Activity › Forums › Creative Community Conversations › FCP X hardware performance
-
Walter Soyka
June 4, 2012 at 11:16 pm[Lance Bachelder] “Did you have Premiere set to use highest bit depth (32 bit)? I’ve found renders out of X to be super clean – colors come out exactly like you create them and Blu-rays look better than anything I’ve rendered out of any NLE. I don’t mind waiting a little longer if the quality is as good as it gets.”
Floating-point (32-bit) processing is great. It has three big advantages for visuals:
- Large dynamic range (the difference between black and white)
- Values above and below clipping (values brighter than white and darker than black, very useful when chaining effects)
- Very high precision (leading to reduced quantization or rounding errors, also useful when chaining effects)
With fixed point or integer math, the gap between any two adjacent values that the computer can represent is constant. With floating point, the gap between any two adjacent values is proportional to the number itself — about ten million times smaller than the number. In other words, large values are spaced further apart, but small values are spaced closer together, giving you incredible precision (read: more detail) in manipulations in the lower/darker end of the range.
For many straight editorial cases, though, FP processing can be speed-sapping overkill. You start with 8-bit (and occasionally 10-bit) acquisition, and you deliver 8-bit video. For straight cuts and dissolves, floating point processing offers literally zero benefit for all the additional time and resources it requires. Once you start compositing, chaining effects, or doing colorspace conversions, though, you start seeing improved accuracy.
As for colors looking like they were intended — that’s how it’s supposed to work! Previous versions of FCP (and QuickTime) had disastrous gamma and color management, not to mention somewhat byzantine split YUV (sic)/RGB processing pathways with questionable transforms. FCPX, its ColorSync support, AV Foundation, and fulltime RGB processing is a huge step forward here. (For the sake of completeness, I’ll add that FP processing is a big benefit for accurate RGB/YCrCb transformations and superblack/overbright preservation)
I’d further note that accurate and predictable color are certainly doable with other NLEs as well, but I don’t say this to take anything away from FCPX.
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 -
Jeremy Garchow
June 5, 2012 at 12:16 am[Walter Soyka] “In other words, I agree with you that Windows is the better choice for performance, but I disagree that it has anything to do with a tradeoff for stability.”
Ok. Fair enough. As I said, Walter, this isn’t true anymore. Windows has caught up in the stability department. That was my point earlier in this thread. It’s there to reread.
While agree that Apple markets grand central dispatch, and the “Core” APIs, it seems all of that is built not for prpfessional use (although pro’s can get a lot out of it, like fx factory) its really made so the whole OS can have animations like the “puff of smoke” when deleting something from the sidebar, or making all the app chiclets shake when rearranging app on a iPhone, or the way that everything slliiidddes around in the X timeline.
X’s export is sometimes faster then Pr. Is the GPU export of X float RGB? How do we know? Rendering in X seems to be which is why it’s so slow, but does the Share function share this function? If export is so fast and rendering so slow, why not make the rendering as fast as export? Isn’t it kind of the same process, just different intermediate files? Something is amiss there, but I don’t know enough about X and there’s little “quality control” exposed to the user like there was in every sequence of Legend. It’s not super cut and dry.
Also, I do not think that msoft favors speed over stability, that was not my argument. That’s not what I said or implied, just so we can get that out of the way. Apple and msoft have differing needs. Msoft needs to run the worlds servers (along with Linux), Apple does not. They bowed out of that game.
In my mind, Apple has been developing for small machines to go as fast as possible, but not have the fastest machine. In that race, Apple has done a lot of work, as has intel, but OSX has a lot to do with that as well. It’s efficient and portable.
As far as distributed computing, you’re right Apple had done a bit of work there, but thats not “fast” CPU computing per se, it’s actually spreading the load around to a bunch of smaller lower power machines to break up the work.
I will stop trying to compare Apple system stability vs msoft stability as that’s not what I’m saying exactly.
Henceforth, I will say simply, Apple is not in the CPU speed race. They want small and fast, which by default doesn’t mean the fastest available.
-
Jeremy Garchow
June 5, 2012 at 12:52 am[Frank Gothmann] “What I meant was that Apple isn’t slow to implement new hardware because it migh impose stability issues. hence my remark about Linux.”
Got ya. I didn’t mean to imply that at all. I do not think that Apple waits for that reason.
They wait because they can and aren’t in a CPU speed race on every new proc that comes out.
And since there’s no Linux NLE that works with AJA cards, that pretty much rules it out.
DPX is not necessary for my workflows, but that’s just me.
-
Walter Soyka
June 5, 2012 at 5:39 amJeremy, you know all the criticisms — some justified, some historical but now incorrect, and some totally unjustified — that you get when you suggest that FCPX is a viable solution? That’s what I’m seeing when I suggest that Windows as a viable solution.
