game scales well with the easy upgrades, it's not a problem. If it caps out on the CPU, then it will be much more difficult to push the game world boundaries.

[Rabiator der II.]

[Cdaragorn]

Sadly, one of the other problems when choosing to use DirectX is that Windows XP, which is still extremely common, doesn't support anything past DX9 either.

What really fascinates me is why the discussion always revolves around DirectX. OpenGL is just as capable an API, and has many advantages over it, not the least of which being that instead of being arbitrarily restricted by what software you have (read OS), it is only restricted by what the physical architecture of your GPU is capable of.

Then there's this (yet again) discussion on multi-threaded programming. It's awesome that Egosoft is building a new engine to do that, but also amazing that even with that on the horizon, we still get people here that like to pretend they're somehow "behind the times".

Multi-threaded programming is an ongoing challenge in software engineering. Just because programs have managed to do it in the past does not mean we have somehow magically solved all the problems that are encountered when trying to do it. Any time you try to multi-thread a program, you have to face all those problems in a brand new light all over again. I'll be thankful if they are successful at making 2 concurrent threads. Any more than that will just be a nice bonus. Another problem to consider is that having too many threads in the same program can often cause just as many problems (if not more) as having too few threads. The fact is, the raw speed of each of your cores individually is still far more important than how many cores you have.

[Rabiator der II.]

[Zloth]

"The fact is, the raw speed of each of your cores individually is still far more important than how many cores you have."

That VERY much depends on the problem. If you can divide the calcuations into independent threads then multi-core can be very handy. For instance, figuring the out-of-sector calculations in the current X games sounds like something that might work nicely in multi-threads. Dividing up the player's sector into multiple threads, however, doesn't sound very easy at all.

[SuperG]

Developers in big studios don't use XNA. maybe to fast prototype.
Big studios have Full commersial Console devkits. And a bigstudio inhouse or Comersial big upperclass licensable game engine. With Crossplatform capabilities. And use lots of middleware.
And have a large codebase over time.
These devs already are using SMP.

Its the smaller devs or indie scene that aren't into SMP because there requerment are often low that on core sufice.

The bigger commercial Indie games engine follow often bit late.
like Egosoft latest engine.

I think SMP For X series wil help a lot. As this kind of game resource needs are bit CPU depending.
But has it simulaties are lot of entities and GPGPU uses could mean lot masive concurent computing could put to good use.

[Seyumi]

[SuperG]

It depends how they implement concurent computing.

SMP by give every task its own tread. So Multithreading is then limited
1 ) by number threads. doesn't scale wenn CPU get more cores then aviable tasks.
2 ) Tasks could depend on other task so idle threads due to depending task.

A very non-efficent way of Multitreading for goal for all multi Core load saturation. Because games exist out of few taskes and they are depending.

Valve and iNtel did some pioneers work on efficent mutithreading.

Using a late sync and lots most small executing bloks each task is devided into.

wich means wenn other task are depending as idle waiting on a task. the active task uses msot core by it selfs. The task in most needs would be filler the execution qeue for each core.
There are also fire and forget threads like sound etc wich arent that heavy. And won't scale up with fine grain multitreading.

so the best solution Valve conclude is a hybrid form.

this kind of SMP concurent computing software enginering needs a senior SMP capable programmer.

I don't know what egosoft does but Easy way is better then nothing.

there is also a third point. Some code is very depending like a complexe branching routine. wich is just difficult to multithread unless you have lot of complex independand branching heavy concurrent code blocks.

wich means some kind of games can use SMP very good while others can't.

It depends to the code mix. depending vs nondepending code ratio.

In short Efficent SMP isn't that easy for games.

[Rabiator der II.]

[SuperG]

I don't think it works that way.
First they set a mark of what ever mainstream hardware they aim for.
So making game use 6 core efficent means a dual core can't handle the game wel as it bogs down due to cpu overload.

Lower mainstream is dual core. It would not run well but it could be playable if they aim for this.
Quadcore uper mainstream. Well could aim for this. But this target could still set X:R as a heavy CPU depending game.
So scale up to high-end market of 6 core. Well this target aim means screwing mainstream. Force gamers to upgrade CPU.

Also a option Scalable AI. Don't have the core your AI is more limited then if you got the computing power. Problem with this is. Gamers won't have the same game experience.

Like as if my wingman act like it schould or fly with a dodo kamikaze. On a old gamerig

So what this means is. If they aim for mainstream 6 core will do this split up a 2 or 4 core load over 6 cores if programmed well with most core semi idle.
or use 2 to 4 cores load for the game. If each core is lot faster then mainstream cores they will be more idle.

Wich just would shift frame computing time from CPU to GPU. Because fast 6 core CPU do it job fast so GPU get more time to do its thing.


Well current game took singlecore as mainstream so dual triple quad six eight core just use one core for current x series games.

[half0]