Kahawai: High-Quality Mobile Gaming Using GPU Offload

According to the report the decision seems purly based on technical application restrictions at the moment. Hardware h.264 decoding was better for the test tablet cpu it introduced more latency so was dropped.

Hardware encoding was not possible due to the provided api's not being flexible enough atm.

This end to end software approach limited them to 720p, i assume hardware h.265 is even less mature / flexible and more punishing if implemented in software.
 
That's pretty neat actually. Really neat, lol. I hope this continues to develop well, this is promising at the least, I'm not sure how far this can scale for consumer use in the near future but time will tell.
 
AzBat said:
Found a YouTube from Duke University when I was reading this article....

http://www.geek.com/games/microsoft-cuts-cloud-game-streaming-bandwidth-by-over-80-1623534/


Compares conventional cloud gaming at 1 Mb/s between with Kahawai at 1 Mb/s. They used Doom 3 as an example.

Tommy McClain
Click to expand...

Really cool video, it gives you a good idea of what is possible. For consoles that could mean the difference between high and very high textures. Mmm how would parallax maps work using this?.....
 
iroboto said:
That's pretty neat actually. Really neat, lol. I hope this continues to develop well, this is promising at the least, I'm not sure how far this can scale for consumer use in the near future but time will tell.
Click to expand...
It is neat but it's also limited by the thin client device. That device may not actually be a thin client, it might be a beefy PC or a current gen console but what about smart TVs or cheap tablets? Although my TV actually runs a Linux variant OS, I doubt it's got much oomph under the hood.

So they haven't really reduced streaming bandwidth by 80% (why do I continue to believe internet article titles!?!), they've found a way to stream less which sounds a bit like their cloud ambition for XBO, but from the opposite perspective.
 
DSoup said:
It is neat but it's also limited by the thin client device. That device may not actually be a thin client, it might be a beefy PC or a current gen console but what about smart TVs or cheap tablets? Although my TV actually runs a Linux variant OS, I doubt it's got much oomph under the hood.

So they haven't really reduced streaming bandwidth by 80% (why do I continue to believe internet article titles!?!), they've found a way to stream less which sounds a bit like their cloud ambition for XBO, but from the opposite perspective.
Click to expand...

My understanding is that they stream the same amount as with it not being on (1MB both cases), but the client side fills in the additional details. Or the 80% means how much more bandwidth would be required to stream in the additional details that the client side is rendering.
 
iroboto said:
My understanding is that they stream the same amount as with it not being on (1MB both cases), but the client side fills in the additional details. Or the 80% means how much more bandwidth would be required to stream in the additional details that the client side is rendering.
Click to expand...
You can read the original paper here. They term it collaborative rendering but it's a specific implementation of realtime distributed computing. I'm not denying it's interesting, and it's quite clever how they they both integrated their tech directly into the idTech4 (Doom 3) engine and also got it working with Street Fighter IV by nothing more than intercepting API calls.

I think this last bit is actually more interesting because it offers a possible alternative solution than SLI/Crossfire if this work can be used to distribute effectively rendering to a whole bunch of hardware on the network. :yep2:
 
I don't understand what the advantage is compared to streaming.

For AI it made a lot of sense (high computation, low bandwidth requirements), but with 1MB/s you can easily stream graphics in HD. Might as well just run everything remote?
 
In a game with a long draw distance you could do the rendering for beyond the game's parallax barrier service side and stream it in.

Make the FOV a little wider than the game needs and you could pan around to accommodate for a few frames worth of latency on the camera.

then use the local power for the things in the foreground and overlay. Have an overlap area where you blend in.

An MMORPG where many players might be able to share the same server rendered distant area could see significant reductions in total flops/watts needed to give everyone the same high quality experience.

Damn, what a Kawaii idea. /scoff
 
MrFox said:
I don't understand what the advantage is compared to streaming.

For AI it made a lot of sense (high computation, low bandwidth requirements), but with 1MB/s you can easily stream graphics in HD. Might as well just run everything remote?
Click to expand...
A lot of things get research that never see fruition. And sometimes there are things that are researched that at first are nothing more than hobbies but explode into revolutions.

Like if we look at electricity. Started off as just a magicians trick. Personal computing was just a hackers hobby. Etc. this could very well go no where, or be entirely something. I watched this documentary on Netflix about the large hadron collider and these business/economics guys always ask about the ROI of the device. And the answer is simple there isn't one and when it was built it was never intended to have one. It was strictly built to learn, and maybe one day they will discover something that will lead to business and consumer application.
 
I fully agree with the importance of the research itself and it's interesting, but to be fair the LHC was never advertized as "New high-energy particle accelerator reduces our dependence on oil by 80%". It's just the journalist though...
 
This seems like it would work well with complementing technology.

http://arstechnica.com/gaming/2014/...deling-could-make-streaming-gaming-tolerable/

Project delorian was also achieved in doom 3, i wonder if these two will be merged to form a low bandwidth solution with low lag.

Ultimately having all the assets locally with processing means large enough initial downloads that negates some of the best reasons to stream such as instant on and very low storage and processing requirements.

You could try to do remote frame boosting to target fps for aaa titles but it seems a high infastructure cost for little gain,
 
If used with Numecent's Cloudpaging technology you wouldn't need to worry about downloads.

Tommy McClain
 
You must log in or register to reply here.
Back
Top