Jump to content




Photo

Posting in 4k


  • Please log in to reply
42 replies to this topic

#1 Paul Brenno

Paul Brenno
  • Basic Members
  • PipPip
  • 88 posts
  • Cinematographer
  • North Dakota

Posted 24 May 2016 - 12:32 PM

Just got the Lumix G7, am testing it out, but only have an 8G Windows Premiere Pro system. I havent shot on 4K yet, but other than shooting in 4k, just wondering about editing in standard HD, or do I need 16GB (?)

Any input is great....
  • 0




#2 Jan Tore Soerensen

Jan Tore Soerensen
  • Basic Members
  • PipPipPip
  • 156 posts
  • Director
  • Norway

Posted 24 May 2016 - 02:24 PM

An older Windows system is not going to function well with a 4k timeline. It probably won't be able to play the files in VLC. 

 

I haven't gone 4k yet, but when I do, which is probably next year, I am going for a Xeon setup with 64 gigs of ram. It is just that hard on the system. Keep in mind though, there are major differences within 4k. All the way from 50mbps to 400+. The more information, the better computer you will need.


  • 0

#3 Perry Paolantonio

Perry Paolantonio
  • Basic Members
  • PipPipPipPip
  • 344 posts
  • Other
  • Boston, MA

Posted 24 May 2016 - 03:02 PM

You actually don't need a ton of CPU speed to move 4k uncompressed. an i7 is plenty. What you need there is ridiculously fast storage. Most of our RAIDs can move close to 1.5GB/second, and we can play 4k easily. But it depends a bit on the software as well, how it caches, etc.

 

If you're working with compressed 4k, CPU (and possibly GPU) speed is a much bigger factor because the video needs to be decoded before it can be displayed.

 

With uncompressed files, it's largely about storage I/O. With compressed files, it's less about storage and more about CPU.

 

In either case, lots of RAM usually helps, but not always. Again, all depends on the software being used and what kind of caching it does to speed up playback.


Edited by Perry Paolantonio, 24 May 2016 - 03:03 PM.

  • 0

#4 Macks Fiiod

Macks Fiiod
  • Basic Members
  • PipPipPipPip
  • 433 posts
  • Director
  • OG from DC, Now in NJ

Posted 24 May 2016 - 03:42 PM

 

If you have the cash, look into the GTX 1080.

 

 

Most of our RAIDs can move close to 1.5GB/second, and we can play 4k easily. But it depends a bit on the software as well, how it caches, etc.

How many drives do you have running in that array to get it that fast?


  • 0

#5 Perry Paolantonio

Perry Paolantonio
  • Basic Members
  • PipPipPipPip
  • 344 posts
  • Other
  • Boston, MA

Posted 24 May 2016 - 04:10 PM

Most of them are between 8-12 drives, as RAID 5, using a dedicated controller card. Just standard issue off-the-shelf drives, none of that enterprise or raid-specific drive nonsense and definitely no SSD (too expensive per GB).

 

We are moving towards setting up a centralized storage system though, because feature films as 4k DPX are a data management nightmare. A typical feature is about 5TB for the scan, 5TB for the graded file, and if we do restoration, 5TB for the restored files. And that needs to be on fast RAIDs, which we have in all of our systems, but the file copy time to move between systems is a major drag. 

 

The centralized storage system will have 60TB of RAID 6 (or we may do XFS, still haven't decided that part yet), on an infiniband fabric, with 20-40gbps going to each machine over iSCSI mounts. So that avoids the overhead of typical networking protocols like SMB and increases the bandwidth to 2.5-5GB/second to each node (theoretical max). I don't expect the RAID will be able to keep up with those speeds, but it will easily handle a couple of 4k DPX streams at once. With iSCSI, it's a simple matter of mounting a volume on the scanner and scanning to it. Then unmounting that and mounting it in the Resolve for grading. That would write the output to another volume dedicated to that project, and when that's done, it could be mounted on the restoration system for final cleanup and mastering. All of the mounting/unmounting is done in software since they're all part of the inifiniband fabric - no more cables to trace, no more sneakernet. 

 

We just moved into a new office, so this project has been on hold, but it'm going to get back into it in a few weeks, I hope, once we're all back to normal here.

 

-perry


Edited by Perry Paolantonio, 24 May 2016 - 04:12 PM.

