Page 1 of 1

AMD VEGA is here and OR OpenCL support isnt...

PostPosted: Thu Jun 22, 2017 12:55 am
by dzdiode
Dear Otoy,
im really curious about how do you feel about it? The fact that we (users, Otoy clients, artists, pro of the industry...) have the oportunity to go with AMD as rendering GPU's but we can't. We are very limited with CUDA/NVIDIA and it can be very disapointing and frustrating.
I waited a lot for the OpenCL support but nothing is here... And im going to buy 8x 1080 Ti (have no choice, deadlines are here) for an astronomic price knowing that for the same budget i can have a lot of more GPU power with AMD. I'm loosing a lot and im sure that im not the only one.

Re: AMD VEGA is here and OR OpenCL support isnt...

PostPosted: Thu Jun 22, 2017 1:09 am
by Goldorak
We didn't give up on OpenCL, OpencCL gave up on us. We now have the cross compiler - that's how we're shipping Octane on AMD shortly (win10 currently required - but may work in boot camp on Mac, testing next week). It won't be fast in this first release. It will need months to optimize just as we did for Pascal this year.

Re: AMD VEGA is here and OR OpenCL support isnt...

PostPosted: Thu Jun 22, 2017 3:41 pm
by dzdiode
Goldorak wrote:We didn't give up on OpenCL, OpencCL gave up on us. We now have the cross compiler - that's how we're shipping Octane on AMD shortly (win10 currently required - but may work in boot camp on Mac, testing next week). It won't be fast in this first release. It will need months to optimize just as we did for Pascal this year.

Thanks Goldorak! So when approximatively (days/weeks) we can get this version? Just to know if i buy the NVIDIA cards or wait cause i can wait a few more days and yes im ok with the no full speed for the first releases, for me, the more important right now is my investment $$$

Re: AMD VEGA is here and OR OpenCL support isnt...

PostPosted: Thu Jun 22, 2017 5:00 pm
by FrankPooleFloating
If it were me, I'd stick with the 1080Ti plan... You appear to have a major deadline looming, on an enormous gig. 1080Ti is a sure thing. I have never seen Goldorak or anyone else on team state that AMD GPUs are guaranteed (now, or ever) to be as powerful/stable as NVidia GPUs in Octane. I'm not a betting man.. especially when deadlines and/or money are involved. But at the same time I will admit that I have not researched this, and I am just going by me not remembering any promises where AMD GPUs and also CPUs are concerned. Maybe supporting things other than NVidia GPUs in Octane will be just that.. supporting them.. I dunno. I'm sure someone will set me straight. ;)

Re: AMD VEGA is here and OR OpenCL support isnt...

PostPosted: Fri Jun 23, 2017 10:12 pm
by Goldorak
FrankPooleFloating wrote:If it were me, I'd stick with the 1080Ti plan... You appear to have a major deadline looming, on an enormous gig. 1080Ti is a sure thing. I have never seen Goldorak or anyone else on team state that AMD GPUs are guaranteed (now, or ever) to be as powerful/stable as NVidia GPUs in Octane. I'm not a betting man.. especially when deadlines and/or money are involved. But at the same time I will admit that I have not researched this, and I am just going by me not remembering any promises where AMD GPUs and also CPUs are concerned. Maybe supporting things other than NVidia GPUs in Octane will be just that.. supporting them.. I dunno. I'm sure someone will set me straight. ;)


I have to agree with the above.

When using Octane 3 for production ready rendering, especially on a deadline, please keep in mind the following guidelines:

1) Never rely on features that are not yet supported in Octane 3 with the latest stable release for your host application

2) Never assume peak performance or stability using hardware that is not supported in Octane Bench V3 or later. We only add a GPU to OctaneBench once we are dune tuning Octane on the supported architecture and all is stable.

AMD support is very experimental right now, and I don' see that changing until we are at least a few releases after the first drop to users. We have spent a lot of time just making sure that very first release is feature complete and doesn't crash. I'd be surprised if we are ready to add AMD GPUs to Octanebench any faster than we did with Pascal GPUs (which were much simpler to tune than AMD w/ cross compiler). We will have months of performance tuning and (presumably) bug fixing.