Page 2 of 3

Re: Indigo Renderer 4.0.44 Beta Release

Posted: Sun Sep 04, 2016 8:11 am
by fourzeronine
Asticles wrote:Hi all,

I've make an account because I'm considering to buy indigort to work with blender and amd cards but I've seen some issues:

Indigo:

I'm testing indigo with my two RX480, and seems that having the opengl window opened crashes the program when rendering with opencl, but also crashes when exiting indigo after render with the cards.
I've sent a dump.

I'm testing the memory leakage because some days ago I had the same problem. I've updated drivers but the leak seems to continue(on gpu mem). Also, it does not seem to work with my two cards (indigorender).

IndigoRT:

It does not have the opengl problem, nor the leakage (on gpu mem), but maybe is too small that I cannot see it on gpuz.

When I hit pause the GPU load still is at 100%, hitting close button makes the program not to respond and hangs the computer.

Regards
Id love to see the performance of the rx 480's. are they the 8GB versions?

Be sure you disable crossfire and do a full restart to test both cards, also make sure cpu is disabled. This is the only way I get good performance with my 2 Radeon 7970's.

Re: Indigo Renderer 4.0.44 Beta Release

Posted: Mon Sep 05, 2016 4:02 am
by Asticles
fourzeronine wrote:
Asticles wrote:Hi all,

I've make an account because I'm considering to buy indigort to work with blender and amd cards but I've seen some issues:

Indigo:

I'm testing indigo with my two RX480, and seems that having the opengl window opened crashes the program when rendering with opencl, but also crashes when exiting indigo after render with the cards.
I've sent a dump.

I'm testing the memory leakage because some days ago I had the same problem. I've updated drivers but the leak seems to continue(on gpu mem). Also, it does not seem to work with my two cards (indigorender).

IndigoRT:

It does not have the opengl problem, nor the leakage (on gpu mem), but maybe is too small that I cannot see it on gpuz.

When I hit pause the GPU load still is at 100%, hitting close button makes the program not to respond and hangs the computer.

Regards
Id love to see the performance of the rx 480's. are they the 8GB versions?

Be sure you disable crossfire and do a full restart to test both cards, also make sure cpu is disabled. This is the only way I get good performance with my 2 Radeon 7970's.
They are the gigabyte base 8gb version. I would like to have the sapphire ones but I didn't found them where I live.

Well, the problem is the sysrem memory leak, it fills my 24gb in less of a minute, so I cannot test it properly.

I would like to have a good opencl render with the last technology, because I'm a bit tired to wait cycles developers to put opencl at the level of the cuda cards in terms of features.

This is why I'm testing indigo, but last version 3 is really slow in opencl, and by now this 4th version is stilll in development and unusuable with my cards.

Regarding speed what I've seen with luxmark, I score on the hotel scene 5246.

Regards!

Re: Indigo Renderer 4.0.44 Beta Release

Posted: Mon Sep 05, 2016 8:17 pm
by Asticles
With Indigo unlicensed, just tested the bedroom benchmark with a Radeon R9 280x and I'm scoring on GPU 1.695 Ms/s. Since the maximum resolution is lower. Shouldn't I compare with the other scores?

No memory leak with this card, apparently.

Re: Indigo Renderer 4.0.44 Beta Release

Posted: Mon Sep 05, 2016 8:33 pm
by Oscar J
Asticles wrote:With Indigo unlicensed, just tested the bedroom benchmark with a Radeon R9 280x and I'm scoring on GPU 1.695 Ms/s. Since the maximum resolution is lower. Shouldn't I compare with the other scores?

No memory leak with this card, apparently.
Resolution doesn't really affect rendering speed. What do you get with the 480?

Re: Indigo Renderer 4.0.44 Beta Release

Posted: Mon Sep 05, 2016 8:58 pm
by Asticles
I wasn't able to test it because it seems to start with a very low number and after some time it gets higher. The problem is that it starts to get some speed when I run out of system memory due the leak.

Re: Indigo Renderer 4.0.44 Beta Release

Posted: Mon Sep 05, 2016 9:00 pm
by Asticles
Additionally it seems to work with different speeds if the card is the display one or not. But this have to be tested when the leak is solved.

Re: Indigo Renderer 4.0.44 Beta Release

Posted: Mon Sep 05, 2016 9:14 pm
by Zom-B
Asticles wrote:... it seems to start with a very low number and after some time it gets higher.
Yes, GPU rendering doesn't start with full force but raises over the first ~10seconds to full usage,
afaik a workaround because NVDIA GPUs otherwise crash if utilized directly with 100% from the start...

So your memory leak is on AMD 480 cards only, does it happen if you render with only one card too?
Sending a dump should be very useful, now that weekend is over the devs should be in office again and will take a look on it for sure :)

Re: Indigo Renderer 4.0.44 Beta Release

Posted: Mon Sep 05, 2016 9:18 pm
by Asticles
I sent a dump but I don't know if has all the necessary info.

Yes, the leak happens also with only one card. It's very strange it happens only with the 480.

Regards

Re: Indigo Renderer 4.0.44 Beta Release

Posted: Thu Sep 08, 2016 2:44 am
by OnoSendai
Hey everyone,
I have got a RX 480 and figured out what is causing the problem.
It looks like an AMD driver bug, but quite a specific one.
I'll file it with AMD, and also release a workaround that should solve this problem, in the next Indigo release.

Re: Indigo Renderer 4.0.44 Beta Release

Posted: Thu Sep 08, 2016 4:08 am
by Zom-B
Next time please somebody having an issue with an Titan X
So the devs can update their rigs :P

Re: Indigo Renderer 4.0.44 Beta Release

Posted: Thu Sep 08, 2016 4:49 am
by STUDIO AARON
Sweet!!

