08-06-11 - A case for OnLive -

Reading Sanders' last post he brings up an intersting point that might actually be a case for OnLive.

I've long been a sceptic/critic of OnLive. Basically I think that taking standard games and running them over a network where you add 200 ms of latency and get no benefit is totally fucking retarded.

But a game that is custom-made for an OnLive / cloud is kind of interesting.

Particularly an MMO, because 1. latency isn't that big of a deal ; 2. they already have horrible latency so players are used to it, and 3. lots of players are in the same room at the same time, so you can share computer power on the server. Non-networked games where each player are in an independent world are much less compelling.

Also, if you're thinking of a running a Rage-like texture cache, doing local loads and recompresses is sort of like running an OnLive server on your local machine and serving yourself up compressed data - it adds latency, adds compression artifacts, and generally is very undesirable if it there was another choice.

In particular I imagine a use case like this which makes some sense to me :

MMO game
WoW style gameplay that's not super latency critical
cloud-style computing that dynamically puts more servers where needed
Huge number of players can be in the same room and there's no slow down
  (more servers just contribute to processing that area)

Non-GPU renderer; like maybe REYES or a ray tracer
Super high source content sizes stored on shared servers
  (how to generate massive amounts of source content is unknown)
Since you're just sending frames back to clients, render quality is unlimited
  just requires more servers

Could do real-time GI since you can put lots of servers on it
  (and the result is shared for lots of players so the cost is not prohibitive)
or just have massive pre-baked lightmaps with time-of-day variation
  (something like a spherical harmonic per texel, and store 24 of them, one for each hour)
  since back-end storage size is unlimited

the OnLive-style serving the images actually is an advantage in that scenario. In practice, no game company has the know-how to manage such complex servers. And the cost per player is too high. And being able to deliver massively more content just creates a big problem of how to create that content. etc.

1 comment:

LogicalError said...

I think, of all companies out there, Google would be in the best position to provide a service like this for people to built on.
They have the money, the server infrastructure, the bandwidth, server parks around the globe (which is necessary to reduce latency), the influence and they've already started to make baby steps into the gamespace lately..
They'd need to built server parks with GPU's for sure, and it'll be expensive (potentially too expensive), but it would be possible.. Also, didn't NVidia just announce some cloud products?

old rants