-
Article about Performance in FCPX
Read something interesting in an article on the Ken Stone blog. The article is actually a fairly long winded review of a Raid product and can be found here:
https://www.kenstone.net/fcp_homepage/accusys_raid_pci_express_douglas.html
Now it should be noted that the Ken Stone blog is VERY pro FCPX.
About 2/3 of the way into the article the author, Steve Douglas, does a test to see how many streams of PIP ProRes he can play off the raid.
“In Final Cut Pro X, I was able to play 3 streams of Pro Res 422 at high quality and with the Dropped Frames warning checked in the FCP X preferences. While the render bar showed, I was able to play these 3 streams constructed as a picture in picture without the dropped frames warning coming on. No rendering was ever done.
Once I placed a 4th stream, the dropped frames warning appeared as shown below. This test was repeated several times with different streams and each test ended with the same result.
However, once I changed the Final Cut Pro X preferences to ‘Better Performance’, I was able to place at least 8 streams of picture in picture video without a single warning and could have easily built more streams into the viewer.”
He then does the same test in Legacy. Here’s where it gets interesting.
“Next I moved on to seeing what I could do in Final Cut Pro 7 feeding off the Accusys Raid. Here is where my biggest surprise came to the forefront. With Final Cut Pro 7, I was expecting to get fewer streams of Pro Res 422 video to play in real time without dropped frames due to its being a 32 bit application as opposed to Final Cut X’s 64 bit. Instead, and with FCP 7’s real time settings set to High Quality, not Dynamic, I was able to place 10 streams of picture in picture video in the timeline and it played perfectly. Through the MXO2 box onto the external monitor it looked just fine and played without a single stutter. I am positive that I could have added even more streams and it still would have played like a knife through butter, but I was running out of picture in picture room on my canvas.
My big question then was why I could play fewer streams of video in FCP X than I could FCP 7? This surprised me. What I’ve learned is that FCP 7 relies entirely on your processors and RAM for playback and I have 20 gigs of ram installed.”
Now for the quote that struck me:
” FCP10 is a totally different application. This app was written to deliver excellent performance even with smaller computers and slower (or less) processors, and, for this, it relies on the extra power of the graphics card. In order to have access to this extra horsepower Apple uses Open CL, which some consider to be a better technology than CUDA. NVIDIA cards, like the NVIDIA FX4800 card I am using, are good for CUDA but not very strong in Open CL support, thus, they are less suitable for working with FCP10.”
Can anyone shed more light on this description of the performance behavior of X vs Legacy?
Herb Sevush
Zebra Productions
—————————
nothin’ attached to nothin’
“Deciding the spine is the process of editing” F. Bieberkopf -
14 Replies
-
I guess it boils down to X being heavily optimized for Open CL and performs better there while CPU is a supporting but not the main factor. You know, smaller, lighter, mobile devices. That plus immature code given that AV Foundation is, just like X itself, in its infancy.
I am not sure the whole 32 vs 64 bit thing comes that much into play here. The other day I was playing around at a friend’s place with Edius 6 which, of all the bigger NLEs out there that are still in development, is still 32 bit only and the performance of that thing in terms of RT and render speed even with AVCHD is simply mind-boggeling without any CUDA or Open CL help at all.
I guess a lot is related to how good the engineering behind it all is. -
Herb, these quotes from the article aren’t making much sense to me.
[Linked Article] “This surprised me. What I’ve learned is that FCP 7 relies entirely on your processors and RAM for playback and I have 20 gigs of ram installed.”
As the author pointed out, FCP was a 32-bit application, which means it can only address a theoretical maximum of 4 GB of RAM.
Think of RAM like timecode; if your counter doesn’t go beyond two digits each of minutes, seconds and frames, you can’t edit to or from a point in the second hour of the program.
Likewise, FCP cannot use 20 GB for caching because it can’t count that high.
