OpenCL really need some love ! in preview at least

Hello guys, here i’m again after 1 year, and after we talk with Ton too. @Ton

Amd still participating in develop with @JeroenBakker if i’m not wrong, but to be honest OpenCL fall behind Cuda too much right now, at least using it in preview.

Our case here, is a direct comparison on every day production working, between:

  1. dual Nvidia Titan X (Maxwell) Windows10
  2. 4 Amd Card (phisically 3) AMD Radeon Pro Duo Polaris 32Gb (essentially a dual Wx7100) + 2x Rx 580, all Ellesmere on Windows 10.

While in pure render the machine N. 2 is fastest by a 25%; using it for material, texture and light setup is a pain.

The preview once OpenCL kernel is calculated (faster than before, so tnx), is really too slow if compared with Machine N.1 (like 3 time slower).
It’s so annoying that somethimes we must switch to CPU for a “faster” preview… not so faster…

As @brecht said it’s a matter of time for some more love.

In a while, AMD is boosting his products, their CPU’s are the best right now… Radeon VII was a beast in Blender too, and they move on quite nicely, also with Radeon Pro Render (for both Nvidia and AMD) they push forward.
Have you tried it? have you checked how smooth is in preview with both brand under your software?

Cycle is a great engine, it’s yours, it’s integrated into Blender, but please we beg you, give OpenCL some love, or at least speak clear, so Amd users can decide to buy a better supported Cuda card, or switch to ProRender or Lux Render once for all.

We also tried with some new Amd 5700xt but same story.

I can imagine how busy the guys are there, and how many things they have to do, but hope is the last to die :smiley:

Thank you guys, and good job.
Max

Please do not move the topic back into the development section, non programming questions do not belong in that area.