I checked out that link and it looked like I was stepping back into the 90s. That image on the home page looks like it's a 256 colour GIF! Where's the specular mapping? Everything in those shots looks dead, like a bad phong highlighted raytrace.

There's much more impressive stuff going on with path tracing on conventional GPUs - something that, at least for me, is making a definite case for ungodly improvements in processing power for GPU hardware.

I've got a very-first-generation, USB-hobbled-by-polyfuses-until-I-performed-surgery-on-the-thing 256MB Raspberry Pi. The thing's part of my time-travelling Radio-4-Matic and thus transfers a few gigabytes a day over a little USB WiFi adaptor by streaming radio over the intertubes, buffering it for some hours then playing it back.

Uptime? Right now:
19:12:14 up 52 days, 15:46, 1 user, load average: 0.01, 0.09, 0.12

Last reboot was for a system upgrade of some description; the things are pretty stable now. (There have been many improvements to the firmware and system software.) My other Pi (a more recent 512MB model) is busy being a tiny home fileserver and virtual server backup device (remote stuff rsyncs over ssh to this thing) - I could easily use a spare PC for those tasks, but the result would be a lot less near-silent and much more power-hungry. Plus it can saturate 100Mbit ethernet with file serving - faster isn't much use when most of my stuff is on WiFi.

Make sure you've got a decent power supply. Apparently voltage drops can be a big source of instabilities. Power for my midget fileserver is via a Samsung cube phone charger; the radio's got a hacked-together DC-DC converter running off a mains-to-12V-DC adaptor. (I'm surprised the thing is as stable as it is, what with it solely relying on my impromptu electronics hackery!)

But in the end, it's all nothing but quarks and electrons, bound together and moving in various combinations and patterns, interacting via strong force (gluons) and electromagnetic force (photons).

Actually, below that it's mostly Perl.

I'm using Mail.app with Dovecot as the IMAP server - I upgraded to OS X 10.9 a few days ago, and haven't seen anything weird going on (yet). I sent myself a test email a few minutes ago while watching the Mail Activity window, and numbers appeared sensible. dovecot.index and dovecot.index.cache files on the server aren't ballooning - at 178KB and 11MB respectively.

The Fastmail article mentions Cyrus as the IMAP server. Is it Cyrus-specific, or have I simply not been bitten by this yet? (I get loads of spam, but it gets pre-processed by Spamassassin so Mail.app rarely gets to see any in the main inbox itself.)

The Koch brothers, aka Tea Party, don't really care about science as such.

Actually, one of them makes pretty hefty donations to science-related stuff, including big exhibits at the American Museum of Natural History. Human evolution, and all that.

(The socialist in me wonders if the latter is revealing some belief in social darwinism - survival of the fittest, and all that. Eek.)

You wouldn't believe the hoops I had to jump through to get that thing working - it had to be in my apartment at the time since the office phone systems were too modern, and connecting a (borrowed) vintage US Robotics Sportster 2400 (from 1987!) up to a vaguely modern Linux PC involved tomfoolery with various adaptors. USB to RS-232, DB-9 then DB-9 to DB-25.

System had mgetty listening to the modem, doing appropriate line control stuff - when people connected and entered the username 'backup', it would fling them straight into a hacked-together PHP script (stop laughing!) which asked for a password and then cycled through various plot fragments and home-made ANSI-art conversions of Portal 2 imagery, before kicking them off after a few minutes. (Why did the script itself ask for the password? I'd discovered a bit too late that the 'backup' user was Quite Important in Debian, and instead had to find an mgetty work-around. Which had the interesting effect that if you failed to type in 'backup' in the first login attempt on that connection, subsequent attempts would be tested against the real, no-login-available 'backup' user. Which actually delayed people's successful logins for quarter of an hour or so, since someone had failed a login with the correct username and password that way. Wahey!)

Testing was fun with only one phone line. I had a 'local' version running, with two modems attached together by a short phone cable - but this needed poking at mgetting with various signals to get it to pick up the line. With the BBS modem actually connected to the real phone line, I could call in with my mobile phone, and verify that it would automatically pick up and started squeaking. But knowing that the modem could actually send and receive data over the real phone line? Blinking heck completely untested!

Phone was ringing off the hook for over a week. I have no idea what the phone company thought I was doing - with this newly set-up phone line constantly receiving calls from all over the world...

I saw forum posts wondering what size datacentre we'd set up for this thing. Um....

Video of the thing in action here - PC doing all the work, Mac laptop logged into the logging stuff over SSH.

Recent Eurogamer article here!

An entire PCB filled with parts? This looks like an example of someone too smart for their own good.

The photo seems to be of this thing, which is an entirely different device which apparently 'allows a computer (or "host") to masquerade as a USB "device" to communicate with other USB devices or USB Hosts.'

In other words, exactly the kind of device you wouldn't want to unknowingly connect things to.

