Questions About Cluster Computing

[JakobyGames]

This summer I intend to start a cluster with a combination of desktops I find on Craigslist and a some I build myself for a rendering rig as I do a lot of premiere, photoshop, and a BUNCH of 3D modeling/animation. I have been looking through videos and forums for a few weeks now with SO MANY QUESTIONS about cluster computing and had almost none of them adequately answered. Here are a few.

 

1.) Can multiple gpu's be passed through from separate systems? not for SLI, but for separate use, like CUDA.

2.) Can different nodes have different CPU vendors? (Ex: 3 with Intel, 2 with AMD)

3.) Cores or clocks? Should i rather get fx-6300's for more cores or pentium g3258's for overclocking/price (for multiple nodes)

4.) Proxmox or Xenserver? Xenserver DEFINITELY wins when it comes to simplicity and user friendliness, Proxmox seems like it has more room for advanced functionality and that sexy ability to access from the internet if forwarded correctly.

5.) With multiple cpus from diff vendors, in the VM, what would the cpu come up as (GenericXenCPU1?) and what drivers would I need?

 

one last question that doesn't really have to do with me, but more with everyone else.

7.) Why doesn't everyone use cluster computing? I see all these people who get a new rig and throw the old one in an old room to rot. Why not make them nodes and use them both?

 

Hope the forums can help, as this is my first non-giveaway-related post here even though ive been watching LMG since its creation.

[Embattled]

TBH you probably know more about this stuff it seems than 99.999% of people on this forum, trust yourself I guess?

[LiveTrooper]

1. I dont believe so.

2. I think so.

3. Why not both? But in term of modeling and rendering purposes, I would get more cores.

4. Xenserver for sure

5. I have no idea.

7. The idea scares people away I guess, plus I dont see many tutorials or troubleshooting video online. I guess they prefer the simplicity of a single computer. Although, as a graphic designer, 3d artist, and video editing guru, I completely understand the purpose. Why not just make a render farm? It would cost less money i would think.

[flip1012]

This seems like something that Linus needs to go over in one of his videos, either a guide or an "As Fast As Possible" either way soory that I don;t have an answer for any of those question, try PMing Linus about it, it's worth a shot...

[JakobyGames]

Thank you Trooper for the vague answers. In all seriousness that was more of an answer than ive gotten anywhere else. As for flip1012, i will DEFINITELY try that and it really sounds like a good fast as possible if not a full blown LTT series. "Scrapyard Wars 3: Cluster Computing" Sounds like something that might just work, thanks for all the help, except you embattled, you have failed me.

[LeightonPC]

That just gave me a scrapyard wars idea  

[MrSpaceman]

Thank you Trooper for the vague answers. In all seriousness that was more of an answer than ive gotten anywhere else. As for flip1012, i will DEFINITELY try that and it really sounds like a good fast as possible if not a full blown LTT series. "Scrapyard Wars 3: Cluster Computing" Sounds like something that might just work, thanks for all the help, except you embattled, you have failed me.

 

I had similar questions, being a designer I also work heavily in Illustrator, PS and Indesign, and was wondering if my Spare 3770 rig can handle any rendering/processing workload as its just lying idle.

[Ryoutarou97]

I don't really know too many specifics on the subject, but it intrigues me from a cost to performance standpoint. What kind of scaling is there? How do you think getting a lot of FX 8350's would compare to one or two xeons? Can you easily add more pc's to the system? Is there a limit? 

[trag1c]

This summer I intend to start a cluster with a combination of desktops I find on Craigslist and a some I build myself for a rendering rig as I do a lot of premiere, photoshop, and a BUNCH of 3D modeling/animation. I have been looking through videos and forums for a few weeks now with SO MANY QUESTIONS about cluster computing and had almost none of them adequately answered. Here are a few.

 

1.) Can multiple gpu's be passed through from separate systems? not for SLI, but for separate use, like CUDA.

2.) Can different nodes have different CPU vendors? (Ex: 3 with Intel, 2 with AMD)

3.) Cores or clocks? Should i rather get fx-6300's for more cores or pentium g3258's for overclocking/price (for multiple nodes)

4.) Proxmox or Xenserver? Xenserver DEFINITELY wins when it comes to simplicity and user friendliness, Proxmox seems like it has more room for advanced functionality and that sexy ability to access from the internet if forwarded correctly.

5.) With multiple cpus from diff vendors, in the VM, what would the cpu come up as (GenericXenCPU1?) and what drivers would I need?

 

one last question that doesn't really have to do with me, but more with everyone else.

7.) Why doesn't everyone use cluster computing? I see all these people who get a new rig and throw the old one in an old room to rot. Why not make them nodes and use them both?

 

Hope the forums can help, as this is my first non-giveaway-related post here even though ive been watching LMG since its creation.

1 I don't think so. The nodes don't share resources since they're more or less unaware of each other. (or at least should be unaware.)

2 Yes you can do what ever component configs you can get. However it is generally better to have similar parts that way 1 node doesn't lower the effectiveness of the cluster. e.g. a large complex task is given to a low end dual core rather than a monstrous 18 core.

3 Cores all the way. more cores = more simultaneous tasks.

4 not sure.

5 not sure.

7 (what happened to 6?) I think its more so complexity or unaware/educated.

[trag1c]

CG cookie render farm tutorial