Just like you, I’m trying to help the community here move past perceptions and stereotypes, and see that the situation has changed drastically while no one was looking.
[Jeremy Garchow] “As I said, Walter, this isn’t true anymore. Windows has caught up in the stability department. That was my point earlier in this thread. It’s there to reread. “
Yes, we are totally agreed on this. Windows used to be way, way behind on stability, but that is no longer the case.
[Jeremy Garchow] “Msoft needs to run the worlds servers (along with Linux), Apple does not. They bowed out of that game.”
Microsoft’s server OS is a bit different than their desktop OS, as is Apple’s (and Microsoft doesn’t design hardware whereas Apple does), so I’m not sure where this fits in.
[Jeremy Garchow] “I will stop trying to compare Apple system stability vs msoft stability as that’s not what I’m saying exactly.”
I guess I didn’t make it clear enough in my post, but I do know from our previous conversations about this that this is not what you’re saying, and I wasn’t trying to put words in your mouth. I’m sorry if that didn’t come across.
I did say “it may be unintended, but there’s an implication…” but what I probably should have said is instead that it’s easy for someone to incorrectly infer that if Apple prioritizes stability over performance, and if Windows is faster than OS X, then Microsoft must prioritize speed over stability.
[Jeremy Garchow] “Henceforth, I will say simply, Apple is not in the CPU speed race. They want small and fast, which by default doesn’t mean the fastest available.”
I can only half-agree with this. Apple is very inconstant in CPU speed race. Every Mac Pro is a speed demon at its launch, but after launch, competitors get updates with speed bumps while the Mac Pro stagnates. I see Apple’s commitment to CPU speed as stop and go, and as a user who likes OS X and cares about performance, I’ve found it absolutely maddening. It would be easier for people like me if Macs were either always competitive or never competitive on speed, but while I agree that Apple fails to consistently offer the fastest CPUs, I don’t see how you can deny that their workstations are competitive at their launches.
If Apple kills the Mac Pro, then small and fast will indeed be best description. If they release a dual E5 sizzle core beast Mac Pro next week, then the Mac Pro will again for a time be among the fastest systems available. Apple will be in the CPU speed race until Intel releases the next mid-architecture speed bump that Apple ignores and other manufacturers adopt, and this argument will continue another round.
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 -
Jeremy Garchow
June 5, 2012 at 4:11 pm[Walter Soyka] “Jeremy, you know all the criticisms — some justified, some historical but now incorrect, and some totally unjustified — that you get when you suggest that FCPX is a viable solution? That’s what I’m seeing when I suggest that Windows as a viable solution.
Just like you, I’m trying to help the community here move past perceptions and stereotypes, and see that the situation has changed drastically while no one was looking.”
The funny thing is though, Walter is that I’m not saying Windows sucks like people say FCPX sucks, as a matter of fact, I have said that the two OSes are probably fairly even. My argument is that Mac’s are slower and always have been, yet still I somehow feel that we are disagreeing and that I’m cathching sh*t for it! 😉
Macs are slower and have allowed significantly less simultaneous peripheral connections since I have been using Macs professionally for work. Mac used to advertise that speed was better, but rally what I found this to be was that Microsoft Windows needed more maintenance and needed to be reinstalled more than MacOSX (but don’t let me fool you about stability, cause that’s just ridiculous!). Once a Windows system was rebuilt, the speed returned.
[Walter Soyka] “Microsoft’s server OS is a bit different than their desktop OS, as is Apple’s (and Microsoft doesn’t design hardware whereas Apple does), so I’m not sure where this fits in.”
What I alluded to was that Microsoft has a completely different business model. Microsoft needs to support a wide variety of computer hardware (servers included). It is in their best interest to get as many copies of that software in to as many hands as possible through new computer/OEM sales, and through OS upgrades if the hardware can handle it.
Apple, on the other hand, has a very inexpensive OS, and once you purchase it, can pretty much be installed on any machine you own; totally opposite of Microsoft’s model. Apple is trying to get their hardware in to as many hands as possible through various methods. You don’t even need to own a Mac to have Apple hardware anymore, but I bet those little gadgets sell Macs.
