because I have time for that right now )
so, you opting to depreciate OpenCL in Cycles?
No. I say must keep stability. Stability and performance important than new abilities.
2.80, Cycles works good. 2.81 adds 4D Noise etc., and performance decreased. 2.82 stability is nearly good. 2.83 adds new abilities and stability is gone. This is not normal situations.
And what if they canât do that because of opencl?
Then they can only add for CUDA this abilities. Same thing, if they canât do that because CUDA?
Thatâs certainly a part of the evaluation. It may not be pleasant, but in reality, no one has resources to waste.
Sometimes, another strategy is to keep an API around and to just do whatâs absolutely necessary. If I was a Cycles developer, that would likely be something I would try to do for OpenCL.
Sometimes for keep users, you must waste your resources.
Example, Vray maybe think like this: Our licence is thousands dollars for years. Our potential users can buy a expensive GPU. And we are industry standard. But Blender and Cycles can not say that.
In another thread I compared Blender Opendata benchmarks between 2080Ti and the 5700XT a GPU costing at least 1/3 the price of the 2080TI.
The 2080TI doesnât look great value for money by comparison.
The Tech sites all seem to think AMD with RDNA2 will bring a card that is >x2 the shader performance of the 5700XT plus it will have hardware ray tracing. If it offers similar value for money as the 5700XT then it could be a very interesting option.
The Tech sites almost all think nVidia has underestimated AMD and is struggling with the 8nm process and clock speed as well as enormous TDPs.
Iâm going to reserve judgement until these cards are released but is nVidia really all that better?
2080Ti Optix (CUDA) vs 5700XT
BMW 17s (35s) vs 78s
Class 72s (113s) vs 162s
Fishy 32s (72s) vs 116s
Koro 55s (115s) vs 76s
Barber 815s (490s) vs 302s
Pavillion 93s (217s) vs 474s
so⌠not adding new features that other pars of software have and fixing only bugs, sounds like depreciating OpenCL
Blenderâs release cycles is quite fast. there is less OpenCL users that test new versions, so more bugs pass under the radar
not adding new features only will only deepen difference in userbase as issues with megakernel kinda did it
BTW guys from OTOY said that vulcan is not mature enough yet in their post about metal in OctaneX, but might might it be possible in not so far feature that it will be mature and couldbe the one for all api for Cycles?
Vulcan and Cycles. This is two separate things.
If I can not take render, what should I do new feature? If you can not, then not add new abilities, until they can do this. But I know as a programmer, everything is possible, if you know how do this. Another things are only can be excuse, not anymore. Cycles programmers know good CUDA than OpenCL, and can not program to GPU with OpenCL as good as CUDA.
[vulcan offtopic]
I know that ongoing Vulcan project in Blender have nothing to do with Cycles.
But in future is it tight be?
from https://render.otoy.com/forum/viewtopic.php?f=15&t=75411
Does this mean Octane for AMD/Intel on Windows is now possible?
The work to support AMD and Intel (and others) was completed alongside our Metal work, but that hasnât really brought us much close to shipping anything on those devices. We really need Vulkan to catch up to Metal and itâs not clear when that might happen. As of right now, the Vulkan drivers on AMD and Intel are not mature enough to compile (much less ship) Octane for Vulkan. The only Vulkan drivers that do run Octane for Vulkan just fine (which is how we know it works!) are NVIDIAâs - and we will be using Vulkan+RTX for Brigade on NVIDIA.
So they at otoy tried to make octane on vulcan, am i understanding this right?
[/vulcan offtopic]
Anyway what instability issues are you talking about? I havent heard about it like i heard about gosc shader problems.
And second thing LTS. Isint that what you expect from BF?
There is now information coming out that if you want say a new Nvidia 3080, you will have to rip out your entire PSU and buy a new one and rewire you computer. Because it needs a 12 pin power connector.
Its possible Nvidia will be forced to brute force their way to the GPU gaming crown this generation, with power hungry cards, vs AMDs new efficient RDNA 2 on 7nm EUV. And like i say going forward in the FUTURE, we are stuck with CUDA so Nvidia can charge whatever they want.
And to re-iterate, if things start to unravel at Intel as they appear to be doing, especially with their future plan of big-little architecture which does not benefit power users as the small cores dont have the full instruction set, AMD having the clear lead for possibly quite a few years will definitely up their prices significantly, especially for workstation CPUs. So buying a Threadripper will not be a cheap solution to the CUDA conundrum(if they crank the prices).
Last I checked, Intel retreating to big-little architecture (where you are back to 4 cores and 8 threads for Blender work) is a rumor and has not been confirmed.
Another theory is a bit more optimistic, and that is the specification for Alder Lake hints at Intel officially following AMD in ditching the monolithic chip design.
As for RDNA2, thereâs a rumor floating around as to whether the massive gains is due to AMD figuring out how to make the Infinity Fabric work for the GPU, it would definitely mean more power efficiency and higher yields for starters.
AMD have already made Infinity Fabric work for GPUs itâs not a rumour, the Mac Pro dual Vega GPU boards use Infinity Fabric. Infinity Fabric also connects multiple GPU modules together so itâs already a thing.
Ok I assume youâre not trolling, since you spent so many time writing here (time youâd had spent better fixing Cycles problems on Opencl 
).
So not trolling, but what are you doing then?
Reading here and there youâre basically repeating over and over two concepts:
- Cycles developers are bad developers
or
- Cycles developers are hindering (with malice) AMD
the corollary is that we users are blind or dumb to not see this.
You might be a âCassandraâ, sure.
But when I put together the two following sentences of yours, it actually sounds like you donât really know what youâre talking about, from both a technical and an âinside Blender developmentâ side.
The repeated and documented voices that explain who and why had/has problem with openCL doesnât seem to have a grip on you.
This is not a real conversation. It doesnât let us build anything upon each other posts.
I wish some moderator takes a look at the situation here.
Peace.
edit: post flagged and hidden because of âadvertisingâ???
2 Likes
They use Vulkan for accelerated Ray-Tracing like OptiX.
Sadly, LTS is more bugged than non-LTS versions.
Hi guys,
Please keep the conversation reasonable and respectful please, no personal arguments. Thanks.
2 Likes
Everyone needs to stay calm and constructive, thanks.
1 Like
Yesi but some fanatics come here and attack to persons unnecessarily and not have any knowledge.
One person says I want know lottery, one person says troll. Here is like child park.