Digital storage in modern devices (digital cameras, phones/PDAs, etc) seems to have standardised on SD (Secure Digital) cards, particularly the higher capacity SDHC for digital cameras and physically smaller MicroSD for smaller devices like phones and PDAs. It has almost completely surplanted CF (CompactFlash) for this purpose, and seems to be surpassing Sony's MemoryStick technology too -- although both still exist in other areas. SD was first created by Panasonic, SanDisk, Toshiba in 2001, based on the MMC (MultiMediaCard. It is now standardised by the SD Card Association, who have approved lots of variations, including SDXC (an even higher capacity version), and SDIO which allows using the SD form factor for non-storage devices (eg, camera or wifi modules for PDAs).

As original created the cards had a write speed of 0.8 MB/s and a read speed of 3.4 MB/s (of which the write speed was the most critical parameter). More extensive use of the cards in devices with tighter performance considerations, particularly digital video recording, has led to the creation of faster versions. The SD Card Association defines three speed classes, Class 2 (2MB/s), Class 4 (4MB/s), and Class 6 (6MB/s). These speed classes represent a minimum sustained transfer speed guarantee of the associated speed; a CIR (Committed Information Rate). Best case transfer speeds will usually be much faster, especially reading back from the card (writing often needs to do wear levelling, but reading of course does not. Typically devices that are capable of recording in 1080p recommend a "Class 6" card, where a other recording typically wants a "Class 4" card; but there are a number of "Class 4" cards that are marketed as "Full HD" (1080p) capable, presumably for some definitions of "Full HD". "Class 4" cards are relatively easily obtained, but "Class 6" cards are still fairly specialised and hence more expensive (there are also lots of "Class 2" cards which are used for bulk storage where sustained speed is not an issue).

Unfortunately it appears that while the transfer speeds for the SD Classes are standarised, whether an actual product meets a given class appears to be "self assessed" (ie, the manufacturer decides whether they can claim a given class for their device). Some seem to treat this as a marketing advantage to claim it for their cheap storage ("look, cheap Class 6 storage") and some seem to treat it as a marketing advantage to encourage purchase of a premium version which exceeds the requirements (expensive version is claimed as "Class 10" -- something that isn't standardised, but presumably means guaranteed 10 MB/s). So to a certain extent there's still a difference between different "Class 6" devices, giving an advantage to buying a reputable brand. For the more honest assessments of worst possible case transfer speed it presumably boils down to a combination of wear level algorithm tuning (an area of active research), and how much storage is available that isn't presented to the user to mix into the wear levelling pool (and thus reduce the number of copies required to maintain the right wear levelling); extensively used flash slowing down has been a common problem with early flash devices as the wear levelling algorithms got developed, and I suspect will remain an issue with cheaper brands forever. (I also suspect that SD devices probably don't support something like the trim command which allows discarding blocks from the wear levelling so that they don't have to be copied, just overwritten; it may be the only way to tell the device this information is to format the storage.)

To add to the marketing confusion, the other factor usually claimed for flash storage is a "N x" speed rating. These figures are based on the notional CD-ROM transfer speed (1.2 Mbps -- note bits, not bytes), which works out to 150 KB/s. "Speed ratings" of, eg, "40 x" (48 Mbps == 6MB/s) or "66 x" (80 Mbps == 10 MB/s) are fairly commonly claimed, and even higher figures appear in various marketing (up to "200 x" seems to be semi standardised). However unlike the "SD Class" figures, these "speed ratings" are almost pure marketing, being maximum possible transfer speed (with the claimed figure more likely to be achieved reading than writing). A faster speed rating in the same SD Class is beneficial for uses that don't involve sustained transfer but do benefit from quicker read/write performance (eg, still photography, or file storage/transfer), but definitely need be taken with a grain of salt.

For video recording the storage transfer speeds depend on the resolution (eg, 1920x1080 for 1080p), the frame rate (23.976 (via 3:2 pulldown of 29.97), 24, 29.97) (achieved by dropping two frames per minute, 30, etc), and the compression settings used. Required bit rates between 17Mbps and 24Mbps are common. Assuming a manufacturer of a "SD Class 4" card is being truthful about the minimum guaranteed transfer speed, 4MB/s (32Mbps) should be sufficient to allow the card to keep up, providing the camera is able the utilised the card efficiently; this presumably explains why some "Class 4" cards are claimed to be "Full HD" capable. However even a 25% exaggeration of the worst case performance, or only 75% effectiveness in using the cards transfer speed, would potentially cause the card to fall behind the data being generated -- which seems to be the reason that "SD Class 6" cards are recommended for sustained 1080p video recording. Lower resolutions (eg, PAL at 720x576) ought to have a sufficiently lower data rate that "Class 4" cards would be fine.