  • 0

#6 Landon D. Parks

Landon D. Parks
  • Basic Members
  • PipPipPipPip
  • 1626 posts
  • Producer
  • Cincinnati, Ohio

Posted 24 May 2016 - 04:12 PM

Much of the issue in playback is not processing power, but rather storage speed. A single 7200rpm drive will not play back a 4K file in something like DNx or Prores - at least not without a lot of pre-buffer. On my computer, I can manage to play back 4K H264 files from my GH4 and a single stream of DNxHR 444 12-bit in 2k. Anything over that stutters.

 

Running those files from my Raid, I can play back 1 stream of 4k DNxHR 444 or around 2 streams of 2k DNxHR 444 in real time. Above these numbers, my processor starts to struggle except on Premiere Pro, where the OpenGL helps in playback - but then only to the point that my 2 x 390x cards will supply the power. 

 

Frankly, even starting with 4K - I'd not stick with it. Downsampling the 4k to 2k would be much better option, and result in better picture quality, reduced noise in the image, etc.


Edited by Landon D. Parks, 24 May 2016 - 04:14 PM.

  • 0

#7 Paul Brenno

Paul Brenno
  • Basic Members
  • PipPip
  • 88 posts
  • Cinematographer
  • North Dakota

Posted 24 May 2016 - 08:00 PM

I have no budget, but lucky enough to get a G7, since ive been picking up more projects lately. My main challenge is in post, mostly due to not having a budget yet to upgrade.

Would 16G be better ?
  • 0

#8 Landon D. Parks

Landon D. Parks
  • Basic Members
  • PipPipPipPip
  • 1626 posts
  • Producer
  • Cincinnati, Ohio

Posted 25 May 2016 - 12:11 AM

Would 16G be better ?

In my opinion, I think the starting point for any editing PC should be 16GB of RAM. Mine has 64 via a recent upgrade, but then again I use after affects - which uses a much ram as you can throw at it. So, while 8GB should be fine for simple editing (even at 4K), RAM is so cheap now that an upgrade 16 will really help your system overall. 

 

If you're working with compressed 4k, CPU (and possibly GPU) speed is a much bigger factor because the video needs to be decoded before it can be displayed.

 

With uncompressed files, it's largely about storage I/O. With compressed files, it's less about storage and more about CPU.

 

I can attest to this. It also depends on the flavor of compression. For example, I can play back a 4K h264 file, but not very well outside of Premiere (which helps by also using the GPU). However, I can easily play back DNcHR 444 12-bit files with no issues whatsoever. It's odd because one would think that DNx would be much harder on a system than the consumer-level h264, but on my hardware I'm finding just the opposite. 

 

 

 

An older Windows system is not going to function well with a 4k timeline. It probably won't be able to play the files in VLC. 

 

I haven't gone 4k yet, but when I do, which is probably next year, I am going for a Xeon setup with 64 gigs of ram. It is just that hard on the system. Keep in mind though, there are major differences within 4k. All the way from 50mbps to 400+. The more information, the better computer you will need.

 

My base PC is circa 2012, and features a processor from that time period as well. I am running Windows 10, but along with my RAID I can easily get a 4K stream to playback just fine. 4K is more about speed of the storage drive than the speed of the processor. And descent 4+ core processor from the past 5 years should be adequate to play back 4K. If you're using Premiere Pro, then you can also throw a descent graphics card in the mix and and take advantage of the mercury playback engine. 

 

As for graphics cards, I'd still suggest the Radeon R9 390x over nVidia. Most editing programs can use either OpenCL or CUDA, and as such you don't need to stick to nVidia for most uses. A new 390x is currently around $300 - where the new 1080 from nVidia will retail for around $600.

 

 

 

 

ds


Edited by Landon D. Parks, 25 May 2016 - 12:13 AM.

  • 0

#9 Brian Drysdale

Brian Drysdale
  • Basic Members
  • PipPipPipPip
  • 4743 posts
  • Cinematographer

Posted 25 May 2016 - 01:45 AM

Proxy editing may be a solution on a less powerful machine.


