Slashdot Mirror
Intel To Offer CPU Upgrades Via Software
derGoldstein writes "Intel will again offer CPU upgrades through software. In the past, the upgrades gave you HyperThreading and more L3 cache. This time upgrades will actually increase CPU frequency: 'Intel Upgrade Service offers three different upgrades on second generation Core processors: Intel Core i3-2312M processor, Intel Core i3-2102 processor, and Intel Pentium G622 processor.' The page provides benchmarks of the 3 upgrade options."
Intel will again offer CPU upgrades through software. [snip] This time upgrades will actually increase CPU frequency
Hurray, now we can buy crippled CPUs and unlock them later.
It's like I'm being scammed at purchase, and scammed again at upgrade time.
In before Intel sells 256 core CPUs but requires you to purchase an extra license for every 2 cores beyond the initial 2.
Intel has been doing that forever, from the 486SX, which just had a broken FPU, to todays chips which are numbered/rated by which tests they pass/fail.
With them making Sandy Bridge non overclockable unless you pay extra, this was very likely to happen
According to the FAQ, if you replace your motherboard, the upgrade is no longer valid on the chip. It must store the information in the BIOS or at least use an identifier from the BIOS.
It also says you must be running certain versions of Windows 7 to install the upgrade but does not mention if an upgraded system would work in Linux or BSD or any other OS after installation.
I'm interested in a crack for this not to cheat intel out of money, but to activate it from BSD or Linux and to "fix" it myself if I have to swap out motherboards.
MidnightBSD: The BSD for Everyone
I know I shouldn't be RTFA but I couldn't read it. Slashdotted already?
I just wanted to know if these "upgrades" is done by changing the micro-codes. Or are there some FPGAs in the chips? Just curious, very obviously I'm not a chip designer!
Also, does this mean that someone (who REALLY knows what they're doing), could upgrade a "cheap" chip into something more expensive? Or add new features/try new designs or instructions? Isn't there some "hardware" encoded security aspects to these chips that might become vulnerable (like DRM)?
and Ford, they're going to sell you a car, and you can purchase an upgrade on your fuel economy, cooler air from the air conditioning, and enable the side-curtain airbags and heated seats too, for an additional fee, all as software upgrades.
The issue here is the manufacturers are starting to realize just how much overhead they're spending making so many different models of products, and that it's cheaper to just manufacture one model, the best one, and then cripple it if you don't want to pay for the best.
You could damage it (don't want the run-flat bladdered tires? they'll just shank the bladders with an ice pick near the end of the assembly line) or by disabling it via software. It's only natural to expect buyers to look for ways to re-enable disabled features. And we've seen so many times how manufacturers like to think they still somehow can tell you how you are and aren't allowed to use the product you purchased from them. (they want to sell it to you, but not really sell, as in, it's your property to do with as you please) God I hate that.
I'm really quite surprised that by now we're not seeing manufacturers trying to license physical goods. So you buy a computer. But you didn't really buy it, you licensed the use and Dell still owns it and is just loaning it to you, and can legally tell you how you are and aren't allowed to use it. (or cancel your license for any reason at any time, and demand you return it)
But closer to back on topic, so what's the going wager on whether they'll play the ever-popular DMCA card (for circumventing a protection device) if these get hacked back to top specs? I'm betting near 100%.
I work for the Department of Redundancy Department.
Intel now sells Atom CPUs, with embedded FPGA. Xilinx, the top FPGA maker, offers ARM CPUs with embedded FPGA. Both CPU lines run Linux now.
FPGA is logic gates, the building blocks of CPUs (and other computing chips) that can be interconnected on demand to create different logic circuits - and therefore custom instructions. Logic implemented in FPGA on a CPU can be revised by over-the-network software upgrades. FPGA was typically used by chip designers to develop candidate designs to be burned into hardware, but has become cheap and fast enough to distribute as end-product "reconfigurable computing" devices.
Imagine your multimedia codecs configured directly into logic circuits on the CPU. They'd be really fast, and lower power than moving data across the CPU/RAM/bus boundaries. Upgrades by SW, just like now. Load/unload them as circuits on demand rather than as instruction codes in banks of RAM. Bring the network wires to FPGA pins on the CPU, and the data can route to codec processors on the chip for parallel operation. Of course these features apply to any "media" data, including business data in streams or large datasets.
Intel's move to SW upgrades of CPU microcode is creating the tech and business infrastructure for regular FPGA upgrades to these new hybrids. Soon enough the literally hardwired CPU logic might become the minority of the chip. Already FPGAs with embedded DSPs are like that, so a chip that's mostly FPGA with just some ALU and CLU circuits already optimized to close to their theoretical performance (in speed or power) are foreseeable.
--
make install -not war
Wait till it start corrupting data on "cracked" processors as a form of DRM.
ELOI, ELOI, LAMA SABACHTHANI!?