Re: Indigo Renderer 4.0.44 Beta Release

Posted: Thu Sep 08, 2016 9:12 am
by Asticles
OnoSendai wrote:Hey everyone,
I have got a RX 480 and figured out what is causing the problem.
It looks like an AMD driver bug, but quite a specific one.
I'll file it with AMD, and also release a workaround that should solve this problem, in the next Indigo release.
Great! Thanks!

Is indigo team in close contact with amd for issues like this? for example to get a patch on next driver release.

Re: Indigo Renderer 4.0.44 Beta Release

Posted: Thu Sep 08, 2016 10:36 am
by OnoSendai
Asticles wrote:
OnoSendai wrote:Hey everyone,
I have got a RX 480 and figured out what is causing the problem.
It looks like an AMD driver bug, but quite a specific one.
I'll file it with AMD, and also release a workaround that should solve this problem, in the next Indigo release.
Great! Thanks!

Is indigo team in close contact with amd for issues like this? for example to get a patch on next driver release.
I have just posted on their dev forum, which is where bug reports go.

Re: Indigo Renderer 4.0.44 Beta Release

Posted: Thu Sep 08, 2016 10:45 pm
by Zalevskiy
GPU path tracing initialisation failed: clBuildProgram failed: CL_BUILD_PROGRAM_FAILURE

Re: Indigo Renderer 4.0.44 Beta Release

Posted: Thu Sep 08, 2016 11:53 pm
by burnin
^confirming an issue
GPUs: NVidia Quadro 600 & M5000

Code: Select all

Building OpenCL program for device 'Quadro 600' ...
OpenCL build program failed.  Build log:
<kernel>:220:37: warning: implicit declaration of function 'printf' is invalid in C99
float print_bool_(bool x) { if(x) { printf((__constant char *)"true\n"); } else { printf((__constant char *)"false\n"); } return x; }    
                                    ^
<kernel>:711:1166: error: cannot decrement value of type 'float'
                if_res_32 = (fmax((let_result_49 * (l1ll1ll1l11ll11l11ll1l11l111l1l11_bool__FullHitInfo__vec4__vec4_(lll1ll11l11ll111ll111lll1l1l1l11l_FullHitInfo__vec4__vec4_(ll1111lllllll1l11l1ll111ll111lll1, &l1l111l1l11lllll1111lll11l11l1ll1_arg_61, &l1ll11111l111l1111lllll111l11lll1_arg_62), ll1111lllllll1l11l1ll111ll111lll1, &l1l111l1l11lllll1111lll11l11l1ll1_arg_63, &l1ll11111l111l1111lllll111l11lll1_arg_64) + l1ll1ll1l11ll11l11ll1l11l111l1l11_bool__FullHitInfo__vec4__vec4_(lll1ll11l11l1111ll1l1l1l1l111ll11_FullHitInfo__vec4__vec4_(ll1111lllllll1l11l1ll111ll111lll1, &l1l111l1l11lllll1111lll11l11l1ll1_arg_57, &l1ll11111l111l1111lllll111l11lll1_arg_58), ll1111lllllll1l11l1ll111ll111lll1, &l1l111l1l11lllll1111lll11l11l1ll1_arg_59, &l1ll11111l111l1111lllll111l11lll1_arg_60))), (let_result_37 * l1ll1ll1l11ll11l11ll1l11l111l1l11_bool__FullHitInfo__vec4__vec4_(lll1llll1l1ll11llll11l1l11l11111l_FullHitInfo__vec4__vec4_(ll1111lllllll1l11l1ll111ll111lll1, &l1l111l1l11lllll1111lll11l11l1ll1_arg_45, &l1ll11111l111l1111lllll111l11lll1_arg_46), ll1111lllllll1l11l1ll111ll111lll1, &l1l111l1l11lllll1111lll11l11l1ll1_arg_47, &l1ll11111l111l1111lllll111l11lll1_arg_48))) * --0.1f);
                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                           ^ ~~~~
<kernel>:3053:72: warning: assigning to '__attribute__((address_space(16776963))) void *__attribute__((address_space(16776963)))' from 'const __attribute__((address_space(16776963))) float *' discards qualifiers
}  ll1111lllllll1l11l1ll111ll111lll1.l11ll11ll11ll1llll1111ll1lllll111 = (const float*)l11ll11ll11ll1llll1111ll1lllll111; ll1111lllllll1l11l1ll111ll111lll1.ll1111lllllll1l11l1ll111ll111lll1.ll1l1l11l1lll111lll1l111ll1llll1l.v = (float2)(lll1111lll1l111ll11lll1l111lll111, l11l1l1111l11l1lll111lll1lll11111); 
                                                                       ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
<kernel>:3659:72: warning: assigning to '__attribute__((address_space(16776963))) void *__attribute__((address_space(16776963)))' from 'const __attribute__((address_space(16776963))) float *' discards qualifiers
}  ll1111lllllll1l11l1ll111ll111lll1.l11ll11ll11ll1llll1111ll1lllll111 = (const float*)l11ll11ll11ll1llll1111ll1lllll111; ll1111lllllll1l11l1ll111ll111lll1.ll1111lllllll1l11l1ll111ll111lll1.ll1l1l11l1lll111lll1l111ll1llll1l.v = (float2)(lll1111lll1l111ll11lll1l111lll111, l11l1l1111l11l1lll111lll1lll11111); 
                                                                       ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~