  • 0

#10 Tyler Purcell

Tyler Purcell
  • Sustaining Members
  • 2352 posts
  • Other
  • Los Angeles

Posted 25 May 2016 - 02:29 PM

As for graphics cards, I'd still suggest the Radeon R9 390x over nVidia. Most editing programs can use either OpenCL or CUDA, and as such you don't need to stick to nVidia for most uses. A new 390x is currently around $300 - where the new 1080 from nVidia will retail for around $600.


Just an FYI, After Effects and Premiere use CUDA for almost all of their real-time rendering. Since nVidia are the only chip's that support CUDA, it's smart to use CUDA for over-all performance. Even DaVinci prefers CUDA support over Open CL.

http://create.pro/bl...eal-world-face/
  • 0

#11 Landon D. Parks

Landon D. Parks
  • Basic Members
  • PipPipPipPip
  • 1626 posts
  • Producer
  • Cincinnati, Ohio

Posted 25 May 2016 - 02:52 PM

Just an FYI, After Effects and Premiere use CUDA for almost all of their real-time rendering. Since nVidia are the only chip's that support CUDA, it's smart to use CUDA for over-all performance. Even DaVinci prefers CUDA support over Open CL.

http://create.pro/bl...eal-world-face/

 

After Effects uses CUDA only for it's built-in 3d renderer, which no one uses now that AE has Cinema4D built-in and Element 3D as a $100 plugin. Other than that, the remaining features of After Affects are mostly CPU based, and the playback is RAM based and not based on the graphics card at all. Mercury Playback is not available in After Effects, and as such any graphics-card accelerated stuff will either be CUDA for the built-in 3D engine (which again, no one uses), or any graphics cards for Element3D. Cinema 4D uses CPU and not the GPU to render.

 

In fact, The 390x from AMD has proven to be superior to even the Titan in Blender's Cycle (in many scenes anyway), and in Element3D I see a vastly superior performance on my 390x compared to what I did on my 980. And in fact, with Element3D, AMD better implementation of OpenCL means fast ray-tracing and AO in the shadows. 

Premiere Pro's Mercury Playback Engine is not CUDA specific. At one time it was, but many iterations ago they adapted OpenCL and OpenGL support for AMD, which works as well as CUDA. Under normal playback, I have never seen my GPU go about 20% usage for Premiere Pro.

 

Resolve is an interesting case... I see little use for GPU's inside of it. Neither my 980 or my 390x seem to have any effect on rendering times, and neither one helped at all with playback - since they Resolve doesn't use GPU for playback. 

 

AMD also has the advantage of graphics memory - 8GB in this case, compared to - what - 6GB on the top of line, non-titan cards - with most nVidia GPU's in the 390x price range being 4GB or less. When rendering, the amount of RAM on your graphics card is more important than how fast it is - because if the scene cannot be loaded, it can't be rendered at all.

 

As for programs that use simple GPU acceleration for effects like playback, I see little difference between a top of the line AMD vs nVidia. Both have the processing power to accelerate playback in Premiere Pro... And Resolve doesn't even have a playback assistant with the GPU - it's pure CPU.

 

Back in the day when AMD was having trouble with their OpenCL and OpenGL drivers, no one wanted to work with them. That has long since been corrected, and now most programs that were once CUDA-only are now OpenGL/OpenCL accelerated as well. 

 

FRANKLY, If I were buying a card today, I'd strongly consider the 1080 from nVidia... Back when I made the 390x purchase, the only thing AMD had that might beat it in number of cores and memory was the Titan X - at over $1,200, or 4x what I paid for the AMD. Now that nVidia has started to come to their senses and realized that $1,000 cards are stupid, I would consider them.... Though I'd point out that on a budget rig - a $600 1080 is still a lot of money when a $250-$350 390x will perform most everything as well, and allow you to invest the additional cash in something that will make a difference, like amount of RAM or better storage.

 

The reality is: There are very few programs that are CUDA-only any longer.... Blender Cycles was the largest hold-out, but that has since been opened up to GL/CL and now can show as fast or faster speeds on AMD cards. About the only thing nVidia has to it's advantage now is iRay (which I only seen used in Daz3D), and their name brand power - much like how Intel is claimed to be better than AMD in processors. 

