Fractalforums

I'm going to buy a new GPU soon for use with mandelbulber. The new 20 series of cards from Nvidia has raytracing cores amoung other technologies. Could this tech be used to speed up OpenCL for Mandelbulber? If so perhaps it would be worth the price premium over the 1080ti. Thoughts? 

oops meant to post in mandelbulber

Linkback: https://fractalforums.org/mandelbulb3d/13/rtx-with-mandelbulber/1988/

Nvidia has not yet disclosed the detailed specification of those new hardware blocks added for raytracing. Furthermore the programming interface for these RTX units is, as far as I know, fairly specific to raytracing of typical 3D models. So it is currently unknown if the RTX cores will ever be useful for fractal purposes.

However, the early benchmark data suggests that Nvidia's 2000 series cards are notably better at compute tasks than the 1000 series. The performance improvement for compute seems to be quite a bit larger than the performance improvement for games.

However, the 2000 series cards are also very expensive. Maybe one can get three or more midrange GPUs for the price of a single 2080Ti, and maybe those three have a higher combined compute throughput...?

Thanks for the suggestion, I'm leaning toward the 2080 after looking into its compute scores, If I end up getting it I'll do some benchmarks. I don't believe mandelbulber supports multi gpu setups currently

I posted some price performance comparisons on the Indigo Renderer blog recently: https://www.indigorenderer.com/node/2201

The RT units in the new RTX won't be useful for fractal rendering, but overall it's still a monster for GPU compute and worth the investment IMO.

I'm am not a Mbulber user myself but I remember having read something about getting a system working with more than one graphics card (2xMSI 980) in it - in a bit unorthodox setting, but hey... as long as it works, and it really seems to work just great

Should you wonder if two midrange cards might be an alternative to your 2080, check this thread on the old forum and scroll down to Paigan's post:

http://www.fractalforums.com/releases-b233/mandelbulber-v2-2-08/new/

Paigan is also a member, here, btw.

If you have more than one GPU in your computer, you can load other copies of the program and assign them to the other GPU's.  But you have to manually assign the work for each GPU,  i.e. render frames #0 to #x on one GPU, frames #x to frame #y on another, etc.

Netrendering in openCL will probably be implemented  before the  multi gpu option. This all will take some time.

Maybe one can get three or more midrange GPUs for the price of a single 2080Ti, and maybe those three have a higher combined compute throughput...?

This would be my approach. Depends a bit on work methods.  More than one gpu allows you to be rendering a final section of the animation while making changes to another section of the animation.

Thanks for the advice, taking workflow and current setup into consideration I think the best op at this stage would be to have two GPUs, one mid range and one higher range with the better card for fractal exploration and the lower end card to supplement rendering. It wouldn't take much to get another system going in future for the net render, when that feature is ready. Most of my fractal art has been still images up until now so as I dip my toes into animation I'll improve the setup accordingly. Thanks again!