FCPX, being 64-bit, can access all that RAM, and that’s the critical difference between a 64-bit app and a 32-bit app. They’re not faster; they’re just capable of addressing vastly larger amounts of RAM.
NLEs (FCP/FCPX included) rely on fast disk access for playback. Put a couple ProRes file on a USB thumb drive, and you’ll get dropped frames, even on systems with loads of system RAM.
By comparison, compositing systems like AE and Nuke rely on RAM for playing rendered frames like a flip-book. They are incapable of guaranteed realtime playback of anything that’s not rendered and cached.
[Linked Article] “FCP10 is a totally different application. This app was written to deliver excellent performance even with smaller computers and slower (or less) processors, and, for this, it relies on the extra power of the graphics card. In order to have access to this extra horsepower Apple uses Open CL, which some consider to be a better technology than CUDA. NVIDIA cards, like the NVIDIA FX4800 card I am using, are good for CUDA but not very strong in Open CL support, thus, they are less suitable for working with FCP10.”
I’d say that FCPX was written to be scalable. This means it will still run on lower-powered systems, but it will run better on higher-powered systems. The author is almost making GPGPU (general purpose processing on the graphics card) sound like a crutch to let FCPX run on cheap computers, but it’s really just intelligent use of resources.
I won’t get into whether OpenCL is “better” than CUDA, because that really depends on what “better” means to you.
As for performance, it depends on what the bottleneck is. Is it disk access? Is it decoding the clips? Is it compositing them together?
I don’t know for sure about FCPX, but other GPU-accelerated apps still run decoding on the CPU. An FX4800 (which was the top of the line NVIDIA GPU for the Mac three years ago) is not the best card for FCPX, but it would’t accelerate ProRes decoding on Premiere Pro or Davinci Resolve, either.
GPUs are often used to accelerate other tasks like transformations and effects, though, so a more FCPX-friendly GPU may or may not help, depending on the nature of the bottleneck.
Walter Soyka
Principal & Designer at Keen Live
Motion Graphics, Widescreen Events, Presentation Design, and Consulting
RenderBreak Blog – What I’m thinking when my workstation’s thinking
Creative Cow Forum Host: Live & Stage Events -
Walter –
Thank you for the, as always, enlightening explanation of 32 bit vs 64 bit.
Follow-up question, is the 4 GB limit per application, or is it 4 GB per core or per thread? Also, Would you care to speculate about why Legacy could playback more streams of the same data from the same raid on the same computer?
Herb Sevush
Zebra Productions
—————————
nothin’ attached to nothin’
“Deciding the spine is the process of editing” F. Bieberkopf -
[Herb Sevush] “Thank you for the, as always, enlightening explanation of 32 bit vs 64 bit.”
You’re welcome! There’s a lot of hype around 64-bit apps, so I think it’s important to be clear on what they’ll do for us and what they won’t.
I did oversimplify a bit. For the sake of completeness, there are a couple other notable benefits to 64-bit processing: more (and wider) CPU registers. Some tasks can be optimized by keeping more data in these registers on the CPU itself, instead of wasting cycles moving data in and out of system RAM — kind of like how a computer with limited RAM swaps out to virtual memory on hard disk.
In practical terms, though, larger memory addressing is the big user-facing advantage.
[Herb Sevush] “Follow-up question, is the 4 GB limit per application, or is it 4 GB per core or per thread?”
Per application.
[Herb Sevush] “Also, Would you care to speculate about why Legacy could playback more streams of the same data from the same raid on the same computer?”
Wild, uninformed speculation? Maybe it’s the new 32-bit floating point effects and compositing pipeline. There may still be room for optimization.
If nothing else, compositing with 32 bits per channel instead of 8 instantly quadruples the amount of memory required to store a frame of video as well as the minimum amount of memory bandwidth necessary for compositing in real time. If the compositing is done on the GPU, you have to add time to move the image buffers from system memory onto the GPU, and time to receive the composited image buffer from the GPU.
