Hacker News new | past | comments | ask | show | jobs | submit login
Show HN: Rentaflop – Render your Blender projects without sacrificing quality
63 points by dasokol112358 on Sept 14, 2022 | hide | past | favorite | 55 comments
Hi HN,

My name is David Sokol, and I'm the founder of rentaflop (https://rentaflop.com). We're a crowdsourced render farm aimed at making Blender rendering fast and affordable.

If you've used Blender, then I'm sure you've experienced the pain of waiting around for your animations to render. You've probably even had to sacrifice the quality of your work to reduce your render times. I've been there too.

If you're like me, then you're also disappointed with the alternative solutions: spending thousands of dollars on graphics cards or using prohibitively expensive cloud render farms to get fast render times. Our solution to this dilemma is to leverage low opportunity cost hardware from around the world to allow Blender artists to render their projects quickly, affordably, and without compromising on quality. Since most graphics card owners aren't utilizing their hardware to do valuable work 24/7, we provide them with a way to make money without lifting a finger, while lowering the cost curve for 3D rendering.

We're currently doing a public beta. If you'd like to try us, check out our site (https://rentaflop.com) and render your Blender project quickly and affordably! If you're a graphics card owner who wants to help Blender artists while earning money, reach out to support@rentaflop.com and we'll help you get set up.

We posted about our private beta on HN a few weeks ago. If you'd like, you can check out the discussion here (https://news.ycombinator.com/item?id=32299674).

Please leave a comment below, we'd love to hear your thoughts :)




The big question is the pricing and whether it really does make sense to pay for your service instead of purchasing hardware.

Your pricing is $0.0019 per Octane hour. IIUC, that means that an hour of time on an RTX3070 - scoring 400 - costs $0.76. Cost of an RTX3070 right now is about $550 on Amazon and with a TDP of 220W and an average price (in the US) of $0.15/kWh, the total cost will be $550 + <hours> * $0.15 * 0.22.

Given that, the break even is 757h, which is unexpectedly high.

Do you think your pricing is sustainable?

EDIT: To be clear, I'm aware of the model of using "spare" GPUs, my calculations were just for a point of comparison. What I'm asking is whether, with such low pricing, the company will have enough income to be long-term profitable.


Thank you for your question! We believe our pricing is competitive and tends to be one of our strong points, along with our speed. We charge considerably less than most other rendering services. Our farm will also render your projects much faster than a single RTX 3070 can render them.

For artists/studios who do a ton of rendering, it may be worth it for them to buy up many graphics cards and create their own private render farm. However, this requires considerable up-front cost as well as a lot of technical expertise, which many simply don't have. For a lot of them, it's a much better deal to render with rentaflop and not have the headache and cost of managing hardware.


I want to emphasize something here. Artists in general are not technical. Many don’t have the knowledge and skills to build their own render farm. This is why services like yours exist. Don’t listen to people trying to work the economics of build vs buy when a studio doesn’t have the technical chops to run their own. If they did, they would. Some do. Most don’t. If you’re a Disney/Pixar or VFX shop you probably have it. If you are a freelance artist, you probably don’t.

Considering that Blender is being used more and more for high production shots, you’ll be happy services like this exist. At least until you release your feature.

Keep going. Folks who can build their own render farm, will, but the vast majority of artists won’t.


What I mean isn't whether the price is competitive, I believe it absolutely is, what I mean is can you run a profitable and scalable company with that kind of pricing?

Recruiting spare gamers' GPUs is one thing - those people bought the GPU anyway, so as long as you pay more than they pay for power, they're happy - but I imagine that's a relatively small pool of GPUs.

At some point, you'll run out of those people, so I guess what I'm asking is whether the price you charge now is high enough that investors could profitably build and run their own "mining" hardware , while you still get enough of a cut to run your service.


