To further chime in on what Pat says, think of it like this:

Quadros excel at accurate rendering; Geforce cards are better for fast rendering (think primarily games).

If you're up for a heavy technical explanation from the manufacturer, there's a very in-depth document over here:

http://www.nvidia.com/object/quadro_geforce.html

There's evidence that a lot of Quadro functionality is locked away by drivers and that they can be hacked to enable the features of a Quadro card on Geforce hardware. But if anyone's interested in pursuing that line, obviously be aware that here be dragons. Fiddling with drivers to enable otherwise locked features could result in damage to your hardware.

It might be worth pointing out that one solution for accurate rendering could be to use the software openGL option within Avid settings for any final rendering, which should produce the best quality regardles of the installed GPU. Although not always convenient it is a viable finishing option for systems with only Geforce consumer cards and should theoretically equal or better the results from a Quadro GPU.

A very interesting post and I learnt a few things I never knew. I much prefer the scalable balanced load bearing philosophy of Avid's architecture to say that of BM's Da Vinci Resolve, which is heavily dependent on high end GPUs. Resolve is often a struggle on Quad i7 laptops with  only 8GB or even 16GB, whereas Avid is always (at this point) very usable

Hi Chris,

Many thanks for this post. A quesiton about codec handling as it relates to real time playback. Perhaps Shailendra could offer some thoughts or commentary on the following...

One thing that has been both curious and frustrating comes with how reliable or robust the playback is for a given codec, and how that reliability changes with the codec's container or wrapper.

Example:

If I play DNxHD essence wrapped in an OP-Atom MXF container, the playback is typically extremely robust, assuming one has all requisite resources needed to play that codec i.e. storage that supports the bit rate and frame size of the media and a CPU/GPU workstation that can decode it reliably. 

Yet if you take that same DNxHD essence and re-wrap it into a Quicktime container, playback can suffer sometimes. Same system, same storage, same CPU/GPU, different wrapper/container, and the performance changes. 

To some degree I know the answer, I think, maybe... and that it may be related to how for something wrapped in a any container that isn't OP-Atom MXF, then Media Composer is making a "call" to some other OS resource - perhaps the native Quicktime libraries in this example - and that "call" takes time, consumes CPU cycles etc. 

I know that the gradual move to write native library support for non- Op-Atom MXF containers such as Quicktime is an ongoing effort (the so-called generic plugin for Source Browser). I hope these efforts conclude very soon.

Reliance on OP-Atom MXF workflows lessens every day as workflows become more multi-tool focused, i.e. differing color platforms, online platforms, mastering platforms, and our workflows become more heterogenous in terms of container types and camera sources. Ultimately we urgently need better playback performance for just about anything that isn't OP-Atom MXF. 

But the question stands: why would same essence, different wrapper make such a notable difference?

Many thanks for your article.

Best,

Matt Schneider - PostWorks New York

I am interested in a difinitave "version matrix" on Media Composer for the minimum amount of GPU RAM required to enable GPU rendering (and ACPL).

I know, for example, that MC v8.5.3 requries at least 1.5 gigs of Video RAM to enable GPU rendering.  Somewhere prior it was 1 Gig, but not shure when the higher requirement hit.

Newer versions require more, and I think it would be a VERY helpful document to lay out by version the minimum RAM requiremtns to enable.  Or even just tell us the version ponts where the amount required change, and by how much.

Thank You,

Dave Hogan

Burbank, CA

Yes - a very interesting post - thanks. And I echo Dave Hogan - a version matrix would be helpful!

Mat Linnett:

Quadros excel at accurate rendering; Geforce cards are better for fast rendering (think primarily games).

....[T]here's a very in-depth document over here:

http://www.nvidia.com/object/quadro_geforce.html

Thanks for the link. Intersting reading but it's a 2003 document. It's unclear if the distinctions between product lines 14 years ago remain in effect today. 

I can't provide a citation but I've read in a couple places Quadros support 10 bit output whereas GeForces only support 8 bits. Don't know if this is or once was but no longer is accurate. Also, doesn't that distinction loose merit if one uses a video I/O card (e.g. AJA, BMD, Bluefish) & external monitor?   

Hi Matt

I think in part its about building an optimised workflow. Avid has always tried to lever the most from the least and one to do that was to lock down some of the variables. Thats partly why the Avid media is stored in such an "odd" way.

Its a method that helps with efficient playback. Ans os to does the wrapper and codec in use and I suspect why Avid was slow in making the change to supporting more codecs and more wrappers. The AMA plugin method was a neat solution to that but I don't think its as efficent as the build in media handling. And I can't see it ever being so.

Some other NLEs don't have a native media method and are able to manage a wider range of wrappers and codecs with a variety of performance levels. But none match Avid with native media.

Mercer:

It might be worth pointing out that one solution for accurate rendering could be to use the software openGL option within Avid settings for any final rendering, which should produce the best quality regardles of the installed GPU. Although not always convenient it is a viable finishing option for systems with only Geforce consumer cards and should theoretically equal or better the results from a Quadro GPU.