 

Another issue with nVidia is that while CUDA use to be great, their latest drivers have been REALLY, REALLY bad. A google search will show that nVidia has taken a hit in the driver department, which means their implementation of CUDA is starting to wane with bad driver support. I have no doubt that Adobe says CUDA is better, but also keep in mind that article is from  2014 - and a lot has changed in two years. That article also says Premiere Pro is not OpenCL supported with Mercury Playback, which is no longer the case at all. 

 

Sorry, I'm an AMD guy. Always have been, probably always will be. Never seen an advantage to the Intel/nVidia combination. Used them before and noticed no increase in performance. 


Edited by Landon D. Parks, 25 May 2016 - 03:06 PM.

  • 0

#12 Landon D. Parks

Landon D. Parks
  • Basic Members
  • PipPipPipPip
  • 1626 posts
  • Producer
  • Cincinnati, Ohio

Posted 25 May 2016 - 03:06 PM

I'd also point out that if you're using a MAC or Final Cut Pro, it only supports OpenCL - not CUDA. So if you're on a MAC, nVidia might be a mistake to get.

 

Also, that article you pointed to above seems to be wrong on many levels. It could be because it's almost two years old - but Premiere Pro now supports OpenGL/CL for playback, Red Giant plugins now work with OpenCL/GL and not just CUDA... Sony Vegas also works with OpenGl/CL, and has since they first implemented GPU playback.

 

This article clearly points to a win on the AMD platform as well: http://www.dslrfilmn...rendering-test/

 

Quote from that article:

 

As for the results, the Radeon Sapphire R9-290x 4GB GPU scored the best out of the bunch with a render time of 4 minutes and 46 seconds and CPU usage at 88% in resource monitor. This puts the R9-290x ahead of the GTX 670 by a marginal 6 seconds.

Verdict:

Real time playback of the timeline in full resolution from both cards was virtually identical so it’s a tie in that department. While the Radeon Sapphire R9-290x 4GB did win the rendering test by 6 seconds, I would also call the rendering test an overall tie.


Edited by Landon D. Parks, 25 May 2016 - 03:11 PM.

  • 0

#13 Tyler Purcell

Tyler Purcell
  • Sustaining Members
  • 2352 posts
  • Other
  • Los Angeles

Posted 25 May 2016 - 03:12 PM

Some of your information is incorrect.

The Mercury Engine is CUDA ONLY, it doesn't exist on Open CL. So if you don't have a CUDA card, you aren't running real-time effects with your GPU.

DaVinci has no support for OPEN CL. The only real-time GPU hardware rendering support it has is with CUDA cards. Same goes for all of the JPEG2000 (RED CODE) conversion software like Red Cine X.

I do agree that video memory is a problem and 4GB (which is what most cards have stock) seems not to be enough. However, I've installed a lot of graphics cards (not AMD) in multiple systems, CUDA and Open CL. My results have been the CUDA machines always running a tiny bit faster. Whether it's multi-track real-time rendering, whether it's scaling or even simple effect application, the CUDA cards always perform better. I recently did a huge install of Open CL cards into some clients computers and I was unimpressed with the results. We wound up selling the cards and installing older CUDA cards and the results were night and day. This is something I've been struggling with for awhile since Apple is so Open CL based, CUDA is a very 3rd party thing.

Anyway, CUDA in my opinion is worth the money, even if you only have 6GB of memory.
  • 0

#14 Landon D. Parks

Landon D. Parks
  • Basic Members
  • PipPipPipPip
  • 1626 posts
  • Producer
  • Cincinnati, Ohio

Posted 25 May 2016 - 03:28 PM

Some of your information is incorrect.

The Mercury Engine is CUDA ONLY, it doesn't exist on Open CL. So if you don't have a CUDA card, you aren't running real-time effects with your GPU.

DaVinci has no support for OPEN CL. The only real-time GPU hardware rendering support it has is with CUDA cards. Same goes for all of the JPEG2000 (RED CODE) conversion software like Red Cine X.

 

Don't want to argue Tyler, but the documentation I have says you're wrong about this...

