Physics card now, Biology & Chemistry to follow

[MSIC]

Really not sure what I think about all this....
I understand that dedicated units allow for quicker processing at specific tasks, but surely the benefit of dual cores (and then multi-cores) is to allow for exactly things like this whilst being of benefit in broader situations too.

We will end up with games developers producing games that will
a) be so specialist that only 1% of those who buy the game can get 'the experience' the developers intended
b) require all of us to spend lots of cash on devices that we wont be actually using in 95% of situations
c) piss off the average joe who just wants to play the game but will be so baffled by whether he has 'Device X' in his system (because he bought "a dell. It's a new one. It's got a flat screen and everything") or not he won't bother to buy it.

In principal i am all for the further development of technology, but I have strong reservations about the further compartmentalising / specialising that this is doing.

[schmunk]

The same challenges could be leveled at the 3DFX Voodoo 10 years ago. This amazing and exclusive '3D acceleration' technology is now commonplace and most often built into the motherboard with no add-in card needed (for the vast majority of pre-built systems, folks, not our little specialised build-your-own club).

This is a revolutionary technology, which may/may not have the same impact as 3D accelerators. However, it can't be long before AGEIA do a deal with Nvidia/ATI/somebody else to incorporate this technology directly into a graphics card or motherboard.

[MSIC]

I do see the comparison there, but it's still just 'one more thing'.
But years ago you got CPU's without FPU's and they then got incorporated into the CPUs, and my point is that with the whole multi-CPU thing going on, we might all benefit more from games developers concentrating on allocating one of these to physics.
So...how long before we have the chemistry & biology boards too? French? Religion & philosophy.....
Ok this is getting stupid sorry.

[Kumagoro]

what we want is an open source/free physics API like open GL, companies dont have to
pay to develop hardware for direct x do they?

Perhaps next is more likely to be an AI API.

[directhex]

the way it works is via a generic API, which suddently benefits if handed certain hardware. in the case of the physx, epic have been heavily relying on aegia's API for the physics in UT2007 - the code will run CPU-based on a normal CPU, or use the accelerator if available.

see also: software vs hardware opengl rendering

[DavidM]

Would be interesting if dual core cpus would devote one core to physics processing, in the absence of the physics card.

[schmunk]

Well, of course, we already have the Home Economics add-in...



(piccy is clicky)

[TheAnimus]

which is much more useful than a physics addon

The reason i get perplexed by the physics card is this. A 3dfx card with its own on board memory free's up oodles of CPU time spent accessing that memory because that memory can be faster too. Not to mention the highly opermised design for actual rendering maths.

But physics maths, in game.... why does it need any more than SSE? Conroe can do one per cycle at 128bit. On a dual core as standard too.

IMO its much better spend on something that has 70% of the performance in game, but is useful in everyday use, rather htan just gaming. Now if these where built into graphics cards as standard, then i'd be tempted, but the notion of second card which will not be able to help in every day life... no thanks.

[Kumagoro]

I expext it will be like when 3d cards came out as seperate or expanision cards.
Onboard the graphics card is the only thing that makes sense for mass market take up.

[herulach]

which is much more useful than a physics addon

The reason i get perplexed by the physics card is this. A 3dfx card with its own on board memory free's up oodles of CPU time spent accessing that memory because that memory can be faster too. Not to mention the highly opermised design for actual rendering maths.

But physics maths, in game.... why does it need any more than SSE? Conroe can do one per cycle at 128bit. On a dual core as standard too.

IMO its much better spend on something that has 70% of the performance in game, but is useful in everyday use, rather htan just gaming. Now if these where built into graphics cards as standard, then i'd be tempted, but the notion of second card which will not be able to help in every day life... no thanks.

It isnt 70% of the performance though, if conroe can do one op per cycle, then lets say an entire core is dedicated to physics, and it runs at 2.4 ghz, at the least we need to update each particle at 60hz, (otherwise stuff would jump across frames) so we're at 60kops/update, it takes at least 10 operations to calculate just gravity, so thats only 6000 particles you can track, assuming youre completly ignoring anyting other than gravity. (which you wont be, collisions will take ages to do) now think how few particles that is, especially seen as you can knock at least 2 orders of magnitude off that once you start wanting to model bits of concrete flying around and colliding, how the grass is affected by the shockwave etc.

[TheAnimus]

that assumes you want to model each partical indevidually, you don't need too, you can instead be cunning!

You can group particals, use flocking behavour, interlace particals on updates etc.

[herulach]

that assumes you want to model each partical indevidually, you don't need too, you can instead be cunning!

You can group particals, use flocking behavour, interlace particals on updates etc.

You can, for some particles, (smoke etc) however, you cant apply things like that to say, a stack of barrels falling over, or a craet being split into pieces etc.

[KnightWhoSaysNi]

The Sim City guy is developing an evolution-based game. That could make use of a biology accelerator card.