cancel
Showing results for 
Search instead for 
Did you mean: 

Archives Discussions

npm1
Adept II

Blender Cycles(Opencl on AMD GPUS)

Dear Opencl Developer

Why doesn't the AMD opencl compiler work with blender cycles?

whenever i compile the blender cycles kernel, the system either crashes due to lack of memory, or takes too long to compile the blender cycles kernel(which thereafter comes up with the following error:

opencl build failed:errors in console

calclcompile failederror: creating kernel_ocl_path_trace failed!

can't open file c:\tmp\5688.blend@ for writing:no file or directory

).

When is AMD opencl compiler going to work properly with blender cycles?

Why can't the AMD opencl compiler developers test their compiler against blender cycles?

Seasons Greetings,

npm1,

PS i am as well as others(i assume) are considering to make a switch from AMD GPUs to Nvidia.

371 Replies

The catalyst 14.7 is out and first test with blender cycles show no improvement in render time and compilation time ! no change still at the same place. we need clarification here !

0 Likes

Performance in both Indigo Renderer and Luxrender using opencl is now horrible. I'm really hoping Bob's right.

0 Likes

Please we need to know when this driver will be release !

0 Likes

Yes, Bob, please, give us just a quick update.

0 Likes
cusa123
Adept I

Amd Okay, finally have something fairly solid, I leave a picture of sintel hair test for blender-2.70-win64-918f6a4. And own project rendered with amd 7870.

Goofy is a single modeling in blender. What we hope now is a little improvement in performance cycles.

Sintel

http://subefotos.com/ver/?f6f8fee759b883b43a67b9e181990eaco.jpg

Goofy

http://www.subeimagenes.com/img/goofy-final1-986948.jpg

_________

Muy bien amd, al fin tenemos algo medianamente sólido, dejo una imagen  de sintel para la prueba de pelo blender-2.70-win64-918f6a4 .

Y un proyecto propio renderizado con amd 7870. Es un Goofy solo modelado en blender. Lo que esperamos ahora es un poco de mejoras en rendimiento para cycles.

0 Likes
sharlybg
Adept II

Thing are going well but the speed is to bad For GCN based GPU !

in normal situation GCN must be 40% faster in rendering than kepler but they simply 2X time slower !

We are sure the next catalyst will really improve performance !

0 Likes
cusa123
Adept I

In terms of performance, it is really missing a lot to accelerate OpenCL cycles.

But the big question is, how can we accelerate OpenCL? accelerate if the culprit was the driver of amd.

0 Likes

It could be the Drivers but more likely The Kernel itself.

The Blender Devs have posted on more then one place that the OpenCL Kernel is not yet optimized as they could not even get it to work completely.

My hope is AMD will Get the Drivers Fixed so there are no more Compiling Problems, and again Hoping the Blender Devs will pick The Opencl Kernel up and dust it off. 

0 Likes
cusa123
Adept I

Bob you have any idea how this catalyst for OpenCL cycles. Some improvement in performance?.

The problem I'm noticing is that each version of blender becomes slower.

test bmw "w8.1"

blender 2.69       - 146.83 mem - 00:54.70 sec - 50/50

blender 2.70       - 146.86 mem - 01:13.82 sec - 50/50

blender 2.71 rc1 - 151.78 mem - 01:32.13 sec - 50/50

cusa123
Adept I

Well the performance problem is due to the cache memory. Apparently OpenCL blender amd have to rewrite in memory at least 2 times, to make it faster.

0 Likes
sharlybg
Adept II

Last catalyst 14.6 RC2 tested no improvement nothing nada quedal !

cusa123
Adept I

This is what we wanted Bob, tell us some cycles blender. Great job!

Bob some questions 1 - Do vram problems solved? 2 - Would it be possible to ask blender amd OpenCL enabled (in my opinion it is stable)? 3-the big question why you have problems with OpenCL speed must double working process?

0 Likes
cusa123
Adept I

http://i.imgur.com/NUO9Azi.jpg

The problem with the Bmw in blender 2.71. The stability problem is the use of oc and GPU VRAM.