 

From Adobe's own help documentation, available here https://helpx.adobe....quirements.html, there is this quote:

 

 

 

Recommended AMD and NVIDIA video adapters for GPU acceleration

 

Windows OpenCL:

  • AMD FirePro M2000
  • AMD FirePro M4000
  • AMD FirePro M5950
  • AMD FirePro M6000
  • AMD FirePro S7000
  • AMD FirePro S9000
  • AMD FirePro S10000
  • AMD FirePro V3900
  • AMD FirePro V4900
  • AMD FirePro V5900
  • AMD FirePro V7900
  • AMD FirePro W2100
  • AMD FirePro W4100
  • AMD FirePro W5000
  • AMD FirePro W5100
  • AMD FirePro W7000
  • AMD FirePro W7100
  • AMD FirePro W8000
  • AMD FirePro W8100
  • AMD FirePro W9000
  • AMD FirePro W9100
  • AMD FirePro W4170W FireGL V
  • AMD FirePro M5100 FireGL V
  • AMD FirePro M6100 FireGL V
  • AMD A10-7800 APU
  • AMD Radeon HD 6650M
  • AMD Radeon HD 6730M
  • AMD Radeon HD 6750
  • AMD Radeon HD 6750M
  • AMD Radeon HD 6770
  • AMD Radeon HD 6770M
  • AMD Radeon HD 6950
  • AMD Radeon HD 6970
  • AMD Radeon HD 7480D
  • AMD Radeon HD 7510M
  • AMD Radeon HD 7530M
  • AMD Radeon HD 7540D
  • AMD Radeon HD 7550M
  • AMD Radeon HD 7560D
  • AMD Radeon HD 7570
  • AMD Radeon HD 7570M
  • AMD Radeon HD 7590M
  • AMD Radeon HD 7610M
  • AMD Radeon HD 7630M
  • AMD Radeon HD 7650M
  • AMD Radeon HD 7660D
  • AMD Radeon HD 7670
  • AMD Radeon HD 7670M
  • AMD Radeon HD 7690M
  • AMD Radeon HD 7730M
  • AMD Radeon HD 7750
  • AMD Radeon HD 7750M
  • AMD Radeon HD 7770
  • AMD Radeon HD 7770M
  • AMD Radeon HD 7850
  • AMD Radeon HD 7850M
  • AMD Radeon HD 7870
  • AMD Radeon HD 7870
  • AMD Radeon HD 7870M
  • AMD Radeon HD 7950
  • AMD Radeon HD 7970
  • AMD Radeon HD 7970M
  • AMD Radeon HD 8470
  • AMD Radeon HD 8550M
  • AMD Radeon HD 8570
  • AMD Radeon HD 8570M
  • AMD Radeon HD 8670
  • AMD Radeon HD 8670M
  • AMD Radeon HD 8690M
  • AMD Radeon HD 8730M
  • AMD Radeon HD 8740
  • AMD Radeon HD 8750M
  • AMD Radeon HD 8760
  • AMD Radeon HD 8770M
  • AMD Radeon HD 8790M
  • AMD Radeon HD 8870
  • AMD Radeon HD 8950
  • AMD Radeon HD 8970
  • AMD Radeon R7 265
  • AMD Radeon R7 APU
  • AMD Radeon R7260X
  • AMD Radeon R7M260
  • AMD Radeon R9 280
  • AMD Radeon R9 280
  • AMD Radeon R9 280X
  • AMD Radeon R9 285
  • AMD Radeon R9 290
  • AMD Radeon R9 290X
  • AMD Radeon R9 295X2
  • Intel Iris Graphics 5100
  • Intel Iris Pro Graphics 5200

 

So yes, OpenCL is used for GPU playback. In fact though, I don't believe there to be a such thing as 'mercury playback engine' any longer. I believe it is now called 'mercury transmit' or some such.

 

This article, http://www.dslrfilmn...rendering-test/, not only proves that Premiere Works with OpenCL/GL, but does so as fast or faster than nVidia in rendering.

 

As for Resolve, the very article you posted above shows this quote: 

 

 

 

Blackmagic DaVinci Resolve
  • CUDA Support
    • Real time colour correction
    • Real time de-noising
  • OpenCL Support
    • Real time colour correction

 

So, even at the end of 2014, Blackmagic did support real time color correction with OpenCL, just not denoising - which is not even a feature available in the free version. Although, I highly question this - because I have never gotten ANYTHING to playback in real-time in Resolve on either type of card. 

 

AS FOR RESOLVE, At least in Resolve 12 and higher, it does ALL the processing on the GPU - not the CPU. This is directly from their PDF setup file:

 

Screenshot%202016-05-25%2016.26.10.png

 


 

So according to their own documentation, they support AMD as well. 

