Sharing is caring - Graphics card comparison

[rueyloon]

This post is dedicated to the video editors who is thinking of ways to extend the life or upgrade your machine to make you daily toil more enjoyable, and also to get your editing done AFAP so you can just leave the studio sooner.

The 3 exports represents the 3 most used in my daily exporting
1) Publish to YouTube - Up to 4K
2) Apple Devices - HD Best Quality
3) MPEG - MPEG-2 Program stream, 15Mbps

I was thinking of creating a eGPU for the macpro, but the numbers from the R9 390x does not look promising even extra polating an extra card the numbers looks like it will help the OMP reach near the NMP, but probably not over take it. The other option is the 980ti, but I'm not sure about its Open CL performance.

I see alot of bench mark, but no one seems to be using real numbers to compare the specs. With each version of OX and updates to Compressor and FCPX the performance is going to be different, so I feel benchmarks like Bruce5k does not give a clear comparison as even my own system gave different numbers with different versions of Compressor etc...
[doublepost=1482429876][/doublepost]Someone here is also experiencing very minimal performance upgrade after installing a 980ti gpu

Maybe apple apps just integrate better with their default hardware? Or is there some next that I need to edit?


https://forums.geforce.com/default/topic/846114/geforce-apple-gpus/gtx-980-ti-on-mac-pro-5-1/

 

[ActionableMango]

Many video related tasks rely mostly on the CPU. The GPU will help only for specific tasks that are GPU accelerated like scaling and certain filters or effects.

For example (I could be mistaken), but I believe Compressor is 100% or nearly 100% CPU accelerated, and will not benefit at all from a faster GPU. But it will benefit massively from additional CPU cores and use every CPU clock you can throw at it.

 

[slughead]

This post is dedicated to the video editors who is thinking of ways to extend the life or upgrade your machine to make you daily toil more enjoyable, and also to get your editing done AFAP so you can just leave the studio sooner.

I'm guessing You're using FCPX?

I don't have a lot of info for you, I contacted the guy from barefeats.com (really nice guy, responsive to email). My impression is that even though the 980 is twice as fast the software is definitely optimized for the D700 and therefore that card does better. To some extent the other ATI cards do okay as well. However, the 7970 and I believe the 280x show up as a D700 in OS X.

A couple of things:
- Getting FCPX to use all your cards may be an issue, I read years ago that they had to be matched (may not be the case anymore)
- with eGPU you might not be able to do more than one at a time
- Does exporting use GPU? I thought that was more CPU bound

I honestly have no idea but before making a purchase decision I'd make sure you know all that, the barefeats guy can certainly help

 

[h9826790]

For example (I could be mistaken), but I believe Compressor is 100% or nearly 100% CPU accelerated, and will not benefit at all from a faster GPU. But it will benefit massively from additional CPU cores and use every CPU clock you can throw at it.

If the project send directly from FCPX to Compressor. It will use GPU for encoding. However, I have no idea how much difference, and what the GPU use for. (From memory, this function only exist since from something like FCPX 10.2 and Compressor 4.2)

Anyway, even the GPUs can be used, the export performance still very CPU limiting.

However, the 7970 and I believe the 280x show up as a D700 in OS X.

7970 or R9 280X will show as 7xxx but not D700 if not flashed. Or labeled as whatever you want in the EFI / kext.

 

[linuxcooldude]

If the project send directly from FCPX to Compressor. It will use GPU for encoding. However, I have no idea how much difference, and what the GPU use for. (From memory, this function only exist since from something like FCPX 10.2 and Compressor 4.2)

I did a quick send to Compressor render test just for this post. Using the the 2013 Mac Pro with Dual D700 graphics With latest FCP X & Apple Compressor.

A 2:40 minute video took 2:51 minutes to encode which is almost real time rendering. Using iStat Menus both D700's were used under constant load which varied during the encoding process. Since version 4.2 Compressor does support GPU rendering when using "Send to Compressor" in FCP X that also supports dual GPU's.

Now using a classic Mac Pro with Dual GPU's, your milage may vary depending on video cards used and other factors. I heard of Barefeats doing testing using dual GPU's (Such as the Apples Dual AMD 5770's )

 

[h9826790]

A 2:40 minute video took 2:51 minutes to encode which is almost real time rendering.

How about if switch off the GPU assist in Compressor? How much time it cost for the same video to export?