The most believably "Class 6" cards seem to be the Sandisk Extreme III (a range which has been around since 2004, although the speed ratings received a boost in the last year presumably due to technology advances), and the Lexar Professional, both of which actually claim "Class 10" and priced accordingly. Cheaper brands seem to give people less than the expected performance (eg); and there are various fake cards with overclaimed specifications from less reputable dealers. Tom's Hardware roundup in 2009 agrees with this assessment; they also have comparision charts of SDHC cards which may be useful, particularly the minimum transfer speed charts It appears that the larger cards may benefit simply from having more chips in them (which are presumably interleaved in something like RAID 0 to improve transfer speeds).

From all of this it appears that buying a reputable SD Class 4 device, preferably one that claims "Full HD" video, in a larger size (8GB or 16GB) to get any benefit from interleaving, is probably the most cost efficient option for a digital camera -- unless recording 1080p at high frame rates and quality is the most common things it will be used for. The "Class 10" professional devices should show performance benefits, but they're probably only worth the money now if the device will be used regularly in ways where the extra speed will show. (For anything else, saving the money now and buying a faster memory device later when they're cheaper is probably the better cost/benefit.)

And for still photography, the best option might be an Eye-Fi SD card, which includes WiFi support and can automatically transfer to various online photography services, an FTP server or (I think) a nearby computer with their software running. Unfortunately they seem to only be available in North America at present. Some recent cameras are "Eye-Fi connected" which means that they are able to show when it's connected to an access point, and avoid auto-powering off while transfers are in progress.

And now for something slightly different: also on the specifications geekery front, lens microadjustments. Modern "professional" Digital SLRs have a fine tuning mechanism where by the digital camera body can store minor auto focus corrections for the different lens that are connected to it, and then these corrections are automatically applied when that lens is used. For instance the Canon 50D, 5D Mark II, 7D, and 1Ds Mark III all have this feature, as do equivilent level models in Nikon's range, etc; it is noticably missing from the consumer end of the range (eg, Canon 1000D, 500D, 550D, etc) and clearly being treated as a distinguishing feature of professional bodies. The results are most obvious when viewing 100% crops from a close distance.

The usual way of determining the appropriate adjustment is to select a suitably detailed subject, and take lots of photos with different levels of adjustment repeatedly until one is convinced that one has the best possible result for that lens -- the adjustment units seem to be in steps on the lens focusing stepper motor. (Unfortunately there is only one adjustment per lens and at least in theory it can vary with focal length so the adjustments for zoom lenses need to compromise, and either adjust for the portion of the zoom normally used, or try to achieve a balance between the range of the lens. Aperture may also affect this micro adjustment with similar results.)

A more clever approach is to focus on a LCD screen displaying a pattern designed to generate moire interference (moire pattern, as this helps quickly identify when the lens is focused so that it is able to resolve more detail. They compare their best manual focus (using LiveView with a zoomed in view) against what the auto focus achieves, and then tweak the microadjustments until the auto focus always achieves the same as the best manual focus. Someone else found another pattern more useful with tele lenses.

Taking this minute focusing to extremes, SLR Gear built themselves a complete dolly rig to allow them to maximise focus of the lens that they are testing, so they can always compare the best possible results from the lens without being limited by the camera body. The most useful insight is that they can more easily move the camera back and forth by small amounts (on their specially designed dolly rig) to bring it into precise focus than they can attempt to move the lens focus adjustment by "just the right amount" (the Canon Remote LiveView software seems to help with the "just the right amount", but is obviously still limited to whole steps of the focusing stepper motor).

SLR Gear looks to be a useful source of detailed technical reviews of lenses (their review of Canon 50mm f/1.8 mark II); and they have a few other interesting articles, including how to pick your second lens (depending on what you're doing). It's also useful to know how they test lenses (largely building on the work of DXO labs; see also the link above on the dolly rig), how they measure the effectiveness of image stabilisation. (DXO Labs are the ones that also produce software to "undo" lens distortion, which I linked to last year.)

Also of interest is the variation in the Canon f/1.4 50mm lenses due to manufacturing tolerances (as I mentioned last year) -- wide open there's quite a bit of variation in corner blur, but it seems to pretty much settle down by about f/2 or f/4. DigiLlyod also has an article about this Brand New Blur, which points out that many lenses are less than perfect, as delivered from the factory, even if the designs are wonderful -- complete with detailed writeup on several lens problems amongst his collection, and a guide to testing lenses for issues. He also has a guide to the factors affecting focus accuracy, of which the most likely is that the auto focus sensors aren't exactly where they're marked in the viewfinder, so the focus will be somewhere slightly different (presumably with experience one could learn the small offset for a given model), and the area covered by even a single auto focus sensor can be larger than the specific thing one is trying to get into focus. (But he notes that the auto focus usually gets more focus than one can manually, particularly if the viewfinder diopter adjustment is wrong.)

And of course modern sensors are getting very good at revealing limitations in lenses so these things are likely to matter more -- better living through advanced software, anyone? Or else "f/8 and be there" (since f/8 is often the only optimal aperture).