Then I buy a BenQ HT4050 and a screen and all of a sudden judder is a problem . Watching Gravity and The Martian was horrible. The color and the contrast was amazing, and the size of the movie still blows my mind, but every time the screen hitches I have a mini-stroke. I may have to turn on frame interpolation and see how it goes.
DLPs have much better motion handling capability than LCD TVs so seeing judder more pronounced doesn't surprise me. DLP is just exposing the judder, but fundamentally, it's there.
240hz LCD can't transition fast enough and it's likely only 120hz with maybe a black frame inserted (i.e. the backlight is turned off 1/2 the time). 240hz 1080p LCDs are only starting to trickle out now, with transition times remotely fast enough to see a (tiny) improvement over 144hz.
Simply playing back 24 frames per second multiple times won't make judder go away, it does reduce flicker in the old analog cinemas but that's because of the shutter. But inserting black frames in between duplicated frames (like 48hz = frame 1, black, frame 2, black, or 96hz = frame 1, frame 1, black, black, frame 2, frame 2, black, black) doesn't actually get rid of the fact that the delta between successive frames is too large with not enough inherent motion blur to mask judder, so if you have a superior projector technology like DLP (for motion), then it's more obvious. I can see judder in commercial cinemas easily : as soon as stuff moves quickly, chances are if it's not incredibly blurry it's incredibly choppy or some combination thereof.
If you interpolate 24p to 48p, you bring 180 degree shutter angle worth of motion blur to 360, unless the interpolation engine removes the inherent blur from the source material and then re-adds it in an appropriate amount. The pixelworks FI tech probably works on this principle, I'm going to investigate (this has bearing on my work).
The great thing about HDR is that, despite the fact that it exacerbates the perceptibility of motion artifacts like judder (your eye is attracted to sudden changes in luma channel therefore it's attracted to precisely the wrong parts of the frame : those areas with the most jumpy, high frequency changes), motion interpolation should work better than on SDR, because there's more pertinent luma gradient information to exploit, more separation between neighbouring luma values to inform edge and consequently motion detection. HDR also accentuates noise but that will only make temporal noise removal work better as well. This step is likely done at the studio but not always. Plenty of SDR Blurays are quite noisy with film grain because they didn't want to blur out high frequency detail, but with temporal denoising that's not really an issue : noise like film grain persists only in one frame at the same position so comparing succeeding frames (adjusting for motion) can tell you where in the current frame to ignore in favor of a temporally reprojected pixel in the prior frame. This is in the same ballpark of metadata that FI needs to work, namely motion between frames. So it makes sense to add temporal denoising option to FI for content with lots of transient noise like film grain or live video. Most pro video editing packages have temporal denoising but how good they are at preserving legit high frequency content is up for debate.
Regardless, video pixels have been processed countless amounts before getting to your eye, and sometimes the aesthetic or technical choices that producers or editors make are not those that you would take. I've met plenty of pros who just don't really grok the fundamentals of how the eye works. At 24p the frames are persisted way too long on the retina for ultra high frequency detail to be ignored, such as film grain or noise. Low ISO type noise from digital sensors is similar to film grain and is often visible with black levels fluctuating. But at high enough framerates, those minute differences disappear as your mind synthesizes them into an average. 24p is simply not high enough framerate, even 60p isn't. You need 83hz or above, the rate at which your eyes dart around to combine multiple angles into one synthetic image. Then transient per frame noise will disappear. However that still won't make judder disappear entirely, if objects jump more than 1 pixel between frames i.e. the shutter angle isn't 360 degrees (360 degrees is perfect blur, no judder at all possible).
In an ideal world, when you do interpolation, you should still do it to a whole integer multiple of the input framerate, so like 3:3 (72hz) or 5:5 (120hz) treating each new frame in the 24p source as a keyframe which is displayed with the minimal amount of processing.
There are some interesting papers out of Siggraph this year including variable framerate interpolation and temporal denoising / deblurring / interpolation / supersampling in a single pass (judder is just another way of saying : a shear in the temporal / frequency domain). Since shears can be detected easily, they can be undone. But having better (less banding, HDR), less noisy inputs will definitely let algorithms like FI work their best. Plenty of talk at Siggraph about how HDR source material impacts the performance of many computer vision techniques and improves results (since colour detail is preserved in both dark and bright areas of the frame, detection routines that depend on subtle differences in colour or intensity will all be immediately improved with HDR input).