I'd imagine that it depends on how heavily current GPU/CPU compute systems lean on the 'CPU' side of the arrangement:

If the CPU actually keeps reasonably busy(either with aspects of the problem that aren't amenable to GPU work, or with assorted housekeeping tasks required to keep the GPUs fed and coordinated across the cluster), Intel or AMD offer pretty good prices for chips that provide a lot of PCIe lanes, support tons of RAM, and are supported by most of the world's horrid legacy software. Plus, motherboards and other supporting gear are brutally commodified, which is always nice.

If the CPU is mostly idle, and mostly gets included because it's the cheapest way to get a bunch of PCIe lanes and boot an OS that can run CUDA drivers and a NIC, then a Tesla-like card that includes a weedy little ARM core and can run on a simple backplane, without any PC server components, would seem like a logical thing to produce.

I suspect that they also don't have too much of a choice: the cost and energy savings of die-level integration with the CPU are difficult to ignore(and, even if they were less impressive, AMD and Intel both have pet GPUs that they integrate into most of their cores, and can freeze out anything more tightly integrated than a PCIe device at their whim, as Intel indeed did when they changed Northbridge interfaces). Either Nvidia commits to building SoCs that are all things to all people(a rather tall order), or they allow existing SoC-spinners to choose a GPU architecture with rather punchier PC roots than some of the traditional low-power/embedded guys.

Nvidia has been fabless since the beginning, the only difference with this announcement is that they'll sell you the ability to put their GPU on your die, rather than exclusively buying and reselling TSMC-fabbed GPUs of their design...

I'd be delighted if standards existed in this area; but they don't(to my knowledge, if they do, please let me know).

Keyboard, video, and mouse? Well, VNC is pretty antiquated; but at least it runs on almost anything. Your other options get thin, fast.

USB over network? Assorted proprietary implementations exist, no standard. (Even the capabilities of serial over LAN, as ostensibly standardized in IPMI, can be a bit...uncertain... from vendor to vendor and product to product).

"Secure boot" is mandatory on Windows RT(ARM) devices. I think that x86 Win8 devices are required to support it; but OEMs can do whatever key-fill they like, and can, at their option, support turning it off or end-user added keys.

I'm not saying that they didn't make a mistake somewhere, more than a few locked bootloaders have gone down; but it isn't going to be trivial.

I suspect that the inconvenience offers a built-in deterrent. To use one, you have to plug something into it, and the design offers no means of securing a device(as the pay-charge stations often do, in the form of little 'lockers' or similar that will hold a cellphone until you return).

How long are you going to stand around babysitting your phone in exchange for a few watts of free electricity? It's a convenient thing to have if you are taking a walk and need to top up your phone; but that's a pretty lousy hourly rate.

Aside from pure vandalism, which is possible; but wouldn't be deterred by fees, the only potentially sticky use case I can see would be the homeless. They have the fewest other options, and comparatively low opportunity costs for being near one of these as opposed to elsewhere. I suppose we'll see what team NYPD decides to do if they show up...

Unless the vendor doesn't support some update that you need because of a change on the server side(either a sufficient version bump that the protocol isn't totally interoperable, or something like moving from Citrix to VDI).

At work, we've had nothing but nightmares with HP's support for their thin clients. One of their WinCE models had a mystery timekeeping issue that kept the clock stubbornly out of sync. After a couple of weeks of hammering they escalated it to engineering, who confirmed the problem and then told us they had no intention of fixing it. The next model ironed that one out; but the last citrix client that HP supports is old enough that Citrix support people start making nervous noises when you mention it, and the official solution for VDI is 'buy a new one'. They also have a Linux based build; but that has (and has had for at least four years, despite my attempts to find somebody who would listen) a pathetic excuse for a 'kiosk' interface that allows you to dump unsanitized commands directly to the shell, along with at least one trivial root-escalation technique). On the plus side, these things are just overpriced VIA x86 boxes with limited RAM and IDE flash-disk-on-module units, and a nearly stock AMI BIOS, so we were able to just spin our own minimal Linux image, and most of the thin client software vendors have an x86 linux client freely available.

The hardware has been sturdy enough, only a few deaths over the entire deployment; but it was massively overpriced for its specs, and HP must have recruited its printer driver team to make the software suck so badly.

Why would ASIC be concerned about software-based traders? They know that, while it renders them somewhat inflexible, they are both far faster and substantially more power efficient by doing it in hardware...

"STEM: because your outsourced replacement isn't going to train himself"

I'm hardly saying that it's impossible(after all, 'Nylonase' enzymes were identified in 1975, for a compound that had only existed for ~40 years. Just that it's impressive. If anybody is going to be metabolizing plastics, it'll be bacteria, through sheer numbers and rapid mutation; but evolving, with no assistance, to attack novel compounds, designed for resilience, in less than a century after their introduction is pretty good work...

Given how late to the game plastics are, it is fairly impressive how fast they've moved. Some modified natural polymers go a fair way back; but most of the synthetics that we think of as 'plastics' are under a century old, are reasonably novel(not just a synthesis technique that is cheaper than the organic method for producing an existing material), and are often selected, at least in part, for good resistance to decay.

Also, polymers can be pretty tough molecules to crack: even something like cellulose, which is literally older than (some) dirt, is attacked primarily by a relatively small group of specialist organisms.

For truly efficient brigand-tracking, you want a globally-unique taggant, rather than being limited to the few dozen-ish colors of fluorescent dye...

Just imagine! Incentivize your riot cops during the next protest by mixing a unique DNA tag into each one's pepper spray and then analyzing the detainees. The more dirty hippies with your spray on 'em, the better your chance to win the department raffle!

My chemistry is a bit rusty, no pun intended; but my understanding is that chlorine is a stronger oxidizing agent than oxygen is. Are there any (feasible) conditions under which the chlorine could be persuaded to replace the oxygen in the iron oxide, leaving you with iron chloride and a considerable amount of oxygen?

I'm sure that there are peppy-people-person guidance counselors who haven' gotten the memo yet; but the criteria(or at least criteria for intervention) for most psych disorders includes '*whatever symptoms* are present and cause the patient significant trouble or distress'. Starting "Operation Afflict The Solitary" purely because Being Social Is Good! is pointless and unethical norm-imposition.

Providing a means by which the lonely-but-socially-anxious can acclimatize themselves, by contrast, would certainly be a good thing, even better if it can be done by means that are cheaper, easier to distribute, and lower-stigma than psychologists/psychiatrists.

I'd be a bit sceptical as to whether present-day retail-ready tech is good enough at reading and displaying things like facial nuance that(much to my vexation) are vital to in-person communication; but if they are, this seems like a good thing.

One might be tempted to suspect that the NSA is 100% to be trusted when it comes to securing those giant piles 'o data they are Hoovering up, even in the (vanishingly unlikely) event that they are, as they claim, actually not doing anything illicit with them themselves.