Slashdot Mirror
Planar NAND Development Ends After 26 Years
Lucas123 writes: The non-volatile memory used in thumb drives, SSDs, smartphones and any other mobile device today has at last hit an engineering wall. The major developers of planar NAND this week said now that they've reached 15 or 16 nanometer process technology, they no longer expect to shrink their lithography process any further, as the capacity and economic benefits no longer make sense. Toshiba, which produced the first NAND flash chip in 1989, SanDisk, Intel and Micron said they will turn their engineering efforts to 3D flash trap NAND, 3D resistive RAM and other vertically-stacked non-volatile memories that offer a much longer road map. The manufacturers all said they'll continue to produce planar NAND while developing 3D NAND, which has already doubled previous capacities while also offering two to 10 times the erase-writes of previous non-volatile memories and twice the write performance. Intel and Micron are also producing a 3D NAND, based on floating gate, and a ReRAM that the companies say will increase performance and endurance 1,000 time over planar NAND. Toshiba and SanDisk have come out with a 48-layer 3D NAND that could allow them to produce 400GB microSD cards next year.
I am a game developer.
Indeed many games have color banding, so do many jpeg images. But this has nothing to do with the color depth.
When a game bundles an image, it is normally compressed in a lossy format such as DXT5 or ETC1 (depends on your platform) . These formats are typically much smaller than say a PNG, and are sent compressed to the video card. The video card has hardware that can get a pixel when needed from these images without having to decompress it. This saves a lot of video card memory which can be used for more polygons and whatnot.
These formats like jpeg, do modify the image a little bit if it helps makes them smaller. A somewhat oversimplified explanation is this: suppose there are 5 pixels that are almost the same color, for example: (red, red+1, red-1, red + 2, red +1), the algorithm will change them to be the same color: (red, red, red, red, red), then instead of saving each individual pixel, it will just store: (5 red), which takes a lot less space. A particularly bad effect of this is that gradients end up being not so smooth so you see banding. Reality is a lot more complex than this, but you get the idea.
In addition, when a texture is rendered at a distance, the hardware actually chooses a scaled down version of the image. The farther the texture, the less precision is used until there is only 1 pixel. This is called mipmap. Depending on the algorithm used for blending mipmaps, it can also generate banding.
You could use 128 bit RGBA color depth, and you would still see the same banding due to these optimizations.