Ah, I understand now. I believe there are enough private GPU owners around to sustain and scale rentaflop for many years to come. For instance, did you know there were around 10 million GPUs mining the Ethereum cryptocurrency at its peak? Our current prices allow our hosts to earn ~10x what crypto mining pays whenever they're processing rendering tasks for rentaflop. There are millions more GPUs owned by gamers, AI hobbyists, businesses, etc. that we hope to leverage to make 3D rendering more accessible to artists. If we manage to attract even a tiny fraction of these, we stand a chance at becoming the fastest and most affordable render farm on the planet.


> Do you think your pricing is sustainable?

I think you're kind of missing the point. Most non-professional artists (which there's a lot of!) won't need to come close to the break even point. When I was doing modeling and speed painting in school, I probably used about 80 render hours a year on our farm. I never "went" professional, but I can imagine that smaller shops probably aren't spending that much time doing deliverable renders. I bet a lot of small shops with a small farm under the desk are going to sign up to be providers for this as well.

Not to mention, you're talking about one GPU :) That's very slow. You really need to consider that for complex scenes, having twelve GPUs running for five minutes is way more convenient than one GPU running for an hour. Now, suddenly, that changes the math significantly - you're talking about buying 4x gpus (and the "sleds" required to run them - another $600 or so).


The true cost of hardware is greater than the purchase price of the hardware. It's not feasible for everyone to buy hardware, esp. for intermittent load.


It seems like you missed the entire part where they're not buying 3070s...

People who already bought 3070s and would otherwise not be using them are providing the compute, so they don't need a realistic breakeven on an entire 3070. They essentially need to breakeven on electricity for the owner.

Also they're not charging .0019 per Octane hour, it's per _ Octanebench hour_. One GPU can provide more than 1 Octanebench hour of compute.

Compare their pricing to the rest of the market and you'll see it's not unreasonably cheap: https://garagefarm.net/blog/garagefarm-net-adds-gpu-renderin...


> Cost of an RTX3070 right now is about $550 on Amazon

The GPU market is beyond saturated, with prices crashing. There are predictions of new 3090's being $500 in 6 months as NVIDIA and their partners get desperate to offload 3000 series inventory to make room for 4000 series cards.

Everyone and their cousin has a gaming PC now and even if a service like this were to take off, they'd have no shortage of suppliers (and competitors) and thus prices will plunge.

I can't see this being a sustainable business model past 6-9 months from now.


This might have changed, but in my experience the big barrier to having a distributed render farm is transferring possibly many gigs of textures and resources (sims, point clouds, etc) to each render agent, not necessarily the render time. If you have to do this each time you hit render, it's a lot. If you keep the files online, where both the client and the render farm can see them, you're reinventing dropbox + all the complexity of a render farm on top of it.

Also for commercial production use many VFX houses have contracts that don't let them send the assets out of house.

This would be great for people doing styles that don't need a ton of assets tho, like procedural-heavy (geonodes) scenes.

But as a hobbyist, I could definitely use something like this, and I'll check it out!


A long time ago, I worked on Renderfarm.fi, which was like what you're proposing but free. I was just an extremely junior developer working on the front end. The volunteer computing system itself was based on BOINC. I recall it was quite the effort to keep it running. If I remember right we rendered everything multiple times on different nodes to make sure that jobs weren't being faked.

There's no point to this post other than a trip down memory lane. I wish you the best with this project!


Is that the one which plugged the rendering hardware into space heaters to not waste the…wasted heat?

Was an awesome idea but I haven’t really been following that space for a long while.


No, Renderfarm.fi was completely based on volunteer computing, so there was no standardization.


For me the most obvious issue with it is that, unlike on a reputable render farm, the confidentiality of renders isn't in any way guaranteed, which is a full stop deal breaker for most professional work. Sure this might work for hobby projects but even on my hobby projects I want to retain control of my own IP. If I used you for professional work I would be breaking some of the terms of pretty much every contract I've ever been asked to sign as a creative. Not sure how you get around that unless you run blender in a black box process.


