From: Kragen Sitaker (email@example.com)
Date: Fri Jun 30 2000 - 18:44:25 PDT
Eugene Leitl writes:
> Gregory Alan Bolcer writes:
> > The killer app *is and will always be* the same as it's always been:
> > Internet Gaming
> > It's a huge silicon production movie house with a Beowulf
> > pack of 20,000 Playstation III's doing real time awesome renders of
> > real-time in house Internet gaming spray and playing the
> > whole virtual world out to millions of clients, instead of doing all the
> > high-gpu powered rendering at the client side.
> Disagree. You don't need anything central for a game system, user's
> machines running a distributed simulation should do nicely.
The question is whether CPU power is cheaper in bulk or in little bits,
I think. For most things, it's cheaper to buy in little bits, but
cheaper to administer in bulk. For a few things, it's cheaper to buy
in bulk, too.
If I can pay a thousand dollars per gigaflops for a computer in my
living room, or a hundred dollars per gigaflops for a time-share in
some remote location, I might choose the time-share. At the moment,
though, the choice is more like a thousand dollars per gigaflops for
Beowulf stuff, or ten thousand for Cray. (Anyone who can tell me
whether I'm full of shit on that last figure, please?)
The nice thing about interactive timesharing is that you get more than
you pay for. No one person is on more than about 16 hours a day; the
average person will be on perhaps 1 or 2 hours a day. If it's not the
same 1 or 2 hours for everyone, then everyone gets more than they paid
> The realtime rendering and response aspects of games require a very fat
> pipe to the memory. No way you can farcast it over network, even if
> fiber is looking at you from the living room wall.
1600 (pixels/row) * 1200 (rows/frame) * 24 (bits/pixel) * 150 (frames/sec),
by my calculations, is only 6 912 000 000 bits per second --- 7 Gbps.
Fiber looking at you from the living room wall will do nicely for
today's displays, thanks. (I'll need a direct connection between the
network and the display, though --- my CPU is way too slow.)
Latency might be a more serious potential problem. But I can already
reach Web sites all over the West Coast of the US with 60-millisecond
or better round-trip latency, using primitive copper wire for the last
mile on my end (accounting for half of the latency!). How tight are
the realtime response requirements?
-- <firstname.lastname@example.org> Kragen Sitaker <http://www.pobox.com/~kragen/> The Internet stock bubble didn't burst on 1999-11-08. Hurrah! <URL:http://www.pobox.com/~kragen/bubble.html> The power didn't go out on 2000-01-01 either. :)
This archive was generated by hypermail 2b29 : Fri Jun 30 2000 - 18:47:09 PDT