Flame In The Cloud

If there was a better Flame cloud rendering solution integrated with something like Deadline, and you had some kind of SDS storage solution which could pull the media to and from the cloud only for the most minimal time needed to compete the render, then that would make a lot of sense using remote Flame artists with their own kit. Unfortunately, there is no pipeline rendering solution for Flame bar Burn which I don’t believe is available in the cloud yet and is a far cry from something like Deadline.

We’re using AWS instances with Deadline for Arnold. Shit crushes once you have everything configured for on demand.

License tokens can get expensive if artists act sloppy, don’t check their scenes, blah blah blah but the instance cost for us hasn’t been prohibitive at all.

In fact, it’s hard to imagine going back to any other scenario.

1 Like

Yup… 3D rendering is the killer app for cloud.

1 Like

If you have local licenses of Arnold on premises, after a few hoops you can port forward the local license server to the AWS spot instances - either over the AWS portal link, or VPN.

Then you skip needing the UBL license tokens and are only paying for instance cost (until you exceed the number of license in your local pool, then you are back on the UBL token train).

1 Like

just messaged @andymilkis and think this topic would make an amazing Logik Live video of a day in the life of AWS remote flame artist, ie making a aws computer and how one would go about starting it up and shutting it down, ie archiving, saving the cpu image for day two etc and seeing what the actual costs are.

5 Likes

Thanks! I’ll look into it and let you know.

Hi, wanted to dig this thread up as im curious as to the real world expense. I recently did a freelance job for a company that uses remote cloud flames in AWS. It was great, fast, responsive, scaled to my monitor without issues, not without any faults or lags ofcourse, but def WAY better than expected and more solid that most remote connections ive had with all the major vfx houses in LA. i could be an outlier here but i had a good experience with it. … i came across this thread looking for something else and wanted to add my 2c

4 Likes

One thing that is worth tracking if you know, what the user geography is vs. the cloud data center. Cross country latency can be bad and impact experience. One would suspect a cloud Flame on the same coast as the user to better than a LA remote Flame with an East Coast driver.

Cloud flames are easier to co-locate to the user than remote flames. And data center network is also likely better than heavily loaded facility network.

Back of the napkin math on a G5 from AWS, reserved 24/7, with usage 50 hours of week, $1000-$1200 a month plus $2,000 a month for Weka storage. So, all in, about $3k per month per Flame. Ish.

3 Likes

Based on those parameters (not sure what the specs of the G5 @ AWS is), how many months would you have to work for them, before they could give you a free machine you get to keep afterwards and they’re still ahead cost wise? 2 or 3 months? How many extra months to upgrade your broadband connection? 1 month? You stay 4 months and they throw in something else…

The most powerful workstation available right now, Lenovo P620 w/ ThreadRipper 5995WX, 128GB RAM, A6000 GPU is around $15k USD. Cloud just doesn’t make sense financially. Also, not clear if you can even get as powerful of an instance as the P620. In fact it seems the G5 instance is limited to 24GB of GPU ram, which is not adequate for our workload, which is why we put A6000 in all VFX machines at our studio.

3 Likes

Right, heavily dependent on the work you’re doing and how big a system you need. It seems like system prices would be anywhere between $3K - $25K.

Are you talking about FiC pricing or local workstation? I don’t see how you get to $25K in building a local workstation.

2 Likes

I was looking at local workstations.

Well, that P260 w/ 2x A6000, 2x 4TB M2 and a few other things is $24.7K right now. I recently built a Puget System that is about $20K. It’s easy to get past $20K with a MacPro. And that doesn’t consider any of the peripherals (quality GUI monitors, reference display, extra storage, etc), which quickly come out to another $10K+.

But there is plenty of folks who just run on a Mac Studio or something similar.

So my post above was more looking like, if you had a freelancer connect to do some medium level work, would you get him on a cloud instance or send him a Mac Studio and what point can he keep the Mac Studio and it’s still cheaper than the cloud instance of ball park performance.

Having a top of the line workstation is a different matter. But then you either can’t get those on AWS, or they would be insanely expensive.

useless to have 2x A6000 in a single machine for Flame. The whole idea of local framestore storage is ancient and needs to go. Central S+W only.

In my case the dual GPU is for 3D Redshift renders, though I have dual A5000. Cuts my render times in half.

1 Like

Thank you everyone for the great responses here! It’s been very helpful. I reached out to a few companies that help setup Flames on AWS. Unfortunately Gunpowder told me that they were too busy with several large clients to take on any new business.

However TrackIt and Arch Platform were very helpful. Arch is designed for mid to larger facilities so even though their technology looked cool, it wouldn’t make financial sense for a freelancer.

I also priced out rental systems to get an idea of the going rate for those. But as someone looking to do some occasion side gig work, that seems like a cumbersome solution.

Thanks for the in-depth run-down @Josh_Laurence . Out of curiosity, what are you doing about broadcast monitoring? NDI, some type of transcoding or simply nothing?

(Cue @ALan and his hatred of NDI :wink:)

3 Likes

Great thread, and yes, @Josh_Laurence in-depth post is pretty much the reality of things.

We at Rohtau were born on the cloud and have transitioned to Flame on AWS, the evolution from the beta stage to our current setup has been in credible, in less than a year we were running things in a very inexpensive way. We don’t use WEKA although I love their tech, that said, the data they move is encripted in their own file format so good luck getting out of that… for that reason we decided not to use it… and the crazy costs.

All in all, we are super happy, a few limitations worth accepting, a lot of new found freedom, and overall, my only note worth remembering of this post is that you are not buying the same thing when you work on the cloud than buying your machines, not by a long margin… like very long margin.

So it is about understanding you should compare apples with apples…

my 2 cents

1 Like

NDI is garbage. This is the new thing, but StreamBox is a joke company.

1 Like