(no subject)

[da]

We've got quite the storm right now. Just about pitch black, too.

Over lunch I was checking out new CPU specs and came across a new category of hardware: Physics cards. New games apparently "need" more than graphics acceleration, they should have physics acceleration too, since CPUs are no more optimized for physics calculations than they are for 3d rendering calculations. More busy stuff on the screen at a time, yay. Just what I need, surely. (Oh, wait, I'm not the target market. OK.)

Speaking of which, this video makes me happy.

Comments

[zanate.livejournal.com]

The big challenge facing physics hardware: unlike graphics acceleration, physics acceleration affects the entities within an environment. So, in any multi-player game, you either need every client coming up with the same physics results nomatter what their hardware, or you need a central server coming up with the results and the clients have a lot of latency between whatever lo-fi predictions they do and the hi-fi stuff from the server. Which would consume a lot of bandwidth, if the environment is sufficiently complex to benefit from physics acceleration. So it's not as nicely incremental as graphics hardware.

But, don't dismiss the stuff so fast. What we could benefit from is a nice common physics API... I've used a couple, for fun, but there's no solid-body physics equivalent of OpenGL, for instance. Though I bet the DirectX guys are looking at this issue..

Anyway, the fact is there's quite a large benefit to be had by specialised physics processing. It's a little harder to take advantage of. The demos I've seen are very, very impressive but at this moment in time it's an impressive solution still looking for a real problem.

[melted-snowball.livejournal.com]

Are there that many uses for this that would not be games? I mean, I could totally see scientists getting excited, but I imagine that the physics woul dbe too simple for their use.

[da-lj.livejournal.com]

protein folding? My understanding is that one of these cards can manage 10,000 particles...

[zanate.livejournal.com]

No, you're right, scientific interest would be small-- if that's what you mean by not being the target market, you're right. And from an engineering standpoint, maybe, just maybe there might be some way to apply our structural and fluid solver code to take advantage of such hardware, but I'm not sure it would be to as large a benefit unless this hardware were very flexible, and besides: our customers don't blink an eye about spending tens of thousands on clusters which said solvers already run on anyway.

I gotta think that if the benefit of the add-on's architecture for physics simulation calculation is as great as they claim, then they'll end up as an integral part of the design of any serious consumer VR system.

[melted-snowball.livejournal.com]

Yeah, I dunno. FPGA's can be a huge, huge win (we're mildly interested in applying them to the kinds of problems we work on).

This does push my, "but, dammit, the software folks get peanuts compared to the cost of the hardware" buttons.

[da-lj.livejournal.com]

"impressive solution still looking for a real problem"

Hm... desktop weather-prediction anyone? ;)

I wonder how this might help with protein-folding software.

What I (quickly) read was that one of these cards will handle 10K "particles". I bet there are real physics problems of some sort that they could simplify...

[kraig.livejournal.com]

Snow, mist, fog, and all other sorts of precipitation would seem natural.

Practical applications in modelling weather, obviously, but perhaps also things like how smoke spreads from a fire (maybe?).

I can't think of anything else off the top of my head, but I didn't give it much thought anyway. :-)

[kraig.livejournal.com]

*rereads*

"would seem natural"

...

pun NOT intended.

[da-lj.livejournal.com]

That's OK, I didn't see it as a pun. Which, for me, says something.

Exactly what, I'm probably too tired to figure out. ;)

[kraig.livejournal.com]

Crap, I'm guilty of laughing at my own joke while everybody else is stone-faced.

sigh.

[da-lj.livejournal.com]

Well, now we can all laugh together.

?

[kraig.livejournal.com]

You're laughing at me, not with me, I can tell!

*runs to his room and slams the door*

YOU JUST DON'T UNDERSTAND ME!!!


...

Sorry, channelling the 11 year old creature^Wchild with whom I share a roof.

[da-lj.livejournal.com]

Well, naturally. (No pun intended, implied, or acknowedged. Contents of bag may have settled).

...to quote somebody else today, "I've stopped hitting control-w as frequently since I've gotten a mac. I tend to close windows by accident."

[kraig.livejournal.com]

Windows, or applications? I occasionally miskey, and hit splat-q meaning to have hit splat-w. Fortunately, it's usually in Firefox, and they've finally trapped that when you have multiple tabs open (I don't believe it used to).

It gets really annoying when I try to close a tab though, and realize I'd used a hotkey to get to my Firefox desktop but not changed the focus - usually from IRC. That quits the current channel... sigh.

[da-lj.livejournal.com]

well, just now I hit splat-w instead of ctrl-w (when I really wanted to just delete the highlighted text) and closed a browser-tab, which was conveniently lj-compose, so there was still a draft.

d'oh.

[kraig.livejournal.com]

aiee.

Sometimes it's tempting to switch to Dvorak to avoid the splat-q/splat-w issue.