PC desktop system recommendations for rendering with GPU renderers (like Thea)

advice
gpu
hardware
purchase
pc

#1

Hi everyone,

I’m trying to spec a new workstation, and am looking at Thea for my render engine with Sketchup. Does anyone have any recommendations for what kind of hardware I should be using? Like should I use Nvidia Quadro or Geforce? Is the most important thing the number of Cuda cores? I.e.: Geforce 1080 ti has 3840 Cuda cores, while Quadro M4000 has 256, but their price points are almost the same. There are probably a lot of variables, but is there a simple-ish answer for which GPU I should get? I keep hearing that Quadros are better for CAD work, but they don’t have as many Cuda cores.

Thanks for your brain power, I’m basically empty…

Best,
C


#2

I don’t use Thea, but I understand that it is a CPU renderer so the graphics card has no effect on final render times. If you look for fast renders, I would first try selecting a CPU with as many processor cores that you can afford. Quadros provide no benefit to applications like SketchUp that are not designed primarily to require them.


#3

Hi @cam,
As far as i know, workstations are slightly better for rendering etc. but extremely expensive comparing to desktop type of GPUs.

Click here to take a look at the video for brief understanding

https://www.youtube.com/watch?v=U5Npt1BSF04


Also if you look at the comparison at this site, you’ll see GTX 1080ti is better and newer comparing to Quadro M4000. (And that explains why M4000 has so low core count)

By the way 1080ti has 3584 Cores and not 3840


So maybe comparing GTX 1080ti to a Quadro P6000 would be more ‘fair’. But i don’t even want to look at prices :sweat_smile:


#4

First of all you need to figure out if the renderer is CPU, CPU+GPU or GPU only. If it’s CPU+GPU then a beast of a GPU may be limited by your slower CPU.
Second you need to confirm that the software is optimised for CUDA and not OPENCL or some other instruction set. This may influence Radeon vs GeForce decisions.
Third you need to know how much VRAM you might need - large scenes with lots of hi-res texctures can require more RAM so some users go straight to a TITAN for that reason.
Fourth, you need to decide if you want to render while using Sketchup at the same time - usually the Renderer will hog all GPU resources so that Sketchup runs slowly, if at all. In that case you may want a 2nd video card for rendering.
Finally you could find out if your renderer can address two or more GPUs at the same time.

As for Quadro vs GeForce, GeForce is better value for rendering. Some use Quadros because they seem to be more reliable but you can always underclock a GeForce to bring down temps. (buy a version with excellent cooling, rather than quietness- but be wary of the OC, TI’s and other “high spec” models as they often run hotter). Technically you could use an Nvidia Tesla but those are pricier still.

if your renderer supports network rendering, you may also want to consider a 2nd “slave” PC which is a cheap/basic PC with 2 or more video cards and a really good PSU. This can be better in the long term than upgrading your workstation (if your workstation is a laptop then rendering using that is not recommended).

Unlike CPUs, video cards get substantially faster each year or two, so you don’t generaly want to invest $1000s in a setup only to find next year’s $500 consumer models outpace it by 2x.


#5

there are the specs I am looking at:

Single Intel i7 6800k Extreme 6-core 3.4GHz, 64-bit CPU (total 6 CPU cores, 12 threads, 12MB cache)
Noctua air cooled heat sink
64GB DDR4 2400 memory (16GB x 4 dimm slots)
Samsung 500GB 850 Evo SSD drive x 1 (system)
2TB SATA III 7200rpm drive x 1 (data)
nVidia GTX1080ti GPU
Asus X99-A2 workstation mb
Dual onboard gigabit NIC
onboard 8-channel audio
onboard USB 2.0 port x 2
onboard USB 3.0 port x 10
onboard SATA 6.0 port x 8
onboard SATA 6.0 port x 2
24x DVD burner
Coolermaster ATX workstation black chassis
Corsair 750 watts quiet Power
Microsoft Windows 10 Pro 64-bit


#6

Thea has this feature called “Presto” that uses CPU and GPU at the same time.

https://www.thearender.com/site/index.php/features/engines/presto-gpu-cpu.html


#7

@cam

Thea supports CPU+GPU rendering. In Thea’s case you won’t have support for OpenCL with current version, though the it was announced more than a year ago that Thea already was able to render with OpenCL, for now, you should stick with Nvidia GPU.