Because it costs the same amount of money to make a fast chip or a slow one. But many people wont pay more than $xx for a cpu at a specific performance level.
This sort of thing has gone on in the electronics and computer business for 50 years. Back in the 60's and for several decades IBM offered a single printer that could print at three different speeds at three different monthly lease points. The only difference between them was a rubber belt. You'd ask for the upgrade, IBM would raise your lease fee, and a guy would show up to change the belt.
While some chips get binned lower due to inability to run at a certain speed or having a bad core, most are simply made to run slower at a lower price point.
What really is the alternative? Would you like the chip companies to have separate manufacturing process for each speed level, causing an overall increase in cost across the line? Just charge everyone the top cost and give them all the fastest chip?
I think its a cool thing that you can buy an inexpensive computer, pay a small fee, and have it go faster rather than buy a new computer. Why someone would work overtime to find an issue with this is preposterous...
Are they charging for it? I didn't see that anywhere in the link and I downloaded the installer. Unluckily, I have an i5 so I'm not even trying.
I think I'm going to pick up their new Bulldozer when it comes out. Intel makes great processors but these shenanigans have got to stop.
A NYC lawyer blogs. http://www.chuangblog.com/
Have you looked at the power-to-price curve of AMD and Intel? AMD beats Intel so thoroughly on the performance/price curve that I wonder why anyone bothers with Intel. The only part where Intel wins is the performance of high-end CPUs, but that's only because they pack more effective cores into one unit. Performance of single-threaded programs is roughly equal, so Intel can't claim an edge there as well.
You can care about performance of either single-threaded or multi-threaded programs. In the former case, AMD wins thanks to lower price, in the latter, it still wins as you can pile more CPUs and still get it cheaper. The only case when choosing Intel might be a rational choice is the sudden jump between prices of 1-CPU and 2-CPU systems if your needs are just above the top performance of best AMDs but below the point Intel would need two CPUs as well.
Intel's advertising tries to compare CPUs with different prices. To get a meaningful comparison, you need to compare performance with a fixed price or prices with a fixed performance.
The creatures outside looked from Alt-Right to Antifa; but already it was impossible to say which was which.
Just tried it. It was free.
This has been going on for quite sometime in enterprise world, well sort of. Although not quite the same, Citrix's NetScaler box can be "upgraded" via license purchase. This usually increases throughput and the number of allowed SSL sessions. IBM also sells their P-series server in quite similar manner. They will ship the box with all sockets filled with processors, but only enable the ones that you purchase. If you require additional processors, you will have to pay IBM to enable more processor. In the end, you still get what your money worth. I never consider an overclockability as a feature, I treat it more like a bonus. And if Intel or AMD decides to stop giving bonus, that's fine for me
It seems they want to build in a revenue stream so I wonder if they will be rolling out additional upgrades. So you buy this upgrade now, but in 3 months there will be an additional upgrade to increase performance another 10%.
It's like the DLC for games model. Buy the game. A few months later buy the DLC. A few months after that buy DLC #2, etc...
It usually suggests that competitive pressures on the seller, at least in that segment, are sufficiently low that they derive greater benefit from improved price discrimination than they do harm from making their prices less competitive. Given their fab prowess vs. AMD, it isn't totally surprising that Intel sees themselves doing better by voluntarily cutting the value of low end parts, rather than letting higher-end buyers get away with paying less.
(Secondarily, and specific to this particular instance, it probably doesn't hurt that consumer PCs frequently get crufted up and 'slow' over their lifetime and Joe User has no idea why. It's rarely the processor's fault, so what Intel is selling won't help them; but "make your computer faster!" is a well established product line, and Intel's offering won't technically be a lie...)
That is called binning. http://en.wikipedia.org/wiki/Product_binning
It is standard industry practice. Doing so saves *you* money because it gives customers the option to buy underperforming or semi-functional yields at a lower cost. It is good for the environment because it reduces manufacturing waste. Higher sellable yields improves profits for manufacturers and reduce costs for you. It is a win-win situation!
Some people are willing to shell out more money for a faster processor, while other people are not. It costs more to produce genuinely different CPU's than to just cripple one CPU, so the idea is that they make a large profit from the people who will pay for the faster CPU, and a lower profit from those who won't.
Jesus loves me, he loves me a bunch, because he always puts Jiffy in my lunch.
You're partly right. It costs the same to manufacture a fast chip, with all features enabled, as it costs to make that slower, crippled chip.
The admission inherent in the article is, "We've been ripping you off all these years, but suddenly, we find it necessary for public relations purposes to enable the features that you've paid for!"
This is why I've not paid for an Intel CPU since the original Pentium processors. I feel that their business model is dishonest.