Interesting idea. But I have always read that there are potential visual differences between a GPU render and a CPU render.  I wonder if that would be an issue in such a situation as this?

Jef

Matt Schneider - Very sorry I didn't respond to your post back in November. Just surfaced from two straight documentary edits.

You are correct. As it was explained to me, the wrapper is everything. Contrary to logic, right?!?

Someone in the process of assisting me to write that AvidBlog used an analogy that explains it pretty well. Take a coin. Flip it in the air and catch it. Pretty easy, yes? It's a tangible thing you can interact with. Now deposite the coin into your bank account. You still have access to that amount of money, but you can't hold it in your hand the same way. It's no longer as easy.

I'd love to know more about GPU rendering under MacOS... under Windows it's pretty easy to pull up the compatible video card config file and see what you can use (and as some have done, venture to change it to add a different card or driver). However, this doesn't exist under MacOS that I'm aware. Furthermore, I found either an article or a post here on the forums (don't remember which) from a couple years back that said GPU acceleration on Mac had been completely disabled for a time due to an OS bug. I'd love to know if this is still the case! I just put in a new GTX 1080 on my Mac 5,1 tower and looking at my GPU monitor it looks like it's not being used at all. If MC isn't going to use it or use very little of it, I'm thinking about swapping over to a Quadro K5000 Mac edition instead which can be had reasonably inexpensively (I'll actually come out ahead after I offload the 1080) and has native OS support; the drawback being it's an older chipset and has much less computational power.

And just to further emphasize the wrapper issue discussed previously, I've found entire compatibility issues in other software (Sorenson Squeeze 11 in my case) based on wrapper as well. Encoded media was DNxHD 145 in both instances; it wouldn't see the video stream at all when it was wrapped inside an OP1a MXF, but wrapped in QT it was fine.

The hardware differences between Quadro and GeForce are minimal, mostly clockspeed, set more conservative for the Workstation cards. GeForce timings are set agressively to suit gamers.

The real difference is that you are paying for the driver software in the case of the Quadros CAD and similiar applications require high accuracy for positional reasons and vector calculations.

Consider the difference between a consumer NLE and MC. Much more development and testing goes into MC. The NVidia driver process is similiar.

It is true that MC is not as reliant on sheer GPU power than some other applications.

@gstrudier:

gstrudler:

Furthermore, I found either an article or a post here on the forums (don't remember which) from a couple years back that said GPU acceleration on Mac had been completely disabled for a time due to an OS bug

Happily, that's not the case.

More info: https://support.apple.com/en-us/HT208544

But since you're in *this* forum, you should know Media Composer doesn't use the GPU that much.

You can see some real-world benchmarks of GPU usage in macOS with other apps here:

• https://egpu.io

• http://barefeats.com

You probably know this, but in macOS High Sierra, there's a GPU monitor in Utilities -> Activity Monitor.

You can also see GPU usage in:

• iStat (https://bjango.com/mac/istatmenus/)

• XRG (https://gauchosoft.com/Products/XRG/)

I wouldn't give up on the GTX 1080, Just know Media Composer won't take advantage. It's R3D footage and certain third-party AVX plugins.

I should have been more specific; I meant that Avid had disabled GPU accelaration in MC on the Mac side, not MacOS itself. I also do have iStat, which is where I've been looking to check my GPU usage. I've got the BorisFX suite and uses pieces of it from time to time (Continuum, Sapphire, and Mocha) which I do believe uses the GPU some, but even so I think I might switch down to the K5000 for better support & compatibilty, especially considering that MC doesn't seem to use it much (if at all).

Interestingly enough in regards to drivers, which has long been heralded as one of the advantages of the Quadro series in comparison to the Geforce, the current Mac drivers from Nvidia seem to use the same between both series of cards...

Hi,

I finally got a sapphire pulse RX580 installed in my Mac Pro 5.1 - Avid MC 8.9.2 reports GPU accelerated fx is enabled now ( previously it told me that my hd5870 didn't have enough memory - 1GB) I'm using OS 10.12.6 Sierra .

As I wrote back in October I was having trouble with spectramatte keying (something that used to work very well in realtime) but unfortunately I have not noticed any difference at all ! 

Its a pity there's no official technical guidance from Avid, and very little user experience written about it.

The annoying thing is, MC makes a deal on start up of telling you it's enabling GPU acceleration, the info tab proudly tells you it's using GPU acceleration - but nobody can actually tell us what it actually is or how it actually works ☹️

flashavid:

 the info tab proudly tells you it's using GPU acceleration - but nobody can actually tell us what it actually is or how it actually works 

Ha, didn't realize that was conveniently there! Well I guess the good news is I'm seeing that as well, so apparently it must be making some use of my 1080. That also definitively (positively) answers the question of if the Mac version will take advantage of GPU acceleration; still don't know how much, but at least it does. Guess maybe I'll stick with my 1080 after all...