In Thea, the GPU generally contributes more towards samples per minute than the CPU but that contribution is determined by how powerful they are. I built my rig around the GPU.

The two aspects you should pay attention to are:

  • Number of CUDA cores;
  • VRam amount.

Both of them are important but VRam is what determines how big a scene you can render. For that reason I have a Titan X with 12Gb which has a considerable amount of CUDA cores too.

With the same amount of VRam and CUDA cores, if I was to go for a Quadro at the time, it would have cost me 10x more.

The real difference between GTX and Quadro GPU’s, for what I know, is double precision that exists in Quadro. This is probably important for some CAD applications but it’s key for scientific applications that use GPU as the base computing power, it’s not important for Rendering.

I also read somewhere, when I was investigating the issue on my own, that Quadro CUDA cores were better than GTX CUDA Cores, but, for the price difference, I didn’t consider a Quadro.

Therefore, imho, GTX is your best bet, and I’d recommend one with the most amount of RAM, as the speed difference might not be that important, while the capability to render as big a job as you can is what makes all the difference.


#8

From Thea:

The recommended hardware (for using Thea Render) is a multicore processor, typically an Intel i7 with 8 Gb Ram.
If you plan to use Studio for rendering (studio is at 64-bit) for high-resolution images, then adding more RAM will help (16 Gb for example).

Regarding graphics card, a GTX card is being recommended (instead of a Quadro) - the more cores you have the speed will be better.

Please have a look also at our Benchmark page: https://www.thearender.com/benchmark/index.html
You can find here several machine configurations and their scores, so that based also on your budget will allow you to make the more optimal selection.

In case you need more information, you can contact us again and we will happily help you.


#9

• Passmark : High End Video Cards
• Passmark : High End CPUs
…• Passmark : CPUs - Single Thread Performance (important f. SU performance)


#10

It all depends on what you’re trying to render, but I would consider bigger RAM than the 16gb.

Thea is not a speedy render as it is mainly unbiased. It has a biased engine (BSD), that behaves similarly to Vray’s, but where it stands out is on the Unbiased engines that most users tend to favour (I never used BSD for production though I tried it, I don’t like it).

Unbiased engines are not fast, they are simple to use and deliver ultimate quality though. Thea is one of the fastest.

Every Thea engine works on the CPU so it’s understandable that a nice CPU is the recommended setting and a decent amount of RAM will allow you to render most projects. 16GB, is imho, small. If you intend on rendering exteriors with massive vegetation and 4k textures or multiple proxies and at high resolutions, you might be bottlenecked. There are a couple of very nice Thea features like Relight, or export channel passes to post edit in image editors or compositors, that will consume even more RAM.

Where Thea shines though is at Presto which supports both CPU and GPU.

Presto is also an unbiased engine, but it will get it’s speed mostly from the CPU+GPU collaboration. The GPU, as I said before and as you can look at the benchmarks, usually contributes with more samples per pass than the CPU. The Top GTX GPU is much cheaper than the Top CPU too. Also you can get more GPU’s inside a Desktop than CPU’s so it’s a better investment to think on GPU rendering than CPU, in my opinion.

However the GPU is also bottlenecked as you can only render jobs with the max amount of VRAM you have. If a GPU is out of VRAM it get’s out of the loop and on tougher jobs only the CPU will hold as it will be able to have 16Gb, 32, 64 or 128 and even more. No GPU can follow there.

That’s why, since I bought mine with 12Gb, I will never go much lower than that. A 1080Ti with 11Gb would be the only one I’d consider from the ones available, besides the costier Titan X.


#11

This is very similar to what my staff use (mine is a 6950x) so its a great all-round modelling & rendering PC and excellent value for money.

You probably wont run out of CPU headroom but you may just find the GPU rendering process makes sketchup modelling difficult as it becomes laggy when rendering, so a 2nd video card is the only way to solve that.