 

I'm sorry, but I just don't buy into the nVidia brand as superior. There are still some apps that take advantage of CUDA only, but those are few and far between - and as AMD's drivers keep getting better, and nVidia's keep getting worse - we are starting to see a shift in performance to AMD. 

 

I have been a computer geek for over 20 years, have an A+ and other network certifications from CISCO, worked as a PC tech for years, and spent a lot of time building my own computers and computers for others - both AMD and Intel / AMD and nVidia... nVidia's dominance that once existed (and it DID exist), is no longer there. The dominance of nVidia (and their price point justification) is the same as Intel vs. AMD. Even though AMD processors have shown multiple times to do better at multi-threading than Intel, Intel still has this 'superior' feeling.

Maybe in Games, an Intel Chip and a mVidia card help, because there are some games designed to work best with CUDA... Though I would point out that there are a great many games designed best for AMD as well, like Tomb Raider. 


Edited by Landon D. Parks, 25 May 2016 - 03:35 PM.

  • 0

#15 Landon D. Parks

Landon D. Parks
  • Basic Members
  • PipPipPipPip
  • 1626 posts
  • Producer
  • Cincinnati, Ohio

Posted 25 May 2016 - 03:42 PM

Ultimately, in 2016 - I'd say the cards are not that much different. If I had a large budget, I'd probably go with an nVidia build for graphics, but only because nVidia cards are much easier on heat and power requirements, meaning I could run more than 2 cards in one computer, something that is hard to do with AMD cards. My 390x tops out at almost 500watts of power under full load, compared to what - 200 on the GTX 1080?
 

So while in my opinion AMD and nVidia cards will perform almost the same on most systems, nVidia does still have some advantage is certain areas - like running multiple GPU's and for some application that are CUDA only.

 

I'm simply pointing out here that the OP said he was on a budget - and as such for a budget build, I'd almost always suggest an AMD card. Even today, nVidia is twice as expensive on their new top of the line cards than AMD - though the price point is starting to come down. 


  • 0

#16 Tyler Purcell

Tyler Purcell
  • Sustaining Members
  • 2352 posts
  • Other
  • Los Angeles

Posted 25 May 2016 - 07:08 PM

The spec sheets are actually wrong. I've talked with the folks at Adobe and Blackmagic personally because I setup systems which are used professionally in the industry on a regular basis. So I need to know how to do that and what cards to use. I've had consultants for both manufacturers sit in on meetings to confirm my specifications before installing. These manufacturers never even consider Open CL as an option, it's not even on their radar. Software manufacturers have spent considerable amount of money developing code that specifically works with the CUDA cards. When you use open CL cards, you are NOT getting the use of that code.

Anyway, I've personally tested Radeon (AMD) Open GL cards vs nVidia CUDA cards of similar speeds.

So instead of telling people what you use works... why don't you do the testing yourself? I've done it. I downloaded samples from the internet of Red Code, Arri Raw, Cinema DNG, Pro Res XQ, i-Frame MPEG and .h264 and have manipulated all the codec's in DaVinci using different resolutions and the consensus was night and day.
  • 0

#17 Landon D. Parks

Landon D. Parks
  • Basic Members
  • PipPipPipPip
  • 1626 posts
  • Producer
  • Cincinnati, Ohio

Posted 25 May 2016 - 08:58 PM

First, I really don't think two separate spec sheets from two different companies are wrong. On top of that, I posted a third-party point of view showing that in Premiere, AMD was as fast or even slightly faster than nVidia in rendering.

 

However, I'm also not basing this on spec sheets, I'm simply using the spec sheets to backup my in-person claims. I have used all the adobe suite and their plugins, and Resolve/Fusion for a long while. I have used them across two separate setups - my old nVidia 980 and my new 390x setup.

 

In Premiere Pro, both cards performed the same across media ranging from 4k h246 (GH4 straight), DnxHR (Atomos), CinemaDNG RAW and Proress (Pocket camera). Playback is smooth across all formats and shows no dropped frames (except CinemaDNG - which dropped frames on both cards at about the same rate). In fact though, my 980 always showed about a 60% spike in Premiere, where my 390x never goes above 20%. So while performance was the same, card usage was much higher with the 980.

