coilbook wrote:so basically 6 x 980 TI cards to render for one hour will cost you 10 dollars, That's too expensive and you do not get that many cards.
I would pay this much per hour if I was given like 200 GPUs where you can see fast results like 5 seconds per frame. We currently have 12 X 780TIs and even using just AO it still takes 2-3 minutes per frame. That means if we are given 800 octanebenchmarks per hour for $10 we can only render 20 - 30 frames
maybe one day cloud rendering will be cheap enough where you get 200-300 GPUs for as much as you want and you pay 500 bucks a month for service.
now all hopes to pascal GP100
Pascal may be a amazing and we'd love to see that double speed of local rendering.
ORC is not (at this time) intended to replace your local GPUs for single image rendering or small animations (< 50 fames) . But when you have thousands of frames, taking weeks to render even on 20x Titan X's, then ORC is really useful.
ORC works differently than current local/network rendering when it comes to animation jobs. Frames are rendered in parallel on ORC when possible, not linearly.
As a rule of thumb, you should send ORC 100+ frames per job. In this case, ORC is going to spread all frames across dozens, 100's, or even 1000s of GPUs to complete all frames in minutes, not hours.
Using your case as an example, if it takes an hour to render 20 frames of your scene on your local machine @ 800 OB, ORC will try render each of those 20 frames in parallel, and complete all 20 frames ~3 minutes.
This means ~16,000 OB would be applied in 3 mins, per your $10 bucket. At this rate, ORC will be rendering 400 frames/hr , or ~9 seconds/ frame at a cost of 50 cents per frame. To render more > 20 frames, you turn on recurring payment buckets (+$9.99 increments), and set a max monthly cap that fits your needs.
Let's say you decide to set this cap at $500/month. This would allow you to render 1000 frames in ~150 minutes. To be clear, we manage and queue many thousands of jobs per day at this rate, so ORC is optimized to deliver same day results at this volume, assuming peak usage.
Worst case, if your ORC budget is spent mid-render, before a frame reaches its final sample count, your partial render is paused and the filmbuffer is saved. The render can be resumed later on ORC by adding more credits. Or you could download the film buffer and complete the render locally without spending any further money on ORC.
Pause/resume rendering across both cloud and local renders (via 3.00 film buffer files) means any money spent on ORC render jobs can at minimum be used to augment and granularly complement local GPU rendering power.