Meet the rendering service powered by SONM platform
A month ago we announced turning Crypto-IaaS into a full-featured Cloud platform for startups and individuals. The key idea is to build a set of tools on the top of SONM infrastructure as an abstract layer that hides the complexity involved in provisioning and managing bare instances for various GPU-related computations.
We focus on 10 use cases:
- Rendering (movies, animation, ads, images);
- Machine Learning;
- Industrial 3D modelling (cars, planes, buildings, etc);
- On-the-fly video analysis & image processing;
- Financial Analysis;
- Energy Simulations (oil reservoir, gas, minerals, etc);
- Earth sciences (weather, climate, seismology, etc);
- Medical Imaging (CT/CAT scans);
- Life sciences (running molecular dynamics simulations and other high throughput docking studies);
- HPC Workloads (dense matrix computations).
Our plan is to gradually create a set of ‘one-click’ services and cover all these use cases. You can look at these services as “Uber for rendering”, “Uber for machine learning”, “Uber for Industrial 3D modelling” and so on, since we also collect devices (GPUs) on one side of the platform, and provide a simple service for ordering these devices on the client side.
In our “3 years strategy” blogpost we promised to start creating this set of ‘one-click’ tools with the Video Rendering service (video and image rendering for freelancers and small studios).
Meet the Raysrender
As mentioned in our previous update, we have already developed this service and launched it in the closed Alpha (starting from April 15, a number of video production studios and freelancers are testing this alpha version and providing us with feedback).
We named it “Raysrender” (or Rays) and created a logo that simultaneously looks like an axis of 3D coordinates and a clock:
The service is aimed to compete with traditional rendering farms, which are currently used by the majority of small studios and freelancers.
Nowadays, the standards of this market are:
- Typical rendering farm has 30-500 servers with a CPU. Some also have tens of GPU servers. Usually, you rent a device (or several devices) with a certain CPU/GPU configuration and your job is performed on these resources.
- To get started, you need to download and install special software created by this farm or get remote access to a rented device.
- Rendering farms usually offer several plans to choose from. More expensive plans have higher priority, so if you have a low-priority plan, and all servers are scheduled for several weeks in advance, then you have to wait in a queue.
- Some farms also require an account top-up in the range of $500 – $10,000
We decided to disrupt this concept and offer studios and freelancers a new approach:
1. No software to download and hardware to configure
All actions – from uploading the source file to downloading the outputs – can be done through the web interface.
Our smart orchestrator will automatically analyze the source file and select the most suitable servers from all available on the SONM platform. If the task is not resource-intensive, the orchestrator rents the Nvidia 1060 or 1070. If the task requires more than 8Gb vRAM, the orchestrator takes this into account and rents 1080 Ti.
2. No priorities, no queues
No waiting necessary – we operate massive resources of all GPUs available on the SONM platform, so in most cases, our smart orchestrator will allocate your task to dozens of GPUs.
Renting 10 servers for 1 rendering hour is equal to renting 60 servers for 10 rendering minutes, but the rendering time in the second case is significantly reduced. Therefore, we decided to create a ‘Justice Algorithm’ that always tries to complete any task as soon as possible, in order to free up resources for the next one.
This means that we allocate all available servers to your task. Even if hundreds are free, you get them all, but the price of your rendering will not change.
3. Single price instead of multiple fares
No matter 1 or 100 servers performed your task, you pay only for the rendering hours. Moreover, we decided to enter the market with an unprecedented low price per rendering hour (here is a 137 OctaneBench comparison):
|Farm||Rendering hour price|
|Raysrender||$0.50 / hr|
|RenderStreet||~$2.35 / hr|
|RebusFarm||~$1.38 / hr|
|FoxRenderfarm||~$1.25 / hr|
|Render4You||~$1.18 / hr|
|GarageFarm||~$1.00 / hr|
How it works
To demonstrate the work of the service, we rendered the popular Gooseberry benchmark on the Rays closed alpha.
|Source:||Result (rendered on the Rays alpha):|
|– Blender file with 15 frames
– 2048 x 858 px
– 600 Sample Cycles
|– Rendering time: 1:29 hr
– Total price: $7.36 (~49 cents per frame)
Blendergrid says they rendered this animation within 2 hours (1:48 hr) at the price of $15 (~$1 per frame). If you want to check how long it takes your computer or favourite rendering farm to render this animation, you can download the .blend file and try.