[Walter Soyka] “I can only half-agree with this. Apple is very inconstant in CPU speed race. Every Mac Pro is a speed demon at its launch, but after launch, competitors get updates with speed bumps while the Mac Pro stagnates. I see Apple’s commitment to CPU speed as stop and go, and as a user who likes OS X and cares about performance, I’ve found it absolutely maddening. It would be easier for people like me if Macs were either always competitive or never competitive on speed, but while I agree that Apple fails to consistently offer the fastest CPUs, I don’t see how you can deny that their workstations are competitive at their launches. “
OK, I would see it as you whole agree, not half agree. Of course when Apple launches a new computer with brand new procs, it’s going to sort of compete with a similar specced windows machine (although windows still seems to run video applications faster). Apple doesn’t keep up with every single speed bump or proc refresh. That’s not their game. They don’t care about CPU speed enough to keep refreshing their line. See? We agree. Ha!
[Walter Soyka] “If they release a dual E5 sizzle core beast Mac Pro next week, then the Mac Pro will again for a time be among the fastest systems available. “
Yes, they will choose to use the newest processor as it probably has some architecture gains (perhaps Thunderbolt, perhaps USB3, and all the other upgrades that have been mentioned on the interwebs). But that’s it. It will fit in to their system, and once it fits in to their system and you can connect the things they want you to connect (like their fancy thunderbolt monitor) then speed won’t matter. They care more about the design, the architecture to support their “ecosystem”, and then CPU speed is a result of the advanced technology.
I hope they do make a MacPro type thing that could be tested against the ProMAX One or HP zWhatever so we can have a good test. The current barefeats one is older generation Mac vs newer generation PC. It’s not a “fair fight”. The Mac gets trounced.
Jeremy
-
Walter Soyka
June 5, 2012 at 5:56 pmJeremy, for a couple of people whose opinions are just a degree or two apart, you and I sure spend a lot of time arguing! A few details aside, I think we do agree on the big picture, here and in the other threads in which we’ve recently engaged.
I’m with you 100% if you say that Apple doesn’t care enough about performance to keep their workstation line current like every other computer manufacturer does (even when Intel is the one doing all the work).
I’m not with you if you say that Apple never prioritizes performance. You used some black-and-white language, but it’s an area where I see a couple shades of gray.
Yours sincerely, in PCs and PIOPs,
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 -
Lance Bachelder
June 6, 2012 at 12:13 amYes the 32 bit thing was just for Premiere since there is a checkbox to use highest bit depth. FCPX is using much more than just 32 bit float as you mentioned and the results are the best I’ve sen from an NLE. When I cut in Vegas I leave it in 8 bit mode for speed then switch to 32 bit float for color timing and final render – even if it’s an 8 bit render such as DVD the results are better with the switch though the render times skyrocket.
Lance Bachelder
Writer, Editor, Director
Irvine, California -
David Cherniack
June 7, 2012 at 9:47 am[Lance Bachelder] “FCPX is using much more than just 32 bit float as you mentioned”
He didn’t mention. So where are you getting the idea that FCPX is using much more than 32 bit float? And if you can show some stills of rendered material that clearly demonstrates that FCPX output is better than any other NLE it would help. Otherwise we might think it’s just Lance again, off on one of his enthusiastic, but somewhat delusional, benders 🙂
David
AllinOneFilms.com -
Oliver Peters
June 9, 2012 at 2:01 pmAs a follow-up to this test, I’ve plugged a Quadro 4000 into my 8-core MP. When I did the tests in the original post, I was using the ATI 5870. The 4000 has twice as much VRAM. Render/export is similar to what I had been getting. About 7 1/2 min. for a :65 HD clip with MB Looks applied, versus 8+ with the ATI. However, playback performance in the FCP X UI with these effects applied, but unrendered, is a bit better. More fps (visually) and better skimming. Accessing the external interfaces for filters like Sapphire Edge and Tiffen DFX seems like there is better response with the presets inside those custom interfaces. Otherwise fine, though I did have a crash on close that seems to be related to accessing a CUDA library plug-ins. I believe that some of these third party filters use CUDA acceleration when available, which might cause some conflicts with X.
– Oliver
Oliver Peters Post Production Services, LLC
Orlando, FL
http://www.oliverpeters.com -
Elias Huch
June 28, 2012 at 2:20 pmIn my experience with X over the last year I’ve found that the trick is to use Motion for this type of compositing. It appears that X is not really happy, as was stated above, with having to render multiple frames with several stacked layers (Tiff or not). However, when using a motion Generator my rendering is easy and quick like the simple Generators in X. I have always been an AE guy for graphics, but I’ve found that the “sort of” round tripping with X and Motion is actually pretty seamless. There are some challenges like publishing parameters in the right order, but if you publish the image to be changeable in motion and then duplicate it several times, you can do all of the compositing back in X with no trouble. It’s all a bit cumbersome, but the ability to not have to fully re-render is worth the trouble.
Elias Huch
Motion Graphics Designer
Reelvizion Productions
Reply to this Discussion! Login or Sign Up