Following my move to a HD workflow of the renown Sony PMW-EX1, I
found myself in need for a high end NLE system. My old(er) systems were
a Dual Processor 2.5Ghz G5 and a Dual processor dual core 2.6Ghz Xeon.
Someone offer to buy off my G5 for use as a photo editing system and I
used the opportunity to top up the sales to reacquire another mac
system... this time, the top end 8 core system (standard 2.8Ghz) with
the 8800GT GPU. In the past, the G5 and Quad core systems were more
than sufficient to work smoothly and quickly whilst in a SD workflow of
the Canon XL2. Once I moved on to the EX1 workflow.. everything slowed
to a crawl. I did my research well but not well enough apparently. This
is what happened.
The 8-core mac pro was a CTO system.. on designing this system.. i researched out the best bang of the buck approach...
1) 8 cores - 2.8Ghz.. the cheapest 8 core out there.. the price
jump from 2.8 to 3.2 was a lot and I figured better allocated to a good
capture/disk system...
2) 8800GT - a comparison between the 8800GT and 4500FX... about a
5% increase in processing power but doubling of price.. not quite worth
it... as I think the bulk of the cost of the 4500FX (i may be wrong)
goes into the 3D imaging components... which I doubt we will need for
FCP.
3) 16GB of 2GB 800mhz matched pair ram modules - Apple mac systems
have this thing called 512bit memory addressing whereby the memory
access speeds up a notch if you have all the slots filled with
identical ram and works similar to a RAID 0 system on disk arrays. I
noticed a noticeable improvement with renders with this. This is highly
recommended!
4) Lacie S2S with PCIe esata2 - the esata2 allows 3 gbit of
throughput.. and with a 4 disk RAID 0 system.. and 1.8 TB of space..
allows ample speeds up to: Read in MB/s: 143.4 / Writes in MB/s: 155.4
This in FPS R:W:
8bit YUV422 1080: 36/39
10bit YUV4222 1080: 27/29
10bit RGB 444 1080: 18/19
12bit RGB 444 1080: 15/17
8bit YUV422 720: 81/88
10bit YUV4222 720: 61/66
10bit RGB 444 720: 40/44
12bit RGB 444 720: 135/38
So.. for a 720p workflow up to 444, this array is sufficient for 25p work and up to 10bit YUV422 in the 1080p workflow.
5) Black Magic Design Decklink HD Extreme: The latest addtion to
the capture card world.. the first (IMHO) affordable high end capture
card for all things capturable. This card is simple amazing in that it
costs less than 1k and give you access to:
a) breakout box/interface that you can capture anything on the market today
b) HDMI outputs for using your home LCD for monitoring and
referencing (I color corrected a new Phillips Cineos 32PFL9432/98 for
this)
I initially bought this because I thought it would aid me in my quest for real-time previews for MB looks.. but.... see below...
Now.. one would think this system would be no problem with
real-time previews... there is very little in terms of bottlenecks
here.. fast CPU, GPU, disk array, monitoring.. etc. However, I was
wrong.. the only thing i forgot to include in my research was FCP.
OMG.. when I rendered a timeline... without any filters or with a few
low level correction filters.. it was ok.. but once a color corrector
like MB looks was put in.. the Decklink card would hold the frame..
unable to output the subsequent frames... I spoke to my apple dealer
and researched more into this.
As a result.. I only realised that FCP was still a 32bit
application and could not access more than 4GB of ram and only 2.5gb
were actually used for the sequences. On top of that.. FCP's FXplug
uses OSX's core image technology for its rendering and hence was pegged
to Leopard and does not utilise the GPU directly like a window's
system. Thus with 32bits desperately trying to submit work to a 64bit
OS.. there was the root of all problems.. FCP isnt spawning enough
threads to make use of the immense power at its finger-tips... go
figure.. hardware technology is still far ahead of software..
So before you go round snapping up the latest and fastest.. think
about where the possible bottlenecks would be.. even in things like the
architecture of your NLE and plugins.