Consumption in 7870 is  (139-155 mb).

It would be nice to try some feature in catalyst (3d app settings) What would be the best option for better performance?.

01.jpg

0 Likes
bubleguber
Adept I

Sin nombre.png

Guys we aren't in mid-july and Bob did not provide update in this topic, this is not the patch.

Time is money ! we are waiting again without news !

0 Likes
cusa123
Adept I

Bob something new. I asked Ton Roosendaal, if possible. Enable OpenCL at least for amd. The answer was no! says no and he does not believe this to future versions of Blender 2.72. Something to also ask. I would like to try blender cycles in linux ubuntu, but I have problems so I hope to catalyts 14.7 to install linux.

0 Likes

I asked Ton about two months ago, and he said he needs the driver to be as good as the CUDA-drivers to whitelist it.

0 Likes

For my functionality is almost all good. But lacking a bit in the optimization of vram and speed now. I'll be waiting for an improvement before you leave (AMD Pirate Islands), replaced my gpu.

So I'll have to stick with the beloved 7870.

VincentSC Bob told us about the catalyst, would be for next week? or July was the version the improvements?.

0 Likes
bubleguber
Adept I

Hi maybe it's time for news? release was planned for the middle of this month and we still do not have the update on this post. What's happening? thanks

0 Likes
adamu
Journeyman III

Another remark to devs, regarding using openCL in cycles render in blender:

- Motion/Object Blur did not work, so I opened kernel_types.h and noticed that these settings (among others) are commented out:

#define __CAMERA_MOTION__

#define __OBJECT_MOTION__

so I activated most of them, to be able to utilize my HD7870 OC Black to the full extent for hair, cloth, blur etc., and proceeded to compile the kernel.

But, kernel would not compile if #define __CAMERA_MOTION__ is active. When I commented it out, kernel compiled with all other things without any problems.

Why isn't that (I presume its for motion blur) supported? Why is compiler making an error on 3 lines (SEE LINK)?

Link to blender console errors

Awaiting for your reply and some news regarding the driver/openCL update.

0 Likes

About the error:

# define __ CAMERA_MOTION__

# define __ OBJECT_MOTION__

These two errors in OpenCL already know. But only we would have to wait until tomorrow to have some idea if the driver amd promised will be released.

I hope not only object motion, the most anticipated is the speed, and the burden of vram error.

I concur.

The speed is the main thing, vram being the secondary to what should be optimized first. Capt. Obvious of me, make it like a small reminder.

Anyway, I would like to see that soon Open CL should rival CUDA in performance, with ease.

0 Likes
sdar
Journeyman III

[...The update release has shifted to mid-July but I'll provide a further release update as we get nearer to that date.]

So, any further release update?

0 Likes
sharlybg
Adept II

Maybe they don't have human and financial means to do something for us ! so why do you think they can't fix the bug after 3 Years !

0 Likes
VincentSC
Adept II

I've the checked the previous release dates.

The normal delivery time for drivers has been a week before the end of the month (around the 25th). If not, then it mostly skips one or two weeks to the second week of the next month (standard delay?). So statistically we have to wait until the week of the 11th.

0 Likes
bobwhitecotton
Adept III

I am on vacation but tracking the flurry of activity surrounding this issue.  There has been some good news and some not so good news thus far, which is the source of the recent flurry.  The team is still working diligently to get a new version of the driver out in the next couple of weeks but frankly the performance gain is still uncertain given the deadline for this driver release.  That's all I have for now.

Bob

0 Likes
cusa123
Adept I

may last BETA version of catalyst, this can give some kind of trouble?.

Apparently with newly formatted windows try to do some features walk, unable to make it work. But what I see is that it can be compilaciacion blender and a new blender.

This is way too slow in October have nvidia 880. Amd and should be required in this area to get a solution before the new nvidia generation.

blenderartist : Sdar...  "Does someone know if there's any plans to support unified memory (cuda 6+) on cycles? that could be a huge improvement in some cases even if amd manages to fix opencl"

0 Likes

Afaik the main problem with AMD GPUs and Cycles is not the Software, but the Hardware. It is clear that AMD will release something to counter Nvidias 8xx, I think the most of us should have heard something about "Tonga" and the upcoming 3xx Generation from AMD. Im sure there will be some improvements, but I heard nothing about a fix for the Cycles-Problem, are there any informations which I failed to see? Because that could be a "easy" fix for the Problem.