 

In After Effects, I have never seen any GPU acceleration on anything either card. I do not use the built-in ray trace rendered, and so the most important thing for AE is CPU and RAM.

 

In Resolve, I have never got real-time playback of anything except a straight flat h264 file. This applies to both cards. Once a single grade node has been applied I have to step down to a 1/4 proxy to view it in real time. Forget about viewing anything over h264 in real time in Resolve. I have never been able to make it happen with either card. Then again, Resolve uses the GPU for rendering not preview (like premiere), so I wouldn't really expect real-time playback at full-rez.

 

Going into Premiere Pro, I also use many of Red Giants plugins like Colorista, Looks, etc. These are ALL OpenGL and some OpenCL and not CUDA based processes, so nVidia holds no weight with those plugins. Again, I see no speed issues when using my 390x with those programs. 

 

Going into After Effects, I occasionally use Cinema4D lite, which is a CPU-only render. Because of that, I don't use it much. Most of my After Effects work involving 3D is done in Element3D - though that only tends to be my personal projects and stuff. I never had Element while running the 980, so I cannot test that - but I will say my 390x throws out a mean rendering speed - rendering full 2048 frames with high everything in a few seconds a frame. If nVidia did better I'd be shocked, but it wouldn't be worth the cost to me even if it did. Considering Element3D is also OpenGL/OpenCL and not CUDA based, this should not be shocking. 

 

In Blender (which I use a LOT), I have had mixed results in Cycles. It really depends on the scene. Blenders OpenGL/CL implementation has gotten a lot better - so much so that my 390x actually renders the race car test scene faster than a comparable PC running a Titan X. Some scenes though, Cycles still has issues with AMD. If your goal is using a lot of Blender for Cycles, then nVidia is likely still a better option for now.

 

In Daz3D (which I use for animation, book covers, etc.), nVidia rules. You only have the choice of CPU rendering or iRay, which is an nVidia technology. Because of this, I put my 980 in my other work computer - the one I use for book design and editing - and use it with Daz3D. In reality though, iRay is not that great - and I'd never use it for anything more than stills.

 

In Fusion, which I have been using A LOT for compositing some scenes for my project, I have only used the 390x, so I cannot speak for 980 performance. I will say though that considering Fusion does not use CUDA, I fail to see how nVidia would add any speed.

 

I hear Photoshop can use CUDA to speed things up as well, though I have honestly never seen anything slow about Photoshop running CPU only, so I don't know where that speed increase is suppose to come in at. And I use Photoshop on a daily basis. 

 

THE BOTTOM LINE is that nVidia is the wise choice if you're planning to use programs that use CUDA technology. However, if you're program can use CUDA or OpenGL/CL or only OpenGL/CL, then AMD cards should not show any speed issues in comparison. In fact, in OpenCL applications, AMD is a lot faster - due to AMD's better OpenCL drivers.

 

Now this is all from first hand experience working with these programs. I don't doubt your own experiences, but I have had my own and those are here. Like I said, I have been building computers and repairing them for years... And in my opinion, the whole nVidia vs AMD thing is about as good an argument as the Intel vs. AMD argument. The only advantage nVidia has is in CUDA-only apps or apps that partially use only CUDA, like Daz3D and Blender.... 

 

the big issue in programs that you plan to RENDER on, and that the program can use either card - is the RAM on the card. The more RAM the card has, the larger scene you can render.

 

Not really the subject of this thread: But the 390x blows the 980 out-of-the-water for 4K gaming in nearly game. The extra 4g of ram helps a lot. 


Edited by Landon D. Parks, 25 May 2016 - 09:12 PM.

  • 0

#18 Landon D. Parks

Landon D. Parks
  • Basic Members
  • PipPipPipPip
  • 1626 posts
  • Producer
  • Cincinnati, Ohio

Posted 25 May 2016 - 09:07 PM

I'll close by saying this: If I was building the system today, I'd likely go with a GTX 1080 (or two, probably). The GTX 1080 now has 8GB of RAM, which matches the 390 series. It also runs cooler and uses a lot less power, meaning I can stick more of them in my computer over time. The only issue I'd have is the cost. $600 is still a lot, roughly doubt the cost of a 390. So, if someone asked me about a budget build where they where using something like Premiere only, I'd suggest they use an AMD card. If you have $600 for a 1080, by all means go for it. I'm simply pointing out that my first-hand experience is that AMD cards perform the same or better than nVidia in non-CUDA application - which most video editing platforms are. 

