Render Server for Daz?
memoriiv
Posts: 64
I am working on a long term project with some others. I am the artist for this project.
But I have an AMD GPU and cannot afford a Nvidia GPU. Thus I render on a CPU
so we were wondering if there is an afforable render farm that we may be able to use. we have searched ourselves to no avail.
many thanks!
Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2025 Daz Productions Inc. All Rights Reserved.
Comments
http://www.jacktomalin.com/iray/
I haven't used it so I can't endorse it, but other people seem to like it.
There's also this https://rentrender.com/iray-render-farms/ .
honestly, you'd probably be better off putting the budget towards a cheap used gpu and system upgrades.
Depending on the cost of the render farm, you could easily pay for a gpu in the span of only a few frames.
As a for instance, if you used the Tomalin farm kitsumo lists, in 3hrs on the PAYGO, you've bought a tesla K20x(ebay) or p106-100(taobao).
On the monthly's, 3 months on the "standard" tier and you've got a tesla k40, a pair of k20x, a pair of p106-100 or a used/refurb gtx 1060.
The main problem you'll have with any render farm is their availability. If you time it wrong, you may not be able to render when you need to, or the job may be kicked down in priority depending on the tier you're paying for resulting in a delay.
And you can easily be kicked down if a higher paying customer comes along.
Another consideration is if you don't have comparable hardware to the farm. If you don't, you can't benchmark a frame to get an idea what it's going to take and figure your necessary budget, so you could easily under estimate and blow your budget very quickly.
Going back to Tomalin's system specs, he has too little system ram for almost anything i'd throw at it. I can easily hit 40GB of system ram and still fit within my 6GB render gpus(barely) and never exceed the 2GB of my primary video card. And that's without using any optimizations.
If i'm going full cpu, that's not even half enough the ram i need.
Lastly, it's your money, spend it how you see fit.
If the render farm makes more financial sense, or just physical space sense, for your situation go for it
Just because i have the space to put in a server rack and run a spaghetti mess of cables for my gpus, doesn't mean you have to do it.
I'll preface this with saying, yes, paying for a server isn't for everyone. The render server currently has a nice little group of people who do all seem to get a lot out of it.
A few points though, just to clarify
Correct, but the price of a 3 month sub, you have 3 months unlimited access to 4x1080Ti's.. so it's not really comparible in terms of price/performance.
It's true that priorities might mean you have to wait, but it's the only way really to offer a subscription like that at such a low price. I'm not sure people using mine are in such an absolute rush, they just want it rendered within a day or so. The servers usually quiet too, so wait times are usually less than a couple of hours.
Can't say I've ever had an issue with any render - even big scenes that dont fit in the GPU ram, fall back to the CPU fine... so:/
Finally, I had a user who had similar reservations, such as paying for a service when he could just save up.. then he actually looked up the price to buy a similar rig, and yea.. he's now a happy subscriber.
I would say, if it's render/render times that are holding anyone back, there are options out there to solve that problem.
Cheers
@jack Tomalin
Did I get it from the FAQ right, nobody, even you has access to the results which are rendered there?
Are there plans to upgrade to RTX GPUs, as soon as 4.11 is out?
The images I render took usualy around 8 hours on a GTX 1070 (resultion around 2800x2200). With 4 GTX 1080 TI this would be propably less than one hour, for each image. I think this is still fair use, right?
Jack, i just used your numbers as an example because they were easier for comparison purposes than trying to get into the math that some of the larger farms use. I really didn't want to get into explaining GHz-hrs, octbnchpts-hrs, node multipler costs, etc..
There's two issues here.
First, "unlimited access", not unlimited usage, big difference.
If i'm in possession of the hardware, a gpu in this case, i have unlimited usage, as long as i pay my electric bill.
I can, as i do now, run the card(s) ~24/7/365.
With your service, or any cloud based service, you add in layers of potential failure. My isp, your isp, data corruption between the two, your local electric, server failure on your end, higher priority jobs coming in, etc.
Second issue.
I wasn't doing price/performance, I was doing cost/benefit.
I'll freely admit that 4x1080ti will absolutely smash a single k20, k40 or a p106-100. On a 1 to 1 basis a 1080ti benches in at around 2.6x the performance.
That performance difference is irrelevant if i can't connect, or if there's 10k jobs in front of me, or a higher tier job comes in dropping me down the queue, etc.
The reason i did cost/benefit as opposed to price/performance is that the op doesn't currently do GPU rendering. As such, just about any GPU could potentially be a massive improvement.
This was more intended as a general statement and not specifically aimed at your service. I'm glad your startup hasn't had these issues, yet.
Personally i hope you're successful enough that you do have issues like this, or at least have to plan for them as the business grows.
Something got lost in translation here.
I wasn't talking about dropping to CPU rendering, i was referring to system ram(ddr2/3/4) requirements as it relates to GPU rendering.
The render would still fit in the GPU's VRAM and be GPU rendered, but the system ram would be exceeded and windows would switch to the page file for the necessary additional ram.