"Windows is like the faint smell of piss in a subway: it's there, and there's nothing you can do about it." - Charlie Br
I think you don't understand what's going on. Intel is giving everyone more options. There's no way this can make you worse off. You probably don't realize that Intel doesn't make separate "1.8 GHz" and "2.0 GHz" chips. What they do is make many of the same chip, test each chip, and then set the clock frequency depending on how well each chip handles things. Now imagine many people would rather buy a 1.8GHz chip (it's cheaper and they don't need the extra speed), but the manufacturing process is good and makes mostly 2.0Ghz chips. Intel now has three choices:
Under the last scenario Intel is happier (they got the money of the people who want cheaper parts and got to charge a premium from the people who want faster parts). The consumers are also happier (they got the processor speed they want at the price they want). Why should the people who wanted 1.8GHz speed care that the part they got could in theory run at 2.0GHz? that's not the speed they wanted in the first place.
You misunderstand. They can, and do, sell faster processors as "crippled" slower processors. Their testing just identifies the maximum standard speed, and then the chips can be packaged and sold as any slower chip they need.
What's interesting here is that Intel is saying that all chips of these types are capable of running at a faster speed.
Right. AMD does exactly the same 'ripping off'. They speed limit for binning purposes and disable perfectly good cores and perfectly good cache to turn out lower end cpus. The only thing they dont do is give you a sanctioned upgrade path that maintains your warranty. Thats much better!
Years ago manufacturing yields werent as good and materials variability caused vendors to have to speed and function test all the products and bin according to what their capabilities indicated. These days manufacturing yields are excellent and materials variability and processing are also much better. That means you sell everyone an expensive, fast cpu or you artificially bin, the latter being what every single cpu manufacturer does. Consider also that when you're buying a cpu thats limited to 1/2 or 3/4 of its actual performance limit, its going to be a lot more reliable than a cpu thats running at 95% of its actual speed capability.
I hope you dont have a Playstation 3, because that'd really tick you off.. Those have 8 cores that are almost always all good, but they turn one off because half of the yield has one bad core, and they dedicate one core to the operating system even if its not using it! Those @%^#@$'s!!!
But it is true that every silver lining has a cloud, so keep looking for it.
... to enable the features that you've paid for!"
Which features that were listed in the product spec when you bought the chip did you not get? I can see that as a problem along the lines of fraud. But specs that were not disclosed? You never paid for them. Since you picked that particular model, it seems you didn't even want them.
So you're saying that because I like to play games (besides tux racer and rogue clones) at a reasonable speed on reasonable hardware, and so I deserve to be targeted by criminals.
Stay classy, sir or madam.
Job? I don't have time to get a job! Who will sit around and bitch about being broke and unemployed then?
I'm pretty sure that there is not as much variation in the speed of processors coming out of a fab plant as there is in the size of eggs that hens lay.
Having worked for a major chip company, I can tell you that your assumption here is incorrect. There really is that much variation in chips coming off the assembly line. In fact, I'd wager that the variation in chips coming off the fab is more varied than the size of chicken eggs.
It's the side effect of pushing higher and higher density within the chip. Modern equipment could probably push out extremely consistent 90nm designs. But 90nm chips were out of date 5 years ago. In order to be competitive, you have to push the most density out of the equipment you have. And that means that you get significant variation in your product, even within one wafer. The companies build in flexibility to the chip to allow for this variation. There are "fuses" built in to each chip specifically to disable the broken parts of the chip. If the L3 cache on a processor is totally hosed, they will blow the fuses for it and completely turn it off. But since the rest of the chip is fine they can still sell it, albeit at a discount. Even the maximum speed can be fused into the chip. The testing procedures ramp up voltage and clock frequency until the chip starts failing. Then they step it down a notch or two and fuse it there.
AMD has never produced a 45nm dual or triple core design. I'm not sure they even made one in 65nm. The x2 and x3 processors are just x4 (or x6?) chips with one or more dead cores and maybe less cache, depending on the specific chip. Intel does the exact same thing with their core series processors. That's just the way processor companies do business. It's been that way for decades.
You can get the top level performance for the hardware. You just need to pay a bit more.
Or, look at it from the other way. If you don't need top performance, they'll give you a discount.
FPGAs are too expensive and take too much power.
FPGAs are very transistor-inefficient and thus are very expensive and power hungry. To give you an example, programming an ARM Cortex A8 into an FPGA requires a multi-thousand dollar FPGA and takes double or triple digits of Watts of power. While a regular ASIC one costs less than $20 and takes a Watt or so. Also the FPGA one runs at perhaps 50MHz and the ASIC one runs at 1GHz.
Intel's reason for the FPGA is because they don't license their IP, the only way to integrate your logic with theirs without multiple chips is to use this. But that's a weak solution. With ARM you can license their IP and integrate it yourself in an ASIC, you'd be a fool to use an FPGA in a large-scale deployment, you're just throwing money away. In short-run deployments FPGAs make a ton of sense.
Use of FPGAs with DSPs is more common, programmable analog/digital logic can be very useful, like Cypress' PSOC (8051 based though, not ARