"Compositing", from the modern definition of the word, likely refers to the code in Photoshop that puts together your layers to show you a preview on your monitor of what your document is going to look like. If it's been updated, then we can presume the prior implementation must be outdated. Remember, Photoshop's internal structure was invented nearly 30 years ago, when computers were less than 1% as powerful as they are today and didn't even have Graphics Processing Units! If the compositing process has been updated / replaced, it stands to reason it would be done in such a way that makes use of your GPU. Every modern device has a GPU. On this presumption, we can imagine some very good things that could come out of this change, such as better display (or even printing/saving) performance... Are you seeing layered documents displaying more quickly/responsively? Is painting more responsive? Do files save more quickly? I haven't really been able to sense much of a difference myself, but I'm kind of spoiled - my system was already quite fast. We can also hope for better quality... Who hasn't dealt with posterization in a document when zoomed-out below 64%, or effects that only look right at 100% zoom? Are we seeing improvements in that realm? I composite image data together in my own plug-in software using GPU programming, and because of my experience I can name some of the big advantages: 1. All the pixel math in a GPU is in floating point, so accuracy is high and loss of precision due to integer calculations is avoided. 2. Because complex calculations like gamma conversions can be done so quickly, math operations can be done in a linear color space (that's how my plug-ins work internally). Inaccuracy because of working on gamma precompensated pixel values (i.e., what we have in virtually all images today that are less than 32 bits/channel) can be avoided. 3. GPU-based operations can proceed at a speed of 50x faster than they can be done in even the best CPUs. This is from a combination of very high parallelism (whole pixels processed at once in groups by lots of GPU cores) and super-fast GPU RAM. Besides additional display responsiveness this means that things no one would have considered doing can now be done in real time. For example, resampling an image using sophisticated, high quality algorithms (e.g., that don't introduce jaggies) can be done to zoom the display. Based on what I've seen so far, some of the things Photoshop COULD benefit from a GPU-based compositing process aren't there yet, and it's possible it's just such a big task to change out the guts of Photoshop that it has to be done in stages, and we're seeing just the tip of a new iceberg. I have high hopes for where we are likely to see this going, and it also says one subtle but important thing: Photoshop is continuing to be a platform of innovation. Like D Fosse​ so far I've been unable to see troubles with the new default (non-legacy) compositing in the new release. However, I have mostly worked on simple photos, with only a few layers, so far. Given the complexity of Photoshop layering, it's not hard to imagine that specific operations may be broken, or even more likely - presuming the GPU is now involved - that there could be differences from one hardware platform and driver implementation to another. I run a decent nVidia Quadro card, which provides 30 bit color and fast and stable operation (using Optimal Drivers for Enterprise). In my general experience as a plug-in developer in recent years nVidia seems to be leading the pack in quality of graphics drivers. -Noel
... View more