In my example, from some recent benchmarking, I could easily reach 40GB of system ram(ddr3) and yet still remain within the 6GB VRAM(GDDR5) of the GPU. Specifically 4.8GB vram used, per SMI.
If i scale that to your 11GB of VRAM, the system ram would be in excess of 80GB roughly. Your systems are 16GB short, which means windows will start using the page file and system performance will tank. Potentially BSODing or requiring a system restart.
Used $2250, new roughly $7k. I'm leaving out the water loop as the cost of that is too variable.
That's 3.75 years @ $50 a month of your service, or 93.75hrs(2.25hrs short of 4 days) at the paygo rate, to buy the same system used.
At purchase price to performance, you could get comparable gpu rendering for around $800-1600, the determining factor is either 6GB gpus($800), or 12GB($1600).
The caveats to that are that it'll blow your electric budget, at around 1200w/hr when you get to the full rig, as it'll be 12 GPUS@~100w+ per, take up a significant area, it'll also be loud with all those fans running and hope that the bios and o.s. or choice allows more than 8gpus
The advantage is, you don't have to buy it all at once.
I'll agree with this 1000%
The question is, What is the best option to solve that problem?
The answer will always depend on the individual.
For some, render farms are a perfect solution.
For others, such as myself, system upgrades make more sense.
For others, a mix of the two.
For others, change render engines.
etc, etc, etc.
Like tom said there's options, try em all and find what works for you.
peace
Only I can see the renders, but the data itself is all self-contained in a single cache file.. so I can't access it. I usually clear the cache file out every month.. so it's pretty secure.
Potentially I might...well, as it happens I have my work rig which has 3 2080Ti's.. so I'll either swap them over, or something. All depends on the server usage really.
Yea, the fair use thing is kinda fluid.. it's just to try and stop it getting bogged down.. hour long renders are fine, just try not to dump like 10 on there at once.
Just to be super clear - this isn't a business for me.. it's just a small service I offer to people who want it. I saw a lot of comments from people who were complaining about render times, and it seemed a shame to sit and watch the machine sit idle.. so just seemed like the obvious thing to do.
Totally agree it's down the individual - I'm like you.. it's not an issue to build pc's etc, but for a lot of people who aren't tech savvy, they don't want the hassle.. which is exactly what the server offers them, access to a decent rig without all the worry.
Unless your internet connection is VERY fast, upload as well as download, you need a local rendering solution, you need a Nvidia GPU with as much memory on it as you can stretch to. (The amount of memory on the card controls the size of the scene you can render before it falls back to CPU only)
I'd love to try Jack's Iray server out but the upload speed on my broadband is barely 256Kbps, it's like treacle, it would take ages to upload the scene. Fortunately for rendering I have 2x Asus GTX1080Ti GPUs so it flies, it's just a shame my artistic talent is so lacking...
Best Wishes
Steve.
Point of clarification on the Tesla K40 suggested above. I bought two of these form Ebay a few months ago, specifically the K40M. It is designed to go into a server case with large high-capacity cooling fans, not a home desktop system or e-GPU setup. According to Nvidia, these have to be configured to run in a server by someone who knows what they're doing. Mine are sitting in a box, because 1) I don't have a server and 2) I don't know how to configure them.
There is the K40C, which has been reported to work in a home desktop system, and is the same card as the K40M except with a built-in fan, however, the price of a used K40C is, on average, about double what used K40Ms go for.
Nvidia is also very tight-lipped about whether the K40M can be modified in any way to behave like a K40C, and I cannot find any detailed info at TechPowerUp.
Rather than throw money at a Quadro, the GTX series are typically less expensive for the same CUDA core count, although you do take a hit on the VRAM.
The K40 (2880 cores) is the same as a 780ti (2880 cores), however the K40 has 12GB of VRAM, whereas the 780ti only has 3GB. Cores stack, VRAM doesn't, at least with GTX models.
4x 780tis on an Amfeltec quad-GPU cluster is 11,520 CUDA cores, but still only 3GB VRAM.
However, the Titan Z is a fine alternative. It is essentially 2 6GB 780tis in one rather large housing (triple-width GPU). 5760 CUDA cores total with 6GB spread over 2 halves. This is not 12GB usable VRAM for rendering, as the marketing would imply, but it's intended for gaming, not rendering, so the memory optimizations are for games. It also has specific power and 12v requirements which you must meet.
The 1660s look nice for budget multi-GPU rendering, but the low core count and low VRAM (as opposed to my Titan XP and 1080ti that I'm used to) put me off of them. It's a step backwards for me, but a considerable step up if you're coming from below that. If I were just getting into it, and didn't already have what I have, I'd look at 8GB GPUs with 1500 CUDA cores for around $300 each. 4 of them would be 9000 CUDA cores. Still "only" 8GB of VRAM, and a $1200 total cost, but you can get one today and add another over time, probably for less money each time depending on how long you wait.
That's a murderous rendering rig for a hobbyist IMO, and easily worth the expense.