4 ms·

I have no expertise whatsoever in this area, but I‘m curious: Is there not any timesharing / cloud solution for this? The Hardware you mention would be taken ad

I have no expertise whatsoever in this area, but I‘m curious: Is there not any timesharing / cloud solution for this? The Hardware you mention would be taken advantage of 1/3 of a day at max, right?

davidivadavid 6y ago
I'd assume the time cost of uploading/downloading assets would kill all productivity in the most interactive workflows. Otherwise there are solutions for batch processing.
KaiserPro 6y ago
Now that is a good question! There are solutions, but they are not overly compelling from a cost perspective (yet) For a g4dn.8xlarge (32 cpus, 128 gigs of ram) plus one GPU its $1000 a month (or $355 if you're only doing 12 hours a day 5 days a week, no over night renders source: https://calculator.aws/#/createCalculator https://calculator.aws/#/createCalculator) Thats still $4k a year. seeing as you need a remove viewing solution too, so you need a machine to host the Keyboard Video and Mouse, its still not great. Compared to a hp z640 with 48 threads(E5-2673 V3), 128 gigs of ram plus nvidia 3070 is about $2200 - $2900 (https://www.bargainhardware.co.uk/hp-z840-tower-a-configure-to-order https://www.bargainhardware.co.uk/hp-z840-tower-a-configure-...) Where the cloud is useful is if you are hiring someone for a 3 month contract, or you need to render something fast, as 100 hours on one machine costs the same as 1 hour on a 100 machines.
- WanderPanda 6y ago
  Very interesting! I wouldn‘t have expected such a large difference considering the economies of scale (you probably need to also calculate a small cost for maintenance but that should be minor I guess). Maybe in the end it also comes down to the ergonomics of working with a remote software. If you would have a very good client I think that could abstract most of the limited latency and bandwidth away. Onboarding the data might be a hinderance but considering it is only done once (and also considering the services like aws snowcone) I think thats also minor. Given the name creative cloud I would expect adobe at some point to do the move and also provide the compute for their products (or rather services by now)
  - KaiserPro 6y ago
    > economies of scale Its interesting, there was a VFX company that was called Atomic Fiction, that worked on Looper, Sully and a few other big budget films. They were entirely cloud based. in about 2014 it did look like everything was going to the cloud. Upgrading a render farm takes mega bucks($1-4 million for a large company) so converting that to a pay-and-go solution looked interesting financially. A lot of high end "live" editing/colour correcting is done remotely, because the machines are too noisy to have in the same room. (obviously 10ms lag is trivial compared to 45 plus loss) I think the thing that stopped the cloud rollout was how revisions are made. A scene in a film typically will be priced based on the number of people hours the company thinks it'll take +x%. However a film _producer_ will demand n number of revisions to change things. Those are normally not charged for. So if you have your own compute, it only costs you staff and power. _but_ if you are pay and go, it also costs you CPU rental as well. I think you are right, I suspect that at some point either Amazon, or more likely google are going to start giving away software with CPU time. (something like https://www.zyncrender.com/ https://www.zyncrender.com/)

davidivadavid 6y ago

I'd assume the time cost of uploading/downloading assets would kill all productivity in the most interactive workflows. Otherwise there are solutions for batch processing.

KaiserPro 6y ago

Now that is a good question! There are solutions, but they are not overly compelling from a cost perspective (yet) For a g4dn.8xlarge (32 cpus, 128 gigs of ram) plus one GPU its $1000 a month (or $355 if you're only doing 12 hours a day 5 days a week, no over night renders source: https://calculator.aws/#/createCalculator https://calculator.aws/#/createCalculator) Thats still $4k a year. seeing as you need a remove viewing solution too, so you need a machine to host the Keyboard Video and Mouse, its still not great. Compared to a hp z640 with 48 threads(E5-2673 V3), 128 gigs of ram plus nvidia 3070 is about $2200 - $2900 (https://www.bargainhardware.co.uk/hp-z840-tower-a-configure-to-order https://www.bargainhardware.co.uk/hp-z840-tower-a-configure-...) Where the cloud is useful is if you are hiring someone for a 3 month contract, or you need to render something fast, as 100 hours on one machine costs the same as 1 hour on a 100 machines.

WanderPanda 6y ago
Very interesting! I wouldn‘t have expected such a large difference considering the economies of scale (you probably need to also calculate a small cost for maintenance but that should be minor I guess). Maybe in the end it also comes down to the ergonomics of working with a remote software. If you would have a very good client I think that could abstract most of the limited latency and bandwidth away. Onboarding the data might be a hinderance but considering it is only done once (and also considering the services like aws snowcone) I think thats also minor. Given the name creative cloud I would expect adobe at some point to do the move and also provide the compute for their products (or rather services by now)
- KaiserPro 6y ago
  > economies of scale Its interesting, there was a VFX company that was called Atomic Fiction, that worked on Looper, Sully and a few other big budget films. They were entirely cloud based. in about 2014 it did look like everything was going to the cloud. Upgrading a render farm takes mega bucks($1-4 million for a large company) so converting that to a pay-and-go solution looked interesting financially. A lot of high end "live" editing/colour correcting is done remotely, because the machines are too noisy to have in the same room. (obviously 10ms lag is trivial compared to 45 plus loss) I think the thing that stopped the cloud rollout was how revisions are made. A scene in a film typically will be priced based on the number of people hours the company thinks it'll take +x%. However a film _producer_ will demand n number of revisions to change things. Those are normally not charged for. So if you have your own compute, it only costs you staff and power. _but_ if you are pay and go, it also costs you CPU rental as well. I think you are right, I suspect that at some point either Amazon, or more likely google are going to start giving away software with CPU time. (something like https://www.zyncrender.com/ https://www.zyncrender.com/)