0 Likes
cusa123
Adept I

It would be interesting to know whether the announcement was made amd or Arnold, it seems that some users blenderartist not have much confidence in amd.

Catalyst wait to the day of my birthday and walk.

Blenderartist user :  m9105826

"Let's not spread that around as disinformation. It's a limited toy renderer running on AMD FirePro by a person who just happens to work for Solid Angle. It's unrelated to Arnold core, nor are there currently plans to integrate it. Seems AMD is trying to drum up hype with misleading information. Not that it's surprising after this driver debacle."

0 Likes

The announcement was made by Thomas Dinges, who replaces Brecht van Lommel together with Sergey Sharybin as developer of cycles since June 2014.

It was made in this (german) video: BlenderDay 2013 · Cycles Rendering (Thomas Dinges) - YouTube   2:44 -> 3:36

My (certainly not perfect) translation:

OpenCL is a standard verified by the Khronos Group. Fundamental there are three companys who use it. Intel, Nvidia and AMD. Cycles is working well on Intel CPU & GPU, Nvidia GPU and AMD CPU. The main problem is, that cycles is one big kernel and the AMD GPU architecture can't really work with that. Thats the problem.

0 Likes
cusa123
Adept I

The problem is if amd does not help with the hd gcn architecture since 7000 at least.

Not find the problem before it gets out the new series r300 serious for amd another failure, almost three years without a solution. If you could know what is happening with the driver or to pass it to "free linux amd driver?".

0 Likes
bubleguber
Adept I

I just switch to nvidia, AMD keeps giving me problems on CPU and GPU, out of the big fail from OpenCL, not coming back anymore, nor will recommend anyone to buy it, good luck to those who stay with AMD you will need it.

0 Likes

I agree.

Until now AMD was faster in crypto (SHA,WPA,... hashing -- integer operation stuff, which is not common for GPUs). But when NVIDIA's new 800 series comes out (Maxwell architecture), it may be faster than AMD in crypto. The first card with promising results is the 750ti (a 700 series one, but still Maxwell architecture, just not enough cores, only 640). Per core it is faster than a AMD 280,290 in hashing. So if the 750ti would have the same number of cores, like, let's say the 280x, then the 750ti would be faster in hashing (hashcat is a known tool)! GTX 750 TI

NVIDIA has also better Linux drivers.

The only bad thing about NVIDIA here is that they seem to have dropped the OpenCL support and recommend to use their proprietary CUDA 😕

0 Likes

I really want that that my r9 290 will work with blender cycles and v-ray rt or i will have to change it to gtx 780. Waiting for new revision of drivers.

0 Likes

I'm starting to raise money for a nvidia 880 But I can not believe with amd drivers can not, you see that the problem persists in materiad and hardware. Firepro not understand how OpenCL works, but is unattainable for the average user.

He waited for more than 2 and almost 3 generations of gpus, I think it has reached the time to put the boat amd. Amd must reach finish before the r300 driver. At this point to make it clear that you have some kind of problem, in hardware seriously, we do not understand. From my point you could leave the OpenCL linux as I said before and try to see the problem from the more advanced user.

In terms of linux could be the holy grail of possible solution.

Also progress in this year were many, but lack igualemente improved "performance" and blocking software (blender) from this point of view lacks a more drastic 70% OpenCL improvements.

This includes: 1 render speed. 2 VRAM improved load. 3 functions full materials. 4 problems overclocking difficulties rendered vram and gpu. 5 stability problems gpu cpu performance (in this case the "cpu works with 98%" - "65% gpu works").

On behalf of amd users to have no clear roadmap is just pull and test drivers. That's why many users have already given up on blenderartist all amd gpus type. For more flow and the stability that the last advertisements of Solid Angle, Lumiscaphe P3D. amd firepro are being considered with typical "amd fanfare."

Ads Roy on twitter are clear "firepro" but ask amd radeon ?.

There is no answer.

To be understood, is not nothing berate my radeon 7870 gpu asus is very good.

These comments are fair? think so, and that new users with inecesarios dislikes too.

We expect an improvement that was promised last from June, in my case as a designer eh tried to help other users with videos on youtube: - Blender Cycles instalar opencl Amd Radeon 7870 - YouTube  - that have helped many users at least a little better performance.

If this does not progress and the Blender Foundation does not enable OpenCL in operating systems for lack of it "unusable" amd OpenCL. This will continue so forever.

Greetings users Amd Develope Central.

John.

___________________________________________

Estoy empezando a juntar dinero para una nvidia 880. Pero me cuesta creer que amd no pueda con los drivers, se ve que el problema persiste en materiad e hardware. No comprendo como opencl firepro funcione, pero es inalcanzable  para el usuario medio.

Se espero por mas de 2 y casi 3 generaciones de gpus, creo que se a alcanzado la hora de dejar el barco amd. Amd debe alcanzar a terminar el controlador antes de la r300. En este punto dejara en claro que tiene algun tipo de problema, serio en materia de hardware, que no comprendemos. Desde mi punto se podria dejar el opencl a linux como dije antes y tratar de ver el problema entre los usuario mas avanzados.

En materia de linux podria ser el santo grial de la posible solucion.

Igualmente los avances en este año fueron muchos, pero igualemente falta mejoras en "rendimiento" y bloqueo de software (blender) desde este punto vista mas drastico le falta un 70% de mejoras opencl.

Esto incluye: 1 velocidad de render.  2 mejoras en carga de vram. 3  funciones de materiales completas .  4 problemas en overclocking  en renderizados dificultades con las vram y gpu. 5 problemas de estabilidad en rendimiento gpu cpu ( en este caso el "cpu trabaja con 98%"- "gpu trabaja al 65%").

De parte de amd hacia los usuarios no tenemos ninguna hoja de ruta clara es simplemente sacar drivers y probar. Es por eso que muchos usuarios de blenderartist ya desistieron en gpus de todo tipo amd. Por mas que funcione y sea estable los ultimos anuncios de Solid Angle, Lumiscaphe P3D. amd firepro estan siendo considerados con los tipicos "bombo y platillos de amd".

Los anuncios de Roy en twitter son claros "firepro" pero al preguntar amd radeon?.

No hay ninguna respuesta.

Para que se entienda, no es recriminar nada mi gpu radeon 7870 asus esta muy bien.

Estos comentarios son justos? creo que si, y que nuevos usuarios no se lleven disgustos inecesarios tambien.

Esperamos una ultima mejora que fue prometida desde Junio, en mi caso como diseñador eh tratado de ayudar a otros usuarios con videos en youtube:  -   Blender Cycles instalar opencl Amd Radeon 7870 - YouTube   -  que han ayudado a muchos usuarios al menos mejorar un poco su rendimiento.

Si esto no avanza y la fundacion Blender no habilita opencl en los sistemas operativos por falta de que sea "inutilizable" opencl amd . Esto seguira asi por siempre.

Saludos a los usuarios de Amd Develope Central.

Juan.

0 Likes

I agree. We put in 2 r9 280x's into my workstation at my request. If I threw away money than I know we'd have to go Nvidia.   Gamers only buy 1 card at a time, we use 2 and 3 at a time.   Blender is used by many small production houses that don't want to spend that 4k for maya and than put in renderman.

It's just cost effective to go with AMD for multiple gpu's and Blender cycles for rendering. 

Pixar has a whole department writing code for proprietary software to make everything render faster. We don't have the resources to have a team to rewrite cycles or create an amd specific tenderer like MANTLE.   I've tried using luxrenderer and Indigo both support opencl and these latest drivers have cut into render time.

I was even thinking of making up my scene and dropping it into unity and rendering it there!

Excuse the rant, I'm sure everyone understands we are not talking about personal preferences but budgetary constraints.

