VP14 and CUDA

Comments

NormanPCN wrote on 12/15/2016, 8:11 PM

I learned the first CUDA myself, but when NVIDIA came up with a completely new CUDA, I quickly decided it wasn’t worth my time.
 

What did Nvidia break in your old CUDA kernels to the new CUDA?

Red Prince wrote on 12/15/2016, 8:42 PM

I learned the first CUDA myself, but when NVIDIA came up with a completely new CUDA, I quickly decided it wasn’t worth my time.
 

What did Nvidia break in your old CUDA kernels to the new CUDA?

They don’t work with the next generation of NVIDIA GPUs.

He who knows does not speak; he who speaks does not know.
                    — Lao Tze in Tao Te Ching

Can you imagine the silence if everyone only said what he knows?
                    — Karel Čapek (The guy who gave us the word “robot” in R.U.R.)

NormanPCN wrote on 12/15/2016, 8:57 PM

They don’t work with the next generation of NVIDIA GPUs.

Old CUDA code most certainly can run on newer architectures. Binary code may not be compatible but you just do an on the fly compile of your kernel(s). It's what you do.

Red Prince wrote on 12/15/2016, 9:14 PM

CUDA on the fly? Are you sure you don’t mean OpenCL? I’m talking about many years ago when I decided it was not worth it anymore. There was no on the fly compilation of CUDA. It looked like assembly language, which was fine by me (I love assembly language), except they changed the language with each new generation.

OpenCL, on the other hand is compiled on the fly and it does not matter which GPU you are using or who made the GPU, or for that matter, you don’t necessarily even need a GPU, at least not on a PC, as both Intel and AMD created OpenCL drivers that will compile your kernel to run on the CPU.

He who knows does not speak; he who speaks does not know.
                    — Lao Tze in Tao Te Ching

Can you imagine the silence if everyone only said what he knows?
                    — Karel Čapek (The guy who gave us the word “robot” in R.U.R.)

NormanPCN wrote on 12/16/2016, 12:26 AM

I'm not talking about compiling the CUDA C code on the fly, but rather the PTX code generated by the Nvidia CUDA C/other compiler. The driver JIT compiles the PTX to architecture specific binary and caches the result. Driver updates invalidate the JIT cache.

PTX is assembly but it is a virtual ISA. Maybe PTX came after what you did.

starain wrote on 12/16/2016, 7:57 AM

Vegas programmers don't have to write their own implementation, they just need to create visual settings (as they did for mainconcept codec and many others) for 3rd party encoder. And it should be not OpenCL for nvidia, but (in case of render) - NVENC, best implementation of which now - from rigaya. Or, at least, they can take debugmodeframeserver code from github: https://github.com/satishsampath/frame-server and ajust its FOURCC part, so it could work in new 64-bit toolchains world, so anybody could use then their favorite render tool like megui/staxrip/avisynth64/etc. I've tested recently rendering on NVENC in staxrip, for my GTX 1060 for fullHD it give insane results like 400-600 FPS render! Compare with ~10-20FPS on CPU. In adobe world we have dude nvenc4everyone and (kinda buggy) implementation for all this stuff, but for vegas - nothing. They still in year 2009.

Red Prince wrote on 12/16/2016, 8:26 AM

PTX is assembly but it is a virtual ISA. Maybe PTX came after what you did.

Yes, probably. It was years ago.

Thanks for bringing it to my attention.

Last changed by Red Prince on 12/16/2016, 2:53 PM, changed a total of 1 times.

He who knows does not speak; he who speaks does not know.
                    — Lao Tze in Tao Te Ching

Can you imagine the silence if everyone only said what he knows?
                    — Karel Čapek (The guy who gave us the word “robot” in R.U.R.)

animagic wrote on 1/28/2017, 3:57 PM

I checked if there was some update on this issue, but it seems I'll need to stick with Vegas 13 for the foreseeable future. Just for the record, I have reasons to use an NVidia card due to the 3D software I deploy in my work. It has no problem rendering using NVidia, so yes, I am one of those interested in NVidia support. I switched from Premiere to Vegas years ago for a reason and I have not regretted my decision. I was sorry to see Sony gave up on the product.

EDIT: To clarify, as there seem to have some misunderstanding of some of my earlier posts: I'm talking about GPU support to achieve faster rendering. I use Sony AVC/MVC as that is the most useful for me.

astar wrote on 2/6/2017, 1:53 PM

"I have reasons to use an NVidia card due to the 3D software I deploy in my work"

So why not cater a separate machine to work correctly with Vegas? You already cater a machine in another software's favor. Makes little sense to me to expect Vegas to change, when you do not ask the 3d software to support OpenCL correctly.

I never understand why people get so hung on Chevy vs Ford or AMD vs NV. Use what makes the software work the best, and forget the childish Red team Green team allegiance.

Wolfgang S. wrote on 2/7/2017, 2:15 AM

Just to say so: to use Vegas itself for s3D one needs a Quadro card, and that is not available from AMD. So in one of my machines I have a Quadro for Vegas with nvidia 3D vision, in the later machine I have both an AMD and another Quadro card for s3D. While I do not recommen others to mix AMD and nvidia in one machine, since it is tricky and drives some issue, it can be done.

That SCS had not performed an updat for modern nvidia cards in the last years - very true. But that was near to the time when it became obvious that SCS may not wish continue the development of Vegas. That does not mean that this will continue.

Desktop: PC AMD 3960X, 24x3,8 Mhz * RTX 3080 Ti (12 GB)* Blackmagic Extreme 4K 12G * QNAP Max8 10 Gb Lan * Resolve Studio 18 * Edius X* Blackmagic Pocket 6K/6K Pro, EVA1, FS7

Laptop: ProArt Studiobook 16 OLED * internal HDR preview * i9 12900H with i-GPU Iris XE * 32 GB Ram) * Geforce RTX 3070 TI 8GB * internal HDR preview on the laptop monitor * Blackmagic Ultrastudio 4K mini

HDR monitor: ProArt Monitor PA32 UCG-K 1600 nits, Atomos Sumo

Others: Edius NX (Canopus NX)-card in an old XP-System. Edius 4.6 and other systems

animagic wrote on 2/10/2017, 6:52 PM

@astar: I have one transportable PC allowing me to work while traveling, hence my interest in CUDA support.