Hi,

I have a 6 core 12 thread Intel machine and run poser 11 pro. It's great. That said, when I render, with Firefly, it buries the cores. Granted it's still pretty quick, but animations are easily overnight affairs.

AMD just released a 16 core 32 thread wonder chip. Supposedly unbelievable as a rendering engine. Intel has too (soon), but twice the price. (18 cores for $2k, and not released yet. Faster for sure, but...)

My question: What's better for rendering? Cores/ Threads CPU? GPU?

It becomes murky to me when people use Superfly, because they can use their Video Card with tons of cores. I have a fairly decent card, lots of cores, but I don't see much difference in speed when rendering high quality, though you do get to keep using your computer as the CPU isn't buried.

Anyone have one of these Ryzen Threadrippers yet?

I'm feeling the need for speed.

Cores x clockspeed.

The Intel 18-core is pretty pokey on clock speed. And the price....

The core i3/i5/i7 processors cannot run in parallel, i.e., you can't run two i7s on a motherboard. As far as I know, the same applies to the AMD processors (corrections very welcome). I would be especially interested if someone has dual Ryzens operational; that would have great potential, if Ryzens can run in parallel.

GPU rendering is good for ^some^ rendering tasks: portraiture, for example. But if you want to render a large enviro set with lots of geometry and complex materials, it will fail. GPU rendering is limited by the video cards' onboard VRAM. Also some render engines need AMD GPUs, but others need nVidia, so you have to choose.

The relatively cheap way to get rendering performance is to use a workstation/server motherboard and ^two^ Xeon processors. Used X5690s are about $140 each; each X5690 gives you 24 threads @ 3.46GHz (3.73GHz turbo). You'd need a motherboard with dual 1366 CPU sockets. While you might -at considerable cost- get a new processor which will outrun ^a^ X5690 Xeon, it isn't easy to outrun ^two^ X5690 Xeons.

RAM only affects the render in a pass/fail manner; if you have enough, the render succeeds, and if you don't have enough, the render fails.
So, installing lots of RAM is good, because it expands your capacity for larger scenes and larger buckets, but it doesn't speed the rendering.

Hard drive speed will have minimal effect on render speed.

If you want render speed, you want cores x clockspeed.

"My question: What's better for rendering? Cores/ Threads CPU? GPU?

It becomes murky to me when people use Superfly, because they can use their Video Card with tons of cores. I have a fairly decent card, lots of cores, but I don't see much difference in speed when rendering high quality, though you do get to keep using your computer as the CPU isn't buried."

For me when rendering my GPU is a lot faster at it than using CPU, but from my understanding this is different depending on graphic card and CPU etc. I don't have a fast computer, but an ok graphic card. But as Seachnasaigh said, you can run into problems with the GPU once the memory is filled. Meaning textures wont load and so forth. So I just switch to CPU when thats the case. Luckily they have started making graphic cards with a lot of memory, but as far as i know they are still pretty expensive, but guess in a few years they will be cheap.

Regarding Superfly and speed, its definitely not the fastest render out there, but there are some things you can do to speed it up.

These are my settings for Superfly which works well for Preview renders that are decent quality and support glass (Reflections and refraction)

If you need high quality simply increase the pixel samples to 40+ something, just leave everything else as it is. You can check "Caustics" if needed.

Second thing you should do, is always remove the default ground plain (Turn it invisible), did some test on this trying to render a single model and but removing this it decreased the render time by 70% or something. Why the ground plain is there by default is not really beneficial I think and have no clue why they would do that in the first place. But my guess is that most people just leave it there and build their scene inside it. But what SmithMicro have done, is pretty much just to add a huge object that need to be rendered and even though its grey it will take a lot of time to do especially when you start adding your own objects as well, also a good idea is to always render with an HDRI map in the background so you get something to reflect in metals, glass etc. The ground plain is in the way of that as well :). So only render things that need to be in the scene

Thank you both very much. Both very helpful. The dual xeons sound tempting. 48 threads is better than 32 ;) according to my math brain. I'm going to price a used dual xeon system and see what I can find.

Many thanks.

Get on eBay, search "2x X5690".

Other H/T hex-core Xeons:

Oh, that search will bring up a boatload of server blades. Try "2x X5690 T7500" and "2x X5690 z800".

T7500 is the Dell full size desktop workstation chassis; z800 is the Hewlett-Packard.

Hey thanks yet again! I was looking for the xeon chip you mentioned and came up with very little. This will help a lot.

If you mean that you had previously searched for the Xeon processor but wanted workstations, then yes, search using the two models of workstations, as in my previous post.