While we take measures to secure and bolster data privacy, it's possible for hosts to be malicious. As a result, you should never run workloads containing sensitive or confidential information on rentaflop. In the future, we may decide to offer a special service that renders your work only on rentaflop-owned hardware.

With that being said, there are plenty of Blender projects that don't have this requirement, and we aim to focus on these types of projects.


But David, almost all production work requires NDA and IP protection. This is something you should focus on if you want studio work to utilize your service.


It's not really clear how this works from their site but I think they're basically just sending your raw blend files to random users without any protection or guarantee, legal or technical, that either the blends or the renders will be deleted afterwards. That's going to go down well with my clients. "What the prototype renders were leaked by some random teenager from Birmingham? No, I've no idea how they got hold of them. Honest guv!"

Ah well time to add another clause the contract I use when subcontracting, something like "I affirm that I will not make use of Rentaflop during the fulfilment of this contract" should do it.


I think it's pretty clear that the service isn't for professionals. Possibly too niche to sustain a business.


I agree, right now. In the future that could change.


I love the fact that there is a program to actually pay GPU hosts that opt into the network. I'm planning to travel a bit this fall, and it would be a fun project to set up my system for this purpose whilst I'm away.

Mining bitcoin feels gross to me (no flame wars please)... but sharing GPUs with artists seems way more altruistic.


I'm glad you love it! If you'd like us to help get your system set up, please reach out to support@rentaflop.com


According to this page: https://portal.rentaflop.com/blog/hosting

It sounds like this is a cryptocurrency miner plus rendering farm, and participants will be paid in cryptocurrencies?


Many of our early hardware providers (hosts) are crypto miners, as they often have many graphics cards. As a result, we pay our hosts in crypto in order to attract them to our network. Our software, rentaflop miner, doesn't behave like a normal miner, however. Its goal is to take existing cryptocurrency miners who mine currencies like Ethereum and give their hardware more useful work such as rendering.


Are you experiencing significant performance penalties due to miner use of 1 lane PCIe risers?

I did some testing for a version of such a service in 2017 and it was determined the typical GPU miner configuration using x1 PCIe risers reduced I/O throughout and therefore performance of GPU hardware to the point of being practically useless for anything other than mining.

I have no experience with the characteristics of rendering, I/O, etc but it seems this also applies.


Useful work, fantastic !


>We're able to maintain fast rendering while charging ~50% less than most render farms by crowdsourcing. We leverage underutilized graphics cards from hardware owners across the globe to keep rentaflop fast and affordable.

Seti@home meets Blender.

I wonder if that would work for machine learning...


Crowdsourcing for machine learning is certainly possible, and it's one of the things we've thought about a lot. One issue that current Deep Learning algorithms, for instance, have with crowdsourcing is that they require periodic communication between devices to sync neural network weights. This is typically not an issue when devices are physically located in the same place with little network latency but becomes a bottleneck when the devices are scattered across the globe. With that being said, there are a few startups I'm aware of that are doing this sort of thing.


There is https://vast.ai which is a generic GPU crowdsourcing platform.

There are limitations that make it difficult to set up their machines as infrastructure however.


How do you deal with the fact that you have to transfer Blender projects to your crowdsourced computers ? (time to transfer, protection of private projects etc.)


We currently limit blend file size to 500MB and ensure our host hardware providers have the necessary bandwidth to handle transfers of this scale. We take measures to secure and bolster data privacy, but advise our users not to upload sensitive or confidential information that they can't allow others to see.


I'm trying to compile a list of all crowd-sourced compute platforms that aren't exclusively for crypto purposes. So far I have:

https://vast.ai

https://rentaflop.com

https://www.sheepit-renderfarm.com

Does anyone else know of any others? Thanks!


Thanks for the list, I ran across runpod.io yesterday (no experience with them though)


https://www.golem.network/


https://sonm.com/


Have you tried this? Do you know how well it works?


Sheepit was fun. Ran a cluster of machines a while ago for just rendering for other artists.