A couple of years ago I modified my old EOS 350D, replacing the IR-blocking hot mirror in front of the sensor with a filter that only allows IR through. I've taken loads of photos with it since then (please excuse the increasingly crap Flickr) - pretty much all hand-held with available light. Depending on the conditions (metering still works on visible light) I might be +1 or +2 stops up on typical outdoors scenes, while -1 or -2 stops down on near-IR-bright scenes like under forest canopies.

Images are generally pretty much direct from camera, all using the same white balance (set off a piece of white paper under tungsten light when I first did the conversion) - blue tones vaguely correspond with longer wavelengths. In-camera contrast is whacked up to the maximum, but little else. (The custom white balance is kind of weird - with a 'normal' setting, pictures come out looking fluorescent pink.)

I haven't noticed any magical see-through-clothes abilities from the camera, although I haven't really checked...

http://www.geekbuying.com/item/MK808-Dual-Core-Android-4-1-Jelly-Bean-TV-BOX-Rockchip-RK3066-Cortex-A9-Mini-PC-stick-307415.html

$42, but free shipping so I'll let that slide - no ethernet or GPIO, but does have built-in WiFi and 8GB flash storage and includes a mains adaptor. Will run Linux (with hardware-accelerated OpenGL ES) via the unfortunately-named Picuntu.

Interesting. Anyone got anything better?

If someone finds me a similarly-sized Raspberry Pi alternative for $35 (plus reasonable postage and packing) that has a semi-decent GPU, USB, ethernet, audio and GPIO, and runs Linux - I will buy it and report back.

(The BeagleBone Black looks interesting at $45. I think I might get one of those anyway...)

I put my Raspberry Pi in a box and it appeared on national radio. :-(

(Full documentation here. It's a 1970s transistor radio with WiFi, streaming Radio 4 over a SSH tunnel to the UK, time-delaying audio playback by eight hours or so, in order that everything gets played back at the correct local time in Seattle.)

I don't know what you mean by "old", but my father's old Canon (film) SLR's EF-mount lenses pop right onto my relatively new Canon EOS Rebel T2i (EOS 550D for you non-Americans) which takes EF-S-mount lenses.

EF-S is a subset of the redesigned-from-scratch EF lens mount from 1987 - still considered terribly modern 'cause it's fully electronic with no mechanical linkages between the camera and lens. New EF lenses are definitely still being designed, but yes - EF-S lenses won't fit on an EF-only camera, be it film or full-frame digital.

Canon's 'old' system is the FD lens mount, from 1971. The newer EF mount is almost completely incompatible - you'd need that overly-complicated-adaptor-with-included-optical-elements to get an FD lens to mount on an EF camera.

Compare Nikon's F-mount - lenses from 1959 are potentially mechanically compatible with the latest Nikon dSLRs, but there are huge compatibility charts describing which features may or may not work from any particular lens on any particular camera.

I have other CanonDSLRs that do the "video thing" out of the box, but it sounds. like an interesting experiment.

The particularly exciting thing about this hack is that it's not just a previous non-video-capable camera recording video, it's a camera recording 14-bits-per-channel linear uncompressed RAW video. Much better highlight and shadow recovery, white balance defined afterwards, much more information to work with in general. Some really tricky shots are now possible.

It does not have enough RAM to buffer frames continuously at uncompressed DNG format rates for continuous video recording to SD card, whereas other cameras that were designed specifically for video recording have enough memory to be capable of doing this.

The buffer is important, but it's more about being able to stream a metric shitload of data to a unwholesomely speedy memory card - once you can do the latter, the buffer helps smooth over hiccups but won't let you record indefinitely. The 50D's CompactFlash interface probably shares a design with a higher-end camera, Canon not wanting to waste effort in building a second, deliberately crippled version.

Thus my interpretation is that this camera model's hardware specs were deemed insufficient by the manufacturer for this specific capability, and considering that it can only do burst mode up to $X$ frames before capping out its memory buffer, the manufacturer may have been correct.

Being able to record RAW video is a pretty new feature on any vaguely consumer-oriented camera - it's more sheer luck that Canon's dSLRs have features which make it possible, albeit in a hacky manner. I get the impression that on the 50D, it's grabbing data from the sensor in a manner intended for the rear display or for feeding into the (non-existent) H.264 encoder, and then streaming it out to a big file on the memory card before the memory runs out.

When you've captured the data, it's in a big, opaque file that needs post-processing on a PC to do anything with it - in this case, it gets split into sane DNG files for further processing in software like Lightroom or similar. You can record the video on the camera, but you can't (unless I'm horribly mistaken) play the video on the camera - you need to do plenty of subsequent processing to get it into video form.

Don't get me wrong, it's an incredibly cool hack - partly because it gives access to a feature which few high-end cameras have even today. It's not the manufacturer deliberately locking users out of an easily-implemented feature, it's the manufacturer not even realising that such a feature was possible - albeit in a restricted, but still usable, form.