You may find the 500gb ssd becomes full after lots of modelling & rendering files clog it up.
If you can upgrade to a 1000w power supply then I would recommend that, and get a gold or platinum as they tend to be better built


#12

[quote=“AK_SAM, post:11, topic:45131”]If you can upgrade to a 1000w power supply then I would recommend that
[/quote]

really? for what?

rule of thumb: TDP of CPU + GPU(s) + 50-75% for the rest = 140 W + 250 W = 390 W + 75% = 683 W

running PSUs with low load decreases the efficiency = wastes energy.


#13

I’ve learned a bit through trial and error…and PSU reviews are usually awesomely detailed. I’ve had a few dead PSUs on PCs in my render farm as they get pretty stressed (would add that a good quality surge/spike protector is essential). I also prefer buying case, psu, monitor etc to last a long time whereas the cpu, mobo and gpu are more frequently upgraded.

PSUs total output falls as they get older & are used at high output levels. Temperature, dust and power input fluctuations also pay a part. (some of my rendering PCs begin to have coil whine as they age)

Most quality PSUs operate at MAX efficiency at somewhere around 50-70% of their max output.

For a minimal cost (eg $40 to upgrade from 750 to 1000w) you are futureproofed to add USB-C devices, extra hard drives, even another GPU (common for rendering PCs).

Finally, peak power demands can be higher than rated power, eg the 1080ti can be up to 295w peak draw (more still if overclocked). I think Titan X Pascal’s are slightly higher again.


#14

Everyone,

Thanks for the detailed and informative replies. I love how PC specs can make people have somewhat bitter arguments, but it’s really just about what you can afford vs what you need/want.

Cheers and best regards,
Cam


#15

I am slightly confused about the gpu and cpu rendering thing. Because I was told on another thread that my rendering was slowed down when I enabled “GPU acceleration” was because my graphics card couldn’t keep up with my CPU (750ti and i7 4790). However in my head it doesn’t make a lot of sense because when I enable GPU acceleration isn there not more raw power available? because when instructions are sent to both cpu and gpu I assume they don’t take turns but rather they execute the instructions simultaneously?

Also I am putting together a system for vray rendering now with a Ryzen 7 1700 and I am pairing it with a gtx 1050ti. It seems unless I get a GPU on the same level as the processor, then GPU acceleration “works” but it will then only render the same speed as just cpu rendering with no GPU acceleration?

Or does GPU acceleration mean only render with gpu?


#16

buy you a GeForce GTX 1070 and everything will be fine.


#17

Are you using Thea Render? It’s the only engine for Sketchup that I know of, that is capable of Hybrid GPU+CPU rendering. This is not either GPU or CPU is CPU+GPU.

If you have a Renderer that is capable of GPU rendering, it doesn’t mean it is also using CPU. If you GPU is weaker than your CPU then your GPU rendering will be slower than your CPU rendering.


#18

I use vray 3.4 okay so gpu acceleration is not really GPU “accelerated” but more like gpu rendering only


#19

In that case am saving much more money if I just buy a let say a gtx 1080ti and slot it into my current system than building and entire new one?


#20

If you’re using Vray GPU rendering you have to be aware:

1 - It’s probable that a GPU render is faster than a CPU render depending on how fast is your GPU and CPU. I have an i75820K CPU and a TitanX. The Titan X contributes more to a render job than the CPU. That is using Thea… I don’t know about vray.
2 - As Vray GPU engine is a recent development it might be unable of rendering with the same settings or quality as the CPU engine which is probably the most mature in the market. So I’d investigate and compare results first and only after start investing in GPU to replace CPU.
3 - The last thing to consider is that Thea, for instance, is also capable of multi GPU and multi CPU setups, I don’t know about Vray. The advantage is that if I need more computing power, I can buy a new GPU and put it in my system. I have space for 4 Titan X GPUs there, whilst I cannot insert any new CPUs in my motherboard. Thea doesn’t even require all GPUs to be equal. I can buy the best available at the time and insert it in the box. This makes my system scalable for years as it’s not the CPU that is doing most of the work and so I can probably multiply my rendering power +4x relativelly cheaper than buying whole new systems.