I was interested, but having to run an entire OS off thumb drive is a deal breaker for me. I'm naive in this area, but would it be that hard to implement it under WSL2?


My familiarity with Docker tells me this wouldn't be unreasonable on Linux, but Windows would probably be challenging if not impossible. I'm much more familiar with Linux so take that with a grain of salt.

I agree that I'd like to be able to do this without having to reboot and set things up before starting to mine/render.


It's on our roadmap to expand support to other systems such as Windows.

May I ask why using a thumb drive is a deal breaker? Typically the setup for our software only takes around 15 minutes to complete, and we can have someone from our support team help walk you through it.


Its having to shutdown, and having to boot into another OS. That's fine if you're targeting a mining rig, but for a home pc, you can't do anything else with the machine. Hence the deal breaker.


WSL2 only supports Nvidia and, specifically, CUDA for rendering... not OptiX.


Rent-a-flop? That already exists and is called Amazon Prime.


With the cost of electricity going in the direction it is, I don't think this will be worthwhile for anyone in the UK.


Are you also planning to offer other GPU-intensive tasks (e.g., Stable Diffusion)? Seems like an obvious expansion.


I am currently building this! You can try it out right now at https://computerender.com


Seems to work pretty well! Thanks!


It's on our long-term roadmap to offer other services such as scientific computing, monte carlo simulation, AI, etc. One issue that arises from Deep Learning tasks such as stable diffusion is that the various devices in the network need to communicate periodically, especially when training models, which presents problems when the devices are scattered across the globe.


Yeah, I can see that distributed generation of a single image would be challenging, but from what I've seen it's very common to generate dozens or hundreds of independent images, then have a human select the best ones.

There's no need for coordination in that case.


what.... does $0.0019 per octane hour pricing mean?

how much is a graphics card on offer currently? say RTX 3090ti or 3050ti or something else for example?

$0.0019 is not explanatory without giving examples.

can you improve your pricing page to include such details?

like say

$0.0019 in terms of graphics cards mean

RTX 3090 is $x/hour.

RTX 1660 is $x/hour.

titan x is $x/hour.


Thanks for pointing this out! Octane benchmark is a rendering industry standard for pricing hardware. It's useful because when someone launches a rendering job, it doesn't just run on one device, say an RTX 3090 or 1660, but rather on a whole host of devices.

Using this benchmark and our pricing of $.0019 per Octane benchmark hour, a 3090 would cost about $1.24/hr. The important thing to remember, and the reason we don't display our pricing like this, is that you aren't buying raw graphics card power from rentaflop or renting a single GPU. Instead, you're paying to have your Blender project rendered by dozens or even hundreds of graphics cards in parallel.


Price wise the offering is not great (though not horrible either). Without any discounts (longer commitments or volume) you can rent e.g., RTX3090s for 1.30$/hr at genesiscloud.com. Those GPUs are hosted in proper datacenters, have fast and stable internet connections, and non of the security issues mentioned in the other comments.

It seems to me that you are attempting to position yourself mostly via your pricing while you might have much better chances highlighting ease of use. Purely price driven users do have better alternatives but those driven not entirely by price but also convenience is who you want to capture.


ok. that makes sense that way but.... if i have a render job, can i know beforehand how much time/octane benchmark hour will it consume?


Great point. We're working on a render pricing estimator that allows you to upload a blend file and get an estimated price back. Until we have this, one way to see the estimated price of your render job is to upload your blend but cut the animation short. For instance, if your animation is 1000 frames, render the first 100 and extrapolate the price by 10x for a rough estimate. Then you can later render the rest by setting the animation in the blend file to only be frames 101-1000.


Sweet, maybe I can finally render volumetric clouds in Cycles now!


rentaflop.com is a terrific domain name. with out jumping the gun, have you thought of expanding to services not related to blender?




Consider applying for YC's Fall 2025 batch! Applications are open till Aug 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: