HD For Indies

A few new things going on over in the Red One camp in the last week-

1 - Red Drive is vastly expanded in capacity - not 40 or 80GB, but 320GB for under $1000

2 - Geoff Boyle of CML (Cinematographer's Mailing List) posted a comparative exposure lattitude test including Red, SI, F950 and others; Red appeared to compare poorly with only 8 1/2 or so stops of lattitude, BUT

3 - After some debate, Red figured out Red team had made an error in the test imagery as presented, and is showing a different testing methodology that shows much greater lattitude, matching their earlier analysis.

4 - A new image of "Spike" - what the Red team is calling their actual, hand-holdable prototype for the Red One camera.

First up - Red Drive is now a 320GB RAID -

1 - One new spec... - DVXuser.com -- The online community for filmmaking

The RED DRIVE is now a mini-RAID for under $1000. It is a better choice because it offers higher speed, the ability to use the entire drive for recording and has a 320GB capacity. The higher speed gives plenty of elbow room for REDCODE RAW and 4 channels of audio... plus metadata.

-Jim Jannard

So the good news is that the price has held, yet he capacity has gone up 4 to 8 fold (we were originally told that worst case was 40GB for $1000, then recently 80GB was mentioned on the Red website).

With 320 GB capacity, formatted capacity would be 296 GB. At the stated 27.5 MB/sec of 4K@24p Redcode RAW compressed, that would be almost exactly 3 hours of recording time.

At 4K@30p (34.375 MB/sec), that would be 2 hours 23 minutes.

That's a goodly chunk of recording time - for most narrative efforts, that is more than enough time for a full day's recording that can be offloaded overnight.

If you're shooting reality or something else that needs a lot of footage per day? Even if you shot 9 hours of footage a day, 3 Red Drives would be enough.

CML's LATTITUDE TESTS:

Geoff Boyle of CML took the time and trouble to run some tests on the posted Mysterium frames that Dave Stump shot last month at Red's offices. Geoff opened up each image, found the 18% grey square, and recorded that value on a 0-255 8 bit scale. Since the sequence of images were shot in 1/2 stop exposure increments, he just mapped where the 18% grey measured for each frame. So the horizontal axis is as the exposure is increased from one sample frame to the next, and the vertical axis is just the brightness measured. If it helps, instead of 0-254, just think of it as 0-100%. That is why everybody clips at 255 - pure white. The sooner it clips, the poorer the response. Here's that chart. 8 bit images were what he had to work with, but it is worth noting that at the bottom end, working with 8 bits can be misleading - you're dealing with the noise floor, so all the info you can get the better.

The Red team, and Red supporters on CML, had two areas of question/concern:

1.) The indicated 8 1/2 stops of lattitude was well below what Red was finding inhouse, and

2.) The Red is still in development and was being compared to shipping cameras without pointing out that difference (since corrected)

After much running around and many comments from all parties involved and interested on CML with the usual bickering, it turns out Red made a mistake in the processing of the files as provided - Jim Jannard (founder or Red) said on CML:

The mystery is solved and I am the one with egg on my face (I certainly don't like it when this happens). We have been turning this place upside down to figure out why the disparity in tests results... Graeme discovered this morning that the images shot by David Stump and posted on CML were processed through the REDCINE prototype with incorrect settings. It was NOT Geoff's testing and it was NOT the way David shot the images. I made an error in processing. Simple as that. Ugh.

So the test was valid, the data provided was incorrect/improperly processed, truncating the information needed for best results.

RED'S TEST - USING STOUFFER 4110CC CHART

Red, after having figured out what went wrong, found another test that they thought quicker and easier to evaluate dynamic range a Stouffer 4110CC wedge chart (that page references a 21 step chart, Red used a calibrated 41 step chart), and posted their own new test chart over on DVXuser.com - lots of discussion ensued.

Red's using the same methodology as used by dpreview.com, aka Digital Photography Review. A single image is taken of a test chart that has 41 carefully calibrated squares that go from white to black from one end to the other. The goal is to be able to discern as many squares from each other as possible. When you can't tell the difference between two adjacent squares in the captured image, your sensor can't differentiate, therefore is outside of the capabilities of the sensor (tautology, excuse me). The good thing is that it is one frame required, is quicker, can be automated to create the chart, and has less possibility of human error or calibration error due to ND filters, marks on lenses, etc. I've exchanged numerous emails with many of the parties involved and read numerous posts on CML and DVXuser about the pros and cons of each method. Both testing methodologies mentioned here have their uses and strengths, but I don't feel like going into all that arcana here - go read CML or DVXuser if you feel compelled.

I wasn't sure what I was looking at this chart, so I called Graeme, who did the tests and built the chart. Here's my takeaway based on that conversation:

If you look at it, its a diagonal line going up to the right. What the hell does it mean? It is basically a reading of how much light was shown to the sensor, and how much the sensor could register. The vertical lines indicate the floor and ceiling of meaningful response - to the left of the left vertical line, the sensor wasn't responding to the low level of light - it couldn't discern between pitch black and a little bit of light. On the right side, beyond a certain level, the sensor just saw pure white, and any more light was just as pure white. So in between there was the zone where the sensor worked - it could discern light from dark. The total range of sensitivity is about 11 1/3 stops. At the high end, there appears to be a roll-off, the otherwise generally straight line starts to taper off to flat - so a BIG increase in brightness shown on the sensor only generates a slightly higher registered value. (Graeme corrects me on this one - "There's not actually a roll off in real life there. It's just that one data point is just below clipping and one just after and the graph is interpolating between them. It's actually linear up to clipping. That's a graph artifact.") At the bottom end, the graph lifts off at a certain point but records the same signal level for about a stop - not terribly useful information, this is down around the noise floor. So the MEANINGFUL part of the chart is the part that is pretty much a straight diagonal - about 10 stops here.

Bottom line - using the same metrics everyone else does, this test indicates about 11 1/3 total stops of exposure lattitude. BUT...with the bottom all noisy and the top blown out, that isn't the part you'd do good work with. The USEFUL exposure lattitude, based on this chart, seems to be around 10 stops - the flat part of the cart. Both of those numbers are significantly higher than the 8 1/2 stops indicated by the erroneously processed data that Geoff Boyle had to work with. So dynamic range is significantly greater than 8 1/2 stops, definitely.

LATEST IMAGE OF "SPIKE" - THE REAL RED ONE CAMERA (not just an imager test platform)

Remember how Red is always making changes? Here's the latest fully done-up rendering of Spike (what they are calling their prototype cameras), less an LCD or a viewfinder. No, you probably wouldn't shoot with it configured thusly unless on a major production - this just shows ALL of the Red Cage & Red Rail options done up:
(click for larger view)

There's also a lengthy thread over on DVXuser about this image and it's ramifications as well - Spike... latest render - DVXuser.com -- The online community for filmmaking

If you've been following the evolution of the Red camera, you'll notice the thicker (2 drive I believe) Red Drive at the back, with the battery behind it. Lots more buttons on the control surface on back of camera. Rubberized grips on Red Cage as shown in LA last month. Hand grips can be tilted both out and forwards/backwards. Note discontiguous rails at bottom - as sturdy as solid? Solid/contiguous likely as well, nice to have adjustable options. Note mounting holes eh-va-ree-where. Also looks like this one's configured with the high speed optical port. Since the battery stacks on the back, leaves it open to have bigger batteries attached. In this config, access to rear panel is awkward to get to since the battery pack is in the way. But you can rotate the battery pack down, or even underneath the rail since that mount has two pivot points. This thing is really turning into a Lego kit - build it any way you want. It'll be interesting to see what "best practices" emerge for what shooters in what circumstances.

-mike