Will you have enough GPUs for all of us?

[coilbook]

Hi.
If octane cloud is priced well will you have enough GPUs for all of us. What if I rent 50 GPUs then how many will be left for others?
Thanks

[glimpse]

I believe OTOY developers have a possibility to leverade Amazon infrastructure under demand so with this in mind I'm pretty sure there will be more power than You would be able to use =)

[Goldorak]

There's no hard limit. We have thousands of GPUs already running on ORC today. We also have a great relation with AWS, and they are ready to provision more resources if ORC needs it.

[ristoraven]

There's no hard limit. We have thousands of GPUs already running on ORC today. We also have a great relation with AWS, and they are ready to provision more resources if ORC needs it.

So, I could do an animation in 18432 X 1536 cubic resolution, have 2000 frames scene, with PMC kernel and, say, 5000 samples, I could get all the frames back about how soon? Will it render all of the frames simultaneously? So, if one frame takes couple hours to cook (for arguments sake), I'll get 2000 frames within couple hours?

(I can't believe what I am asking right now and actually typing those words.. )

[coilbook]

i bet they will charge you per GPU you use per hour. I hope they wont do it since it will be too expensive
I hope they can give us plans per month Like 100 GPU plan unlimited use 100 dollars a moth or so. 300 GPU plan 250 per month I would sign up

[Goldorak]

i bet they will charge you per GPU you use per hour. I hope they wont do it since it will be too expensive
I hope they can give us plans per month Like 100 GPU plan unlimited use 100 dollars a moth or so. 300 GPU plan 250 per month I would sign up

For live rendering, the cost per GPU minute is what we charge on X.IO.

For ORC, latency is not going to be as important as a live X.IO stream, and results don't need to be processed at the same rate for all jobs or all users. If you need thousands of GPU hours (there are ~720 hours/month, and let's assume useful work happens across several GPUs), then the real question is how fast do those GPU hours need to be processed in parallel? Depending on whether that is months, weeks, days, hours or minutes, your costs will vary. The faster you need a job done, the more you might pay, just like shipping a package costs more if it is overnight vs. next week delivery.

The less urgent the render job, the easier it is to fit that work into the cheapest possible rendering slot between high priority jobs.

[Rikk The Gaijin]

For ORC, latency is not going to be as important as a live X.IO stream, and results don't need to be processed at the same rate for all jobs or all users. If you need thousands of GPU hours (there are ~720 hours/month, and let's assume useful work happens across several GPUs), then the real question is how fast do those GPU hours need to be processed in parallel? Depending on whether that is months, weeks, days, hours or minutes, your costs will vary. The faster you need a job done, the more you might pay, just like shipping a package costs more if it is overnight vs. next week delivery.

The less urgent the render job, the easier it is to fit that work into the cheapest possible rendering slot between high priority jobs.

This makes perfect sense. Now the real question is: how much?