I’ll be curious to know if that same computer and RAID could play more streams via the new multicam feature. If there’s a different processing pathway, it may be capable of more streams.
Walter Soyka
Principal & Designer at Keen Live
Motion Graphics, Widescreen Events, Presentation Design, and Consulting
RenderBreak Blog – What I’m thinking when my workstation’s thinking
Creative Cow Forum Host: Live & Stage Events -
Recently, I’ve found myself wishing that FCPX would use more of my CPU! It’s amazing what it can do on a newer Macbook Pro with the GPU, but when I play back some unrendered stuff and it struggles, I wonder why the CPU is only hitting 10%. When I export video, the CPU really cranks up, which is satisfying.
Jason Jenkins
Flowmotion Media
Video production… with style!Check out my Mormon.org profile.
-
I think too many companies put too much into the whole 64bit thing which on it’s own doesn’t make a program run any faster. I think right now it is actually dragging them down. I’m also not sure why Apple felt slower machines couldn’t handle FCP7. I ran FCP7 on a dual core iMac like there was no tomorrow. Sure the rendering to compressed formats was a bit slow but the actual process of editing was really like slicing through butter. I now have a brand new 17″ MBP quad core that is a dream with FCP7 but is sluggish with FCPX. This is a top of the line Apple system and about the best laptop you can get.
There was nothing wrong with 32 bit software and being limited to 2 GB of ram. FCP7 did just fine and the only thing they really needed to enhance was the render speed. People were editing big budget features on that little 32bit app.
FCPX seems to be taking this new approach of making software bloated and just keep throwing Ram and cores at it to make it run better (cough, Premiere, cough). It is insane to think that a program should have 16GB+ of ram just to do what programs 1 year ago could do with 2GB of ram. Personally I think FCPX is just trying to do too much on the timeline at one time with all the metadata stuff and background processing. Even if background processing isn’t working all the time the program still has to have all those hooks to look out for certain things.
-
[Thomas Smet] ” It is insane to think that a program should have 16GB+ of ram just to do what programs 1 year ago could do with 2GB of ram.”
Absolutely – up until FCPX came along I felt my MacPro was doing pretty well with any current app that I threw at it (including high performance 3D apps and suchlike).
All of a sudden 8GB of RAM is way too little. What’s with that????!!!!
OK, I get it. Someone wants me to think about buying a new machine …
Is Apple going to be selling the new machine that does everything else I need it to do? Who knows? I can only hope so …
Simon Ubsdell
Director/Editor/Writer
http://www.tokyo-uk.com -
Which is why I don’t intend to spend a lot of money on hardware in the next few months – unless, of course, I have too.
Tim
-
This doesn’t surprise me at all. I definitely don’t see any blazingly performance of media with FCP X versus FCP 7. Skimming is a nice parlor trick that takes advantage of RAM, but actual media playback of optimized media files has more to do with the drive performance and internal “pipes” than anything to do with 64-bit-ness. I could get the same performance in older 32-bit versions of Media Composer, which handled media better than FCP 7 or FCP X. Typically about 5 streams of ProRes layered with 2D PIP is the norm on a decent machine with FCP 7. Sometimes more. Media Composer gets about 7 with DNxHD.
A lot of folks have a tendency to talk about working with native H264 files from a Canon 5D or 7D in FCP X and are amazed when it performs better than doing the same thing in FCP 7, which doesn’t handle that well. Once you start comparing ProRes to ProRes or uncompressed to uncompressed and the equation changes.
– Oliver
Oliver Peters Post Production Services, LLC
Orlando, FL
http://www.oliverpeters.com -
[Walter Soyka] “f nothing else, compositing with 32 bits per channel instead of 8 …”
Walter,
Are you saying that FCPX do everything in 32bRGB?
As I interpret FCPX papers, 32bRGB happens when there is some color management going on, but not for general or motion effects. At least that’s my impression.
rafael
Reply to this Discussion! Login or Sign Up