If you want bare processors, just search X5690. Gobs of them, looking for a good home.

There ^are^ other workstation models which use these processors -Cameron is a Boxx 8520- but these would rarely be seen for sale.

The midtower chassis Dell T5500 and Hewlett-Packard z600 are common (Urania is a z600), but be aware that these will generally be limited to "only" 48GB and one or possibly two video card slots.

Thanks. I will search for these, but what is wrong with 48gb? I assume you are talking memory and that seems like quite a bit to me. And why would I care if it has 1 or two video card slots? Isn't the point of buying these to use the multiple cores in the xeon chips? Wouldn't any single slot video card be fine as these boxes would be used for rendering in a queue?

putrdude posted at 2:14PM Tue, 15 August 2017 - #4312229

Thanks. I will search for these, but what is wrong with 48gb? I assume you are talking memory and that seems like quite a bit to me

I've exceeded 48GB when rendering. But then, I'm notorious for making massive scenes. As long as ~your~ usage will not exceed 48GB, you'll be fine. Just want you to be aware of the expansion limitations of the smaller midtowers (T5500, z600) as compared to their larger full tower sisters (T7500, z800).

putrdude posted at 2:14PM Tue, 15 August 2017 - #4312229

And why would I care if it has 1 or two video card slots? Isn't the point of buying these to use the multiple cores in the xeon chips?

Like you, CPU (processor) rendering would be ~my~ priority, but others may also want the option of GPU rendering, and that requires available PCIe x16 slots, and 6/8 pin power supply plugs.

Wouldn't any single slot video card be fine as these boxes would be used for rendering in a queue?

Yes. If you intend to use the "new" machines only as render slaves, I'd go for the 2x X5650 C1100 blades w 72GB RAM for $351
C1100

Hi there

Personally I would invest money to better GPU,with GPU I can render lot faster than with any CPU available at moment in SuperFly,but this depends on scene etc but I do sometimes very large scenes with several props and several figures and I still didn't have issue with VRAM,my current GPUs are GTX1080Ti with two GTX1080 and rendering is really fast,I didn't tried to do animation and due this I can't comment

Getting older Xeons is good way how to get some performance but you need to understand,LGA1366 or X58 is old architecture and going back to X58 from yours 6 core/12 thread CPU will be downgrade or sidegrade in most,in C4D Benchmark dual Xeon 12 core/24 thread X5690 will score same as my i7-5960x with 4.3GHz OC which is 8 core/16 threads and ThreadRipper 16core/32 thread will demolish like my and dual Xeon setup in rendering,my current CPU will score in CineBench around 1600CB and ThreadRipper scores around 3000CB or 3300CB with 4.0GHz OC

Older generation of CPU have different IPC and due with every new architecture like is Haswell etc yours IPC is better and better,Cinebench and few others renderers scale pretty good with every arch etc

Good benchmark is Corona Renderer or CineBench which are most used renderers with V-RAY,there are benchmarks with all of these renderers where you can test yours CPU if its really slow or fast etc

At moment I'm still on X99 but I will be switching to X399 or ThreadRipper just I'm waiting on EK waterblock to be released or something along these lines,these chips can run hot when are OC and I'm looking get that CPU OC

Rendering with GPU and animations you need to understand,previous generation of GPU like has been Titan X or GTX980Ti they're been awesome GPU,but with Pascal rendering in SuperFly is lot faster in my case,I've owned GTX Titan X with GTX1080 that time and switching from GTX Titan X to another GTX 1080 my render speeds has been lot faster than with Titan X and GTX1080,right now have 3x GPUs and rendering is pretty fast,I usually render up to 45 samples for quick renders but usually I end up with 100 samples,Cycles or SuperFly never been optimized for Maxwell GPUs(GTX9xx and Titan X) and due this we are never seen fast render speeds with Maxwell GPUs

Due this I would be vary or careful with choosing next path there,AMD is planning to release their Radeon Pro SSG which should have on board 16GB VRAM and can use SSD up to 2TB as cache disk and in theory you can fit up to 2TB worth of assets in GPU which should be enough for most of the renders I think,just we need to ask SM about the supporting OpenCL and we can use AMD GPU for rendering,if we will be locked in Nvidia environment then we are screwed,same has been with Intel until AMD showed their CPU with affordable prices,Intel would charge us or give incremental upgrades with every arch

if you are on X99 then other option is get Xeon there,I've run 14 core/28 thread E5-2683v3 which has been awesome rendering CPU although if you do use SW which do use only single core you will be struggling but still has been better than my old X5670 with 4.2GHz

Hope this helps

Thanks,Jura