Just think Bob!! once this is fixed you'll have all of us saying  "Without BOB I wouldn't be buying this batch of cards for these workstations! Thank you BOB!" 

0 Likes
gagcl
Journeyman III

Please, don't forget the 5xxx series.

0 Likes
mdriftmeyer
Adept II

Reading all the emotional commentary about choosing Nvidia over AMD for Cycles truly misses the boat on the lack of work invested, by Blender, with respect to implementing OpenCL within Blender proper.

There are dozens of areas Blender could architect to leverage OpenCL on either Nvidia or most importantly, AMD GPGPUs and CPUs.

And yet it is AMD who gets slammed for ``dropping' the ball, while other projects manage to implement AMD's OpenCL stack well within their projects.

I am waiting for the next round of AMD GPGPUs to buy and later will be buying a Mac Pro, so the FirePro investment is assured.

I personally have zero interest in CUDA and complete interest in OpenCL, OpenGL, Metal and eventually Mantle.

It seems with all the heavy lifting that has been committed to LLVM/Clang, and a branch of OpenMP for LLVM-Clang 3.5 now branched, you would hope Blender spends some cycles working on making LLVM a first class citizen within it's project, thus making the likes of AMD, Apple and others that much more interested in pitching in knowing their investment in the LLVM community is assured inside a project like Blender.

Perhaps Blender will have some future GSoC projects centered around these areas, instead of complaining that AMD hasn't done the heavy lifting for them.

After all, the OpenCL quality from Nvidia in Blender is abysmal, but then again Nvidia has zero interest in pushing a superior solution [OpenCL] when it's heavily invested in an inferior proprietary solution [CUDA].

0 Likes

emotional commentary?....

First, this topic alone has more than two years, two years... do you think a company or person will wait 3-4 years for AMD to present a solution...

Second, it is not lack of investment Blender, but in general, as you said, even the CUDA being inferior, can do better in virtually any program or render engine that supports both.

In my house I have an AMD video card, but it really does not make sense to wait longer for a solution.

AMD is not a small company, they have money and should have an interest in seeing your product being used, helps companies implement, throw money in Blender, throw money in the other renders engine...

Only this topic will complete three years and the result still disappointing.

0 Likes


Reading all the emotional commentary about choosing Nvidia over AMD for Cycles truly misses the boat on the lack of work invested, by Blender, with respect to implementing OpenCL within Blender proper.



There are dozens of areas Blender could architect to leverage OpenCL on either Nvidia or most importantly, AMD GPGPUs and CPUs.



And yet it is AMD who gets slammed for ``dropping' the ball, while other projects manage to implement AMD's OpenCL stack well within their projects.



I am waiting for the next round of AMD GPGPUs to buy and later will be buying a Mac Pro, so the FirePro investment is assured.



I personally have zero interest in CUDA and complete interest in OpenCL, OpenGL, Metal and eventually Mantle.



It seems with all the heavy lifting that has been committed to LLVM/Clang, and a branch of OpenMP for LLVM-Clang 3.5 now branched, you would hope Blender spends some cycles working on making LLVM a first class citizen within it's project, thus making the likes of AMD, Apple and others that much more interested in pitching in knowing their investment in the LLVM community is assured inside a project like Blender.



Perhaps Blender will have some future GSoC projects centered around these areas, instead of complaining that AMD hasn't done the heavy lifting for them.



After all, the OpenCL quality from Nvidia in Blender is abysmal, but then again Nvidia has zero interest in pushing a superior solution [OpenCL] when it's heavily invested in an inferior proprietary solution [CUDA].


There is an entire community of developers, free of charge, at Blender's disposal, who are extremely helpful in all matters LLVM/Clang, and yet Blender devs are silent on the LLVM/Clang Dev Lists. Hell, you can even talk to AMD devs on there, including Google, SONY, Apple, IBM, and more about OpenCL.

Blender devs: Silent.

Blender is a beautiful project, but whining that AMD isn't doing the heavy lifting in working around Cycles architecture, instead of Cycles working with groups of experts to evolve its architecture speaks very poorly of Blender.

0 Likes