Hi this is my first question. I'm new to Houdini.

I am setting up 2 PCs for my office. One will be my workstation, which I know needs a decent GPU. The other PC will only need to do basic things, but I do want to use it as a render/simulation node so I can do that on 2 machines at a time.

Does a render node “need” a dedicated GPU? I'll have a node-locked indie license, so I won't be launching the Houdini interface on the 2nd PC. Mantra render doesn't use GPU.. I have read that FLIP simulations etc can benefit from OpenCL on the GPU though, so perhaps it's worth buying a cheap one to speed things up a bit?

Thank you

Eh, I still find OpenCL to have basically a negligible increase in simulation speed. If it fails, it can clobber your entire simulation so I just leave it off to insure simulation completion. Especially if the simulation exceeds your GPU memory, which is easy to do on a GPU with only 4-8Gb of vRam.

For Houdini you want the fastest possible clock speed on a CPU. At least 4Ghz. Then you want as many CPUs as possible for a render/simulation machine. You need at least 32Gb. More is better.

Thanks Evinob

Yeh I looked up some tests of OpenCL, and it does look like not much help… It seems you would need a beast of a GPU and vRAM for it to add much speed.

My main workstation setup is likely to be using a Ryzen 7 1700X, which only clocks at 3.4GHz, but it's got 8 cores, so I figure it beats the equivalent Intel chip with all that multi-threading that Houdini likes so much. The GTX 1050Ti 4GB will do… Best I could find on the budget I have. 32GB RAM.

The 2nd PC which will be where my office admin guy works, has an old i7 4790 - 3.6GHz, and 16GB RAM. It will probably help a bit with rendering in that state… Maybe I'll get some more RAM for it… Yeh I really should…

Anyways. Question basically answered.
When I've got the new gear I'll try see how this whole HQueue thing works.

Did you see this?:
https://www.sidefx.com/tutorials/speed-up-fluid-simulation-using-opencl/ [www.sidefx.com]

Yeah I saw that. It is a shame the poster did not leave a copy of the scene so others could verify the findings.

Yes thanks, I also saw that. Average of 30%-ish increase or something…

It's hard to compare because it's so hardware specific. That GTX 1070 is a fairly powerful GPU… 20 Xeon cores @ 2.4GHz each is kind of epic if all the cores are used… but that's a fairly low clock speed per core. Meh.

I can't afford any kind of decent graphics card for the 2nd PC anyways. As long as renders will at least launch and not give me an error like “Your PC is lame and needs a GPU.”, then I'm as happy as a tortuous.

The only point to remember is that if you design your fx with OpenCL then you should simulate with openCL. The results are slightly different between non-openCL and OpenCL. Your render node should be set to use OpenCL CPU if it lacks a comparable/better GPU than your GUI workstation.

some results from my new machine using TITAN Xp and dual Xeons and pyro solver, no micro solvers involved

OpenCL on GPU - almost no GPU usage, high GPU ram usage, took 12:43
OpenCL on CPU- no GP usage, no GPU ram usage, took 14:25
no OpenCL - after 14:25 i had only 47% done (so its 1/2 slower compared to OpenCL)
OpenCL on GPU , all inputs for dopnet precached (cca 4min to precache all inputs) - almost no GP usage, high GPU ram usage, took 10:55