I don’t think that is a fix nor acceptable.
My render times more than doubled in some scenes like Classroom from 2.8 to 2.82.
They rendered AMD GPUs useless. I’m having a hard time believing in a project that throws a quarter of the GPU market under the bus.
Thing is that Blender Foundation had two coders from AMD and they still couldn’t optimize it even close to CUDA.
The compilation time for shaders is a horrible user experience, some cases I wait 50 sec. OpenCL is complicated and so is Vulkan.
So I started to use Eevee. Guess what. Bug in DoF if you render 4K on AMD cards.
Even the AMD ProRender renderer is faster on nVidia cards than on their own high end cards.
Yet. It’s the only card I have and everything works good enough for me. I also believe that Blender devs will fix the problems we have with the AMD cards. I’m just an amateur and if I was professional I would have bought a 2080ti or above.
AMD’s problems go beyond just Eevee and OpenCL.
There’s reports of Navi owners having the chance of black screens by simply playing a game (even though AMD has already addressed them to an extent). AMD’s primary saving grace is that even Intel’s 10th gen iGPU’s have such bad software that Blender 2.8x (based on a bug report) can barely run at all (with Raja’s arrival doing nothing to improve the overall quality).
You may hate Nvidia’s prices, but they are the only company that really seems to understand how a GPU works. The studio drivers have not failed on me yet.
Based on today’s news of new drivers, one key part of said release is the “black screen” issue among others.
Still that doens’t take away from Blender institute pusing Nvidia performance, Nvida is pushing good funds into the needed development.
Up to recent there was only one dev hired by AMD to support that, and only recently additional funds for devs was supplied.
Still Back on Topic.
They are pushign the fix that is needed, and we’ll see how that goes. But i"m tempted to write to Ton with an overall concern and asking for some clarity.
From 2.80 release to 2.81 release they did limited work to “patch” a drastic drop in performance (initaly was about 2x slow down), and after the patch in 2.81, it was “just” 30% slower then in 2.80… This indicates that their internal testing is either insufficient to catch such major bugs, or they are prioritizing new features over stability and reduced performance… 
Do note what Bigblend’s post said and you will find it could more likely be due to AMD’s relative inability to create a truly robust graphics product than anything else. This notion can be backed up if it is true that a render engine created by AMD with OpenCL in mind actually runs better with Nvidia.
In fact, perhaps the Blender devs. should face facts and expect every Blender user to buy Nvidia if they want to have the best experience. AMD drivers have had issues since the early ATI days and it shows Nvidia all but deserves to become the only option in graphics (with AMD liquidating the Radeon brand and shift their focus entirely on bringing down Intel).
I’d say they should learn from Luxrender but it is another horror show with an application that granted them a lot of wins in the past. Future looks bleak.
I am a Linux, Blender and AMD user out principle. I will stick with AMD because they actually have free drivers for Linux but Blender telling me to embrace Nvidia and their proprietary thingies is the same as telling me not to put up with their shit and dump them for Maya and the gang.
That’s the thing that irks me. Open Software but not OpenCL. Like If anything OpenCL should be priority as it is open. Then again, OpenCL has proven itself to be unlikable, like GIMP.
Statistically speaking (and data does not lie), it is rare for Nvidia users to be subject to a vendor-specific graphics bug when using Blender. Far more bug reports regarding graphics have been sent by AMD users, and it is AMD cards that make use of the most workarounds in the code.
AMD’s drivers are not total garbage though like Intel, so Blender 2.8x can at least be kept usable. It is still a waiting game though, and it nullifies the advantage AMD cards have in terms of performance per dollar.
Blender is the equivalent of GIMP in the 3D software world, so pack those three together. There are far more tutorials, better workflows, nicer artwork and wider selection of plugings for other software.
I wonder what the hell would happen with the Blender Foundation if Blender users had the same attitude towards their screw ups as they have towards their features.
Would love to see where you got the data from.
Looking at :developer.blender.org, i can see equal CUDA and OpenCL BUG reports and tasks
Generally speaking this bug that cause the slowdown was from my perspective a issue of pushing new features into Blender without proper testing across all platforms. And even after testing, they decided to push a new feature over performance/stability across the availble API’s.
Won’t say AMD is clean, hell no, not even close. But Blender team needs to test and hold back a feature (especially a small one like a new noise shader…) when it has this drastic impact. But they didn’t.
So in this particular slowdown issue, it was purely Feature over performance choise made by Blender team, either by conscious decision or insufficient testing.
Main thing now, they are looking at resolving it, so that is the most important part.
Considering how OpenCL did not seem robust enough to get AMD’s own Prorender to a truly professional-grade state, I am not sure just how much improvement can be expected for those using Cycles with Radeon. Now we have OpenCL having trouble with the new adaptive sampling patch, which could be a sign the API has not been able to evolve to where kernel size and complexity is no longer that large of an issue (unlike CUDA). Long term, OpenCL itself might be replaced entirely by Vulkan but that is a major project.
Fortunately, AMD seems to be doing quite well in revitalizing the CPU as a decent rendering device (by forcing the mainstreaming of 8 core 16 thread machines). GPU-wise though, AMD continues the story of rocky launches, subpar software, and questionable moves generating user anger (such as the 11th hour BIOS revision for the 5600XT).
In 2.82, render begin freezes sometimes (with RX 570). Blender Team add OptiX support but broke OpenCL support. And always changes node UI behaviours and this is too boring. When new version publish, every time I need to change my node values.
If Blender team want to change some things radically, please wait and publish after all radical changes totally finished and tested.
Not all changes are caused by the core development team. Blender is an opensource project, so outside developers sometimes make improvements that change the inputs and outputs. This change doesn’t break material setups, but a developer recently made a patch which adds a UVW output to the voronoi texture node: https://developer.blender.org/D7030
This is why they say its best to stick with the same version of blender the entire time you are working on a project. Only use new versions of blender for new projects.
If you use own Node Group, then brokes your setup, tangled your design.
I’m software developer. Open Source or Closed Source, not important, if you want change interface, then must wait, carefully design and change after all this things. You can not change interface behaviour every 3 months. Wait, makes ripe, after that publish.
If one feature change, then all project change. Not every users an amateur user. We want stabilization when develop anything.
This is why they are making an LTS version of blender. Its for folks who prefer stability.
Man, 2.80 version is radically changed everythig. In 2.7x version, most things missing. If highly necessery features is missing, and you add this features in 2.8x version, then you cannot talk about LTS.
It is adding features that causes UI changes. They can’t both add “necessary” features and keep the UI the same at the same time without hiding the new features as console commands.
The best thing anyone can do is just stick to a specific version of blender while working on a project to ensure that they won’t have to make a lot of changes to the shaders in the scene or run into a new bug.
The new LTS version that will come out this year will make things better in this situation because the foundation will continue to fix any bugs found in that version for years, so we won’t have to worry about running into an undiscovered glitch a year into a big project.
So, “What is stable version?”
In programming world, STABLE = STABLE. STABLE not equal BUGGED VERSION.
If publish a stable version, then this means, this is Stable and we use this for serious works. But when I start serious works and always see bugs, then what is this?
I don’t even get what you mean right now. When did I advocate using a bugged version? I was saying that if you need more stability it’s best to stick with one version of blender for a project because you don’t have to worry about new bugs and other issues that come with upgrading to the next version.