Render farm for tight folks

[dean_dmoo]

After reading the xeon 2670 article I came across refurbished hp z620's with 64gb and dual processor for 1200€. Cheaper than building my own so decided to try one. Will hopefully have it next week. ☺

[Garryclarke]

I have a dual xeon E5 - 2670 v3 as my main workstation.
The company I work for buys Dell PC so it wasn't cheap (plus the IT guys tricked it out with a Titan X and all SSDs).

It's a very very nice machine, but surprisingly it's the quietest PC I've ever owned, you can hardly tell it's on even when it's rendering.

I had to turn the hyper threading off as I ended up with so many logical processors (48?) running that some software didn't like it. For example Max 2014, whenever I opened the Material Browser it would dump me to the desktop. Turns out that that part of Max wouldn't support that many and just exploded.

[muoto]

I'm about to get some new dell workstation. What's the matter with the number of cores and HT ? Is there a kind of limit that can be used by max/vray ? Should HT be turned off on the latest v4 xeons ?

[Morbid Angel]

I'm about to get some new dell workstation. What's the matter with the number of cores and HT ? Is there a kind of limit that can be used by max/vray ? Should HT be turned off on the latest v4 xeons ?

its more of a windows limit then vray limit I think. I believe chaos has a version of vray that works on more then 64 cores.

HT improves rendering performance up to 20% on older cpus, new ones probably same or better.

[mitviz]

this is also something am thinking about doing but is it better to have more slaves (maybe 2 or 3) and use them with backburner being that they would be slower for example or use those same slave pcs with Vray DR? i have one extra pc so far which is slower and i haven't found it really beneficial so far to use it with DR since my main and much newer pc will finish rendering much faster and the old one just seems to slow down things at times. would also know what is a decent recommendation for a 2-3 slave setups and at a good price here in Europe and where power isn't much of concern

[dean_dmoo]

Thats one of the reasons I decided to try the refurb dual xeon option above. At the office we have a rendernode which is fast and a few workstations that are regular, but using backburner per image with 4 pc's meant that the fast RN was underutilised. For example if we had 6 renders overnight a heavy shot might end up on the regular stations running the whole night whilst the RN was idle. Switched to BB with DR from then on.

At the beginning of the year I bought a new WS (intel i7 5820k) and was using the old WS (i7 3770) and a RN (amd fx8350) for DR. But really by the time the other stations kicked in and with their speeds it wasn't very efficient. The duel xeon is 30% quicker than those two pc's combined and lessons the network bandwidth as well. Tested it yesterday and its nice and fast, will give it a few projects to see how it goes. My get a 2nd one later in the year.

The other option is to go for a very fast WS, like Bobby and a few others on here, and not use DR. With very heavy scenes this is definitely a nicer option for lookdev as you dont have the network delays. However I like being able to send renders away and continue with my scenes on my WS. With a next round I think I'll do a combination, fast RN like the one I just got and a duel xeon for the workstation to speed that part up.

[Streetwise]

I echo Dean's comments. I like to be able to send jobs away to DR via BB. For me, I work on larger still images, so production-wise, when I have many variations of the same image, it makes sense to do it that way. Of course you have tradeoffs with administrative hassles with more machines, heat, electric, etc., but for me, a beefy workstation just wouldn't work. So I guess it all depends on your workflow. My workstation is an HP Z820 4 core machine (16 threads), and the farm has 5 nodes (3- Z800 X5670, 2 - i7 X3930) They all seem to be well suited and work good together.

[mitviz]

I echo Dean's comments. I like to be able to send jobs away to DR via BB. For me, I work on larger still images, so production-wise, when I have many variations of the same image, it makes sense to do it that way. Of course you have tradeoffs with administrative hassles with more machines, heat, electric, etc., but for me, a beefy workstation just wouldn't work. So I guess it all depends on your workflow. My workstation is an HP Z820 4 core machine (16 threads), and the farm has 5 nodes (3- Z800 X5670, 2 - i7 X3930) They all seem to be well suited and work good together.

was thinking its a either or thing with DR and BB, i know its off topic but would love to try DR through BB for the hell of it, been using BB for a week now or so since i started rendering a tonne more images these days, i used to use DR for a few things but had a tonne of issues which i didn't need back in the day so i left all that alone until now, can you point me to somewhere?

[dean_dmoo]

tbh with the new version DR works great for me. If BB works then its just a case of adding the other nodes in the dr dialog. I needed to use ip adress instead of pc names for it to work for me, but other than that it was simple.

[Sbrusse]

Yep DR through BB works great, that's how we queue all our stills to the farm for the last 6 years and keep our workstations free.
Our workstations are i7 as we keep all the juicy stuff (dual xeons) for the render farm. And if we need some more power on our workstations, we just spawn to a group of nodes.
Usually we send all the stills to queue on one master node that spawn on the other 50 nodes but if we need absolutely two renders at the same time, I made a script that splits the nodes in different groups so we can assign groups to different master nodes.

[Streetwise]

I made a script that splits the nodes in different groups so we can assign groups to different master nodes.

That sounds really useful! Is that available somewhere? Would you mind sharing?