Blender 2.83 on Oracle Cloud Infrastructure — Part 1
Oracle Cloud Infrastructure (OCI) allows you to run all sorts of workloads on the Oracle cloud. I’ve been working on some videos to help teach OCI principles and my weapon of choice is Blender to do the 3D modeling, animation and rendering. I even bought a pretty hefty gaming machine to help speed up all of the graphics calculations (I started on a MacBook pro and quickly outgrew that platform).
Even with all of that computing and GPU power, any animated video, even simplistic ones like mine, take alot of time to render. What would take days on my Mac still takes hours on my Windows 10 machine with 64GB ram, AMD Ryzen 9, 12 core processor at 3.8GHz with a GeForce RTX 2080 Ti. I saw an article on Blender Nation that mentioned render farms. Intrigued, I looked into them. I was disappointed at the pricing though. At the time I was working on a simple render animation that took my machine (the Mac at the time) about 4 hours to complete. The render farm I was considering wanted to charge me $300 to do that render. I am WAY too cheap to do that.
It occurred to me that maybe I could accomplish this on the Oracle cloud (full disclosure, I am an Oracle employee so I have access to all the toys). A quick search revealed that OCI does support special GPU shapes. A shape is simply a (virtualized) hardware configuration. A quick search of the web revealed this page, that shows which regions around the globe have the ability to provision a GPU shape. In North America, only the US-East (Ashburn) region supports shapes with GPUs (check the page often because it’s always changing as Oracle rolls out new capabilities). In Europe and the Middle East, all of the regions support GPU shapes! In Asia the regions in Japan have this capability too. It is not available in Latin America at this time, but there’s nothing stopping you from subscribing to any region on the planet! So no matter where you are, you have the ability to use the GPU instances.
One thing to note, I do not believe that GPU shapes are available on Test Drive accounts. A Test Drive account is a free Oracle Cloud Infrastructure account that you can use to get your hands on OCI and check it out. If. you have never touched OCI before, then I recommend that you get a free account (you can register here) and follow along with the Test Drive labs that I created in Youtube and on Github.
Along with the VM shapes, you can also choose from Bare Metal shapes. Bare Metal is Oracle dedicated hardware offering for the cloud. You would choose Bare Metal when performance is an absolute requirement. Here is a list of the GPU shapes available on Bare Metal.
For specifics on the GPU shapes, check out this page.
So it appeared that I might be able to use OCI as a personal render farm, but what would that cost me? I’m certainly not made of money anymore than most folks are. So I looked into the pricing for both the Virtual Machine and Bare Metal shapes.
Oracle is known for having clear pricing on their cloud offerings (ie no hidden fees like other cloud providers have) so when I found the pricing page and scrolled down to the GPU section I was shocked to see how cheap it was!
As you can see from the screenshot, the cost per hour for a simple VM.GPU2.1 instance is a paltry $1.275. If I decided to use the VM.GPU2.2 that price increases to $2.55 per hour (prices are per GPU).
Now it’s important to remember that unlike other shapes, the GPU shapes still have a cost when you stop the servers (which is akin to shutting them down). However, all billing stops if you terminate (read: delete) the server instance entirely! So if you could provision a Blender instance quickly (and I’ll show you how), use it and terminate it after use, you could, on demand, command some pretty hefty computing/GPU power! In the worst case scenario, if you forget to terminate your instance, your daily cost for a VM.GPU2.1 would come to $30.60 USD. So forgetting to turn them off when not in use is a factor. However, if you RUN a render farm, you’ll likely want to keep your servers running 24x7 to be able to serve your customers on demand.
Ah, but there’s the challenge. How do you provision one of these GPU instances quickly? How long does it take? I’ll go into that in my next article. Be prepared for some Bash scripting and some technical content. You won’t need to figure anything out on your own; I’ll provide you with the instructions and sample commands to get the job done. At the end. I’ll benchmark my OCI render times against my local machine and see how they compare.
Read Part 2.
