3D fractal rendering on 2990WX

Hi I am a 3D fractal artist, you can see examples of my work at www.fractalaesthetics.com.
I have recently built a new workstation to speed up my process (coming from a 4 core 4 thread CPU the difference is mind blowing) and to allow me to do animations, previously I could only do short tiny ones.

Specs

Threadripper 2990WX 3GHz
Custom loop with 360 rad
MSI - MEG X399 CREATION EATX TR4 Motherboard
G.Skill - Trident Z RGB 32GB (4 x 8GB) DDR4-3200 Memory
Intel - 760p Series 256GB M.2-2280 Solid State Drive
Seagate - Barracuda 3TB 3.5" 7200RPM
Radeon R9 290X Windforce
SeaSonic - PRIME Platinum 1000W 80+ Platinum

I use a free program called Mandelbulb 3d, which renders using ‘ray marching’ download here:

http://www.fractalforums.com/index.php?action=downloads;sa=view;down=70

The program is primitive despite being the most sophisticated 3D fractal rendering
software. It is 32bit, CPU bound and has many limitations. The 2990WX has dramatically sped things up however now I’m wondering how I can optimize my work flow. I’m running windows 10. The program has some options for allocating CPU resources under the ‘Internal’ tab on the right hand side. Here you can set Threadcount which I usually have at 60. You can set thread priority to one of four options, idle, lowest, low and normal. I always have it at normal, I’m not sure how the other options effect performance.
There is also an option to disable threadboost, not sure how that would effect things either.

I have a benchmark (renders a large still image), if you would like to run it then copy the following text and click ‘from clipboard’ in the ‘open’ tab at the top then hit ‘calculate 3D’

Mandelbulb3Dv18{
g…o5…3K…w…2E…qxJZZtFFuz4SjdfUIlXyj.lPSJAUPFzngD2cmcdYszQU37FU34Yyj
…NjFH0ZtPV02…Y.3L…EnAnAnAnIG/2…wD
…Uz6/…kr0…M.1/…2EJ1…V0…E8…2/kGp4i9.cD/6…ld3o.dkpXm1…k
./EnAnQD12…2…wz…U5…y1…sD…/…
.wHnAnQDfFPi.wQUXun1.pSlODwWyeIyhObkfCdjXgUAgKwwotvctz0/Ac9dyuEfBi5oR9dDcc9fVVyT
Luv9RfFOPqtWyeISM0.Si7ejU…oNB…kD.2…sD…G…
…oAnAt1…sD…zw1…4Z5SAxkg/3gwQuc9H5…x…k1.
…83iyz1…kz.wzzz1ky.A7.a…M2…EB…m…c3…EE…6/…YZ4…UJl5I
…U.eoZgzjdqZpz8R7vzPeRN/…1c…zzSi…6wyBIC4OWz1…2U.8.UQgk0.
.wUmc2beYz1.dA8E5Exwz0…36UHtjJA3/…cC3LRtKozslFH80817yD.oKcml0/dzP8Xc…
yLjq…UkKZMxkayDM0Dbqxm4Iz9…Kb3L.Euz4U.8.kzzzD…odIWOorVy98Sfl.M71…/…
3Q/2u9/7G/…Qd19UxLoz.EyEnhG57yD.orKJMn0dz9…6EU…zz/6FG1.QY2E7tzD/PX5…KD
v.KzTNie31kT6/qTQs5U./0…bT.026…ly/6EU…Mw5U./0…kqT.026…
…kz/6EU…U./0Izz/6FGHNyzHkqsFLxzZteKwj…
E6…/6…F2U…6…EGiFLNbJaQU.pPrJaQ…c…
…U6./…zj…

…A…3…9…oINiRKNmVISkJaQ.A5…
…6…U.E…kz1…wz…yD…cz9…
…b210gmVKIzXEUJCoWgRmziNaNaNaNavD…E.2…sz0…
…}
{Titel: bench1}

I can finish the main render in around 5:40 (5 mins 40 secs) and the ambient shadows in 1:13. To see how long the render took check out the ‘Infos’ tab on the right. I am running the 2990WX with precision boost overdrive enabled,
RAM at 3200MHz 14-14-14-34. While doing the main render it hovers around 3.45Ghz and stays at 67.75C. Using 70% of my 400W limit. The threadcount is set to 60 in mandelbulb and at normal priority. While calculating the ambient shadows things start to bounce around a lot more. Going from 3.5 to 4GHz rapidly. The program only uses 4GB of RAM.

When I run the benchmark on two copies of the program the time is increased to around 12 mins for both. However doing a quick animated benchmark I can render two copies simultaneously at almost the same speed as one!
If you’re interested in running the animation bench I can send it through.

If you can think of any way to optimize things it would be much appreciated. I’m an artist and lack a technical mind so I would love to know what resources the program is using. Not sure if cache is a big factor? Would also like to confirm that my bench results are typical.

Nice rig.
The updated software is in the new Forum.
Try the v2 2.14 which supports 64bits now and a lot more. Post some fresh benchmarks for us.

1 Like

The new version also supports network rendering. So you can add your old 4core for a speed up of 12% before network latencies. Maybe slower in the real world…

Also some OpenCL code, so your GPU may speed up too (again maybe slower in the real world…)

1 Like

Thanks mb3d! I’ve been playing around with mandelbulber this morning, its definitely faster and has some great features, I’ll get some benchmarks going comparing the two programs. OpenCL is great however when rendering my CPU and GPU utilisation is at around 0.1% (in task manager) and despite rendering quite quickly my PC is running at a whisper. In MSI afterburner I can see my GPU usage hits 100% but temps don’t get above 65C. The CPU seems to be pretty much
useless. I’m planning to upgrade my GPU next week, what works best with mandelbulber? I wonder if Nvidias new raytracing tech in the 2080 could be useful, what are your thoughts? If it can speed up mandelbulber significantly then it may be worth the price premium vs 1080.

First try to upgrade the graphics driver to the latest version.
Then see if you can disable OpenCL GPU rendering somewhere, since you have a monster CPU…
(From the docs: To enable OpenCL in Mandelbulber you need to open Preferences dialog and enable it on OpenCL tab. In this window you can select graphics card which you want to use for computation.)

I would wait for next years 7nm GPUs from both nvidia and AMD. 2080 prices are ridiculus because of the new raytracing feature (not yet supported in any app) not that much faster and still limited VRAM.

1 Like