Here is the process of rendering this file:
Initially, our smart orchestrator allocated 6 servers to this task in the first minutes of rendering, and then increased their number to 15 servers, in order to complete the task as soon as possible.
We have shot several screencasts of the Rays alpha version, so you can better understand the workflow:
1. Drag & drop source file
To manage the rendering process in most rendering farms you need to download special software or get remote access to the rented device. We decided to make it easier and allow clients to simply drag and drop the source file right in the browser.
2. Manage what to render
Studios and freelancers no longer need to prepare the source file for rendering – we recognize all the metadata automatically and you can select specific cameras, scenes or frames in the Rays interface and send them to render. This is similar to the settings for printing documents before sending to the printer.
3. Click “Render” and take a coffee
Rays customers no longer need to care about the hardware – all you need is to click rendering button and our smart orchestrator will process your render on automatically scaled GPU set.
You can take a coffee and just follow the progress of the task. You will see both the overall progress bar and the readiness of each frame.
4. Download outputs frame by frame
Most Rendering Farms allow you to download a ZIP archive with all rendered frames only after the task is completed. We decided to let customers download the first frame as soon as it is rendered and start post-production immediately.
How does this affect the token?
Customers will pay for this service in fiat money using PayPal or credit card. At the same time, we pay to suppliers in SNM tokens, in accordance with the number of performed computing (in this case – rendering) hours.
This means that the more customers Rays have, and the more actively they render, the higher will be the demand for the SNM token (i.e., we will have to buy it on the exchanges more and more).
We will regularly publish reports on the number of customers and the performed rendering hours.
What the future holds?
In the first version, the service will support the Blender 2.8 (Cycles) only, but we are actively working on vRay and RedShift. Adding some other popular rendering engines is planned for H2 2019.
The second important thing in our focus is to create a prediction tool that will calculate the time and cost of rendering before it starts.
This is also very similar to how Uber shows you the time and price of a ride before you call a taxi. Our system will work almost the same:
- At the time of uploading the source file, we will randomly select several frames (for example, one from the beginning of the video, the second from the middle and the third from the end).
- These frames will be rendered in a few seconds in low resolution with ~50 sample cycles.
- We will extrapolate the result of such express rendering to all videos, taking into account the higher resolution and the number of Samples, and will make an offer – for how many minutes and money the file will be rendered.
If the customer accepts an offer, the service guarantees that time and price will not change. Thus, the Rays client will know all the important information before rendering starts.
This should be an important change in the rendering market. About 5 years ago, a person calling a taxi could only guess how long a ride would take and how much it would cost. For this reason, many did not use a taxi.
The emergence of Uber-like services has made it possible to see the ride time and price in advance. It helps to decide whether to order a taxi or not. The emergence of such an option has significantly affected the transportation market, and we hope to do the same with the video rendering industry.
And the third important thing we planned for the second half of 2019 is the promotion of Rays. Our BizDev team plans to actively attract freelancers and small studios. In addition, we are planning a number of PR activities and will test some paid customer acquisition channels.
To expand the reach of high-end GPUs on the platform before launching the Rendering service, we announced elevated rewards (x2 higher than mining) for suppliers with Nvidia 1080 Ti / 2080 Ti.
The elevated reward will bring 1080 Ti owner at least $2 per day (the best 1080 Ti mining option brings about $1 per day).
Please find the details here: https://sonm.com/blog/elevated-supplier-rewards-program/