 

Vegas has been able to use OpenGL since it first made the GPU acceleration available. Premiere CAN use OpenGL/CL for Mercury Transmit as well. It's been proven time and time again by not only me, and by third parties, but by Adobe themselves. Resolve is a special cookie, and I really don't expect a newbie looking for a cheap build to be running Resolve at all - so resolve performance is really not even a subject - though as I said, Resolve shows now GPU acceleration in playback for me, and in terms of render speed - everything renders the same on either card. Neither card uses more than about 50% of it's power or RAM in resolve. The problem with Resolve is that it requires a fast storage medium and faster CPU, since it doesn't do GPU accelerated playback like Premiere does. 


Edited by Landon D. Parks, 25 May 2016 - 09:10 PM.

  • 0

#19 Landon D. Parks

Landon D. Parks
  • Basic Members
  • PipPipPipPip
  • 1626 posts
  • Producer
  • Cincinnati, Ohio

Posted 25 May 2016 - 11:17 PM

Also, if you need more comparisons that AMD and nVidia are comparable in Premiere and Media Encoder, check out this video from Linus Tech Tips where he compares the cards:


  • 0

#20 Tyler Purcell

Tyler Purcell
  • Sustaining Members
  • 2352 posts
  • Other
  • Los Angeles

Posted 26 May 2016 - 12:23 AM

Right, but again, these tests are all exporting tests. I could care less about the speed it takes to encode an .h264 or Cineform video, both of which I rarely encode and have special tools to deal with. The only thing that makes any difference to me is when I'm editing. That's all I really care about because that's what the vast majority of people who own workstations for post production will be doing. They will be using a GUI to create using different software tools. Being able to playback 4k material in real time in DaVinci with multiple nodes of correction is critical. If you can't do that, your workstation is worthless because you can't see how things blend together. I just finished two 4k jobs, one shot with RED and one shot i-Frame MPEG. I colored both with DaVinci with multiple nodes, mattes, in full resolution. It always played back at real time with audio. With my old GPU (Radeon 5770 2GB video memory) it couldn't playback anything, you'd hit the spacebar and it would go to maybe 4fps. So anyone who says DaVinci is CPU only, again is only reading spec cards. The reality is, it's a VERY HEAVY GPU based system. Sure, on EXPORT it's CPU, but that's mainly because it can spread the load across multiple cores. On my system, playback barely makes the CPU's fluctuate, but export spikes them. Again, you only export when done and it can take weeks of work to finish a project, so how long it takes for that project to export is almost completely irrelevant. What matters is how the software works the rest of the week.

Now I edit in Avid, I've not made the leap to Premiere because it requires a newer operating system and I refuse to go there right now. Avid doesn't have ANY GPU support according to their documentation, but again, the addition of the GTX680 was night and day. Avid all of a sudden plays back more layers of video in real time without stuttering, it also allows me to run more composite effects without stopping. It actually performs like an all-new machine, which is pretty incredible since it's NOT a crazy new fancy graphics card and my machine is 8 years old!

I also don't work in consumer formats, I work with RED, Pro Res, Cinema DNG and i-Frame MPEG most of the time. Long GOP MPEG (.h264) and Cineform are the two last things I would ever want or need to touch. I transcode those codec's to Pro Res before editing with them.
  • 0


Abel Cine

Technodolly

Media Blackout - Custom Cables and AKS

The Slider

rebotnix Technologies

Glidecam

Rig Wheels Passport

Tai Audio

Paralinx LLC

Pro 8mm

CineTape

Ritter Battery

Broadcast Solutions Inc

Willys Widgets

Aerial Filmworks

Visual Products

Zylight

CineLab

Abel Cine

Broadcast Solutions Inc

Technodolly

CineTape

Rig Wheels Passport

Media Blackout - Custom Cables and AKS

Willys Widgets

Aerial Filmworks

Visual Products

Pro 8mm

Ritter Battery

Glidecam

Paralinx LLC

Zylight

CineLab

The Slider

rebotnix Technologies

Tai Audio