Tweet
a little theoretical project.. i was musing over this earlier when i should have been doing something else. . what would be needed to get the current state-of- the art vray renders done in realtime, using currently available hardware?
since we are firmly in cloud cuckoo land anyway, here are some random calculations:
i want to be able to render a 1080p image that takes 5 hrs a frame on my machine, at 25 fps.
(5 hrs seemed a reasonable time to include the very cleanest brute force gi, lots of displacement and proxies etc)
my machine is a 4.2 ghz i7.
so to get 1 hr a frame id need 5 times the power.
to get 1 minute a frame id need 60 x 5 = 300 times the power.
to get 25 fps id need 60 x 5 x 25 = 7500 times the power.
thats one powerful PC! assuming i specced up a hugely expensive quad 8 core nehalem EX, i could get approximately 8x the performance of my current pc (assuming i could overclock it to 4.2 as well
so assuming that machine, id need 937 of them.. not cheap but hey im in lala land so i also have the money.
to get responsive realtime performance it would be no good having each machine working on one frame, as the input lag would be laughable. so id have to use distributed rendering.
current limit of 10 pcs would have to go, and a 1000 pc limit introduced. and very small bucket sizes. what kind of interconnect would we need? assuming a fast enough interconnect, would this be feasable? what other issues would there be with such a project? apart from the sheer ludicrousness of it of course..
since we are firmly in cloud cuckoo land anyway, here are some random calculations:
i want to be able to render a 1080p image that takes 5 hrs a frame on my machine, at 25 fps.
(5 hrs seemed a reasonable time to include the very cleanest brute force gi, lots of displacement and proxies etc)
my machine is a 4.2 ghz i7.
so to get 1 hr a frame id need 5 times the power.
to get 1 minute a frame id need 60 x 5 = 300 times the power.
to get 25 fps id need 60 x 5 x 25 = 7500 times the power.
thats one powerful PC! assuming i specced up a hugely expensive quad 8 core nehalem EX, i could get approximately 8x the performance of my current pc (assuming i could overclock it to 4.2 as well
so assuming that machine, id need 937 of them.. not cheap but hey im in lala land so i also have the money.
to get responsive realtime performance it would be no good having each machine working on one frame, as the input lag would be laughable. so id have to use distributed rendering.
current limit of 10 pcs would have to go, and a 1000 pc limit introduced. and very small bucket sizes. what kind of interconnect would we need? assuming a fast enough interconnect, would this be feasable? what other issues would there be with such a project? apart from the sheer ludicrousness of it of course..
Comment