Rhino 6 increase render speed

rhino

(Liis) #1

Hi guys,
I am still working myself through the NVIDIA Quadro Test: Graphic Card Test

Since I am still looking for a new workstation, preferably tower here is my situation:
We mostly get 3D files exported from different CAD or visualization programms. Therefore they often come as mesh objects, no blocks with a file size of approx. 250 MB.
Furthermore we want to do more visualization projects in the near future. For quick ones we want to use the integrated renderer Raytrace. For the others we are thinking about v ray.
I already checked the hardware requirements for both products (the ones for Rhino are higher).
Since the workstation selling firms offer there products with different hardware, here is my question:
Which parameter could be reduced and Rhino, especially rendering would still be fast and fun?

Parameters:
Central Processor: Intel i7 4930K CPU with a clockrate 3.40GHz, 6 cores 12 logical processors
Memory: 8 GB or 16 GM RAM
NVIDIA Graphic Card Quadro P600 or P1000 or P4000
Hard Disc Capacity: 600 GB or 265GB.

Thanks in advance!
Liis


(Pascal Golay) #2

Hello - I would get 16 GB RAM at the very least these days - 32 better. I have a P4000 here and it is great… but, I did not have to pay the bill. I assume you mean GB on the hard disk capacity - I’d get the larger one.

-Pascal


(Nathan 'jesterKing' Letwory) #3

For rendering with Raytraced you want to pick an Nvidia GPU with as many compute cores as your budget can manage. I always like to check the wikipedia page for Nvidia GPUs. Here you an see that the P600 has 384 compute cores. That is quite a bit underpowered. The GTX 1060 I have has 1280 cores - more than 3 times more than the P600.

The P100 has 640 cores, and while almost double that of the P600 it is still half of what the GTX 1060 has.

The P4000 has indeed a good amount of cores, the good stuff for Raytraced.

Personally I like to have two GPUs in my machine - one to drive the display with, and the other dedicated to rendering, but when not having to use the machine I then like to use both cards at the same time for rendering. Up to you of course (:


(Liis) #4

You are right, Pascal, I meant GB. Thanks for telling! I just edited my text right now.


(Liis) #5

Hi Nathan,
when it comes to 2 GPUs: is it a configuration that I can make directly in Rhino, wether I use both GPUs for rendering, or just one? And if yes, can you show me how?
Did you put your PC together yourself? Or did you buy it as all-in-one solution?
Thank you!
Liis


#6

I can’t quite comprehend why nvidia cards are recommended here.looking at these cycles benchmarks tells me that a RX580 would be the better choice than a GTX1060 (which I have). In viewport performance even a RX460 is sometimes faster than my GTX1060. to read the benchmark easier you can just klick on the cards on the top to cross them out. the GTX1060 there is only the 3GB version to be fair but still, the RX580 is that much faster that I’d assume it is still faster than the GTX1060 6GB variant. It is also cheaper. judging from the comparison between the GTX1060-6GB and the RX460 (that I both run Rhino on in two different machines), I assume that the RX580 must be a much better performer when it comes to viewport performance as well.

so I don’t get it.


(Nathan 'jesterKing' Letwory) #7

Probably because you seem to be reading the benchmarks results wrong. When you sort the results from fastest to slowest you’ll see per scene that in almost all cases the Nvidia GPUs are faster than the AMD GPUs. If you want to discuss this in more detail then please do so in a separate thread.


#8

Why in a different thread when it belongs here?

That’s a completely useless statement, of course you can say that all GTX1070, GTX1080 and GTX1080Ti are faster than a RX580 - but they cost much more, double or even tripple as much money. RX580 against GTX1060 is a fair comparison, because they cost roughly the same, so would be a comparison of a Vega64 against a GTX1080 (if we compare raw compute power and gpu die size it would be more fair to compare the Vega64 to a GTX1080Ti).

The old and simple BMW scene is the only one where the GTX1060 is faster that the RX580, in all others the RX580 is noticably faster, in koro even twice as fast as the GTX1060.

Certainly not me who’s reading the graphs wrong.


#9

For the same money, or less, I seriously doubt that a Quadro P4000 would be anywhere near as fast as a GTX 1080, GTX 1080TI OR an equivalent AMD card. The new nVidia RDX 2080 will be out, and so will be whatever AMD brings to the plate.

This is interesting, from https://www.techpowerup.com/gpudb/2930/quadro-p4000

"Unlike the fully unlocked GeForce GTX 1080, which uses the same GPU but has all 2560 shaders enabled, NVIDIA has disabled some shading units on the Quadro P4000 to reach the product’s target shader count. It features 1792 shading units, 112 texture mapping units and 64 ROPs. "

From nVidia:
P 4000
GPU Memory 8 GB GDDR5
Memory Interface 256-bit
Memory Bandwidth Up to 243 GB/s
NVIDIA CUDA® Cores 1792

GTX 1080
CUDA Cores 2560
Graphics Clock (MHz) 1607
Processor Clock (MHz) 1733
Graphics Performance high-17362
Memory Clock 10 Gbps
Standard Memory Config 8 GB GDDR5X
Memory Interface GDDR5
Memory Interface Width 256-bit


#10

yes, for cycles a GTX is always faster compared to a quadro based on the same silicon. that is because the GTX cards reach higher clock speeds and sometimes because of some deactivated shaders on the quadro side (like mentioned in the article you linked). I guess nvidia does that so quadros draw less power, stay cooler and therefore tend to have a longer lifespan. the situation is similar with AMD, Radeon RX vs Radeon Pro.

What makes the quadro and pro cards faster in CAD (viewport performance) compared to their gaming counterparts based on the same GPU is the driver. The gaming card drivers just deactivate some hardware features.

So if you use a GPU based renderer like cycles, having two GPUs, one for display and one for cycles, like @nathanletwory mentioned is a very good solution. a quadro/pro card for the display to get the best viewport performance and a GTX/RX as cycles compute device would be the optimal solution, imo (assuming the mix won’t cause driver issues).


#11

yes, you can select the installed devices in your computer in rhino, this is how it look on my home system:


#12

Generally, CAD-specific cards they are expensive, slow cards without proper cooling. They are a throwback from when Autocad needed special drivers.

Aside from a GTX1080, and the “Quatro K2100” in my notebook, I have a EVGA GTX570/AR, which is warranted…wait for it…for life.


#13

I recommend the Nvidia GTX 1080 Ti graphics card. Available now and at a lower price than the newer RTX cards. Many vendors are offering sales on the 1080 Ti so you may be able to get one for less than $700. My 1080 Ti has been working very well over the last 3 months without problems. Its performance is better than the Quadro P4000 for rendering from the forums I have read.

Regards,
Terry.


#14

A 1080ti?

Sadly, they told me, that you need to put that in a box and mail it to me : )
I don’t know why, I don’t make up these rules : )


(Nathan 'jesterKing' Letwory) #15

Who they?


(Jeremy) #16

Hi Nathan,

Can I ask what combo you have? I was thinking of adding a GeForce alongside my Quadro for rendering but found internet posts that say you can only have one Nvidia driver in a machine (and it would have to be the GeForce one).


(Nathan 'jesterKing' Letwory) #17

On my machine RhinoCycles_ListDevices yields

We have 4 devices
----------
	Device 0: CPU > Intel Core i7-7700 CPU @ 3.60GHz > 0 | False | True | CPU
	Device 1: OPENCL_AMD Accelerated Parallel Processing_Radeon (TM) Pro WX 9100_03:00.0 > Radeon Pro WX 9100 > 0 | True | True | OpenCL
	Device 2: CUDA_GeForce GTX 1060 6GB_0000:04:00 > GeForce GTX 1060 6GB > 0 | False | True | CUDA
	Device 3: CUDA_GeForce GTX 760_0000:05:00 > GeForce GTX 760 > 1 | False | True | CUDA
----------

There is also a built-in Intel GPU (HD Graphics 630) that I sometimes use for testing as well. I have currently only one monitor attached, I believe it is on the WX 9100, but I tend to switch for testing.

I’m generally not too fuzzed about line AA quality, so I tend to actually quite often plug the monitor into the Intel GPU. It may be not the fastest with lots of linework, but it frees up the other GPUs for Raytraced work.

I have even used a multi-device setup with all of the devices listed by RhinoCycles in use at the same time. Not the most useful, since the slowest device dictates the speed, but fun nonetheless. That isn’t trivial to set, not even possible with the regular settings GUI :wink: But it is possible to quickly set for instance the two Nvidia GPUs as a combined render device - the speed will be that of roughly two GTX 760s - and with the monitor in either the WX 9100 or the HD Graphics 630 it is pretty smooth sailing.


(Jeremy) #18

Just tried first render with a 2080 and it was trivially slower than the same render with a P4000, notwithstanding its higher CUDA count. Don’t jump to any conclusions from this one-off trial - but don’t write the P4000 off yet either…

Jeremy

(and before you ask they told me on no account to put the 2080 in a box)


(jesterKing) #19

I haven’t built kernels for these cards yet, because I have not been able to test them. Now that @jeff has such a card I can build them and have them tested too. If all goes well we should have them available with hopefully the next SR.


(Nathan 'jesterKing' Letwory) #20

@jeremy5 since you appear to be the adventurous type, perhaps you’d like to test RTX kernels I just built?

Extract the attached file somewhere, then place the .cubin files in C:\Program Files\Rhino 6\Plug-ins\RhinoCycles\lib. Start Rhino, make sure the RTX is your render device and give it a go.

Let me know how things went.

cuda_rtx_kernels_raytraced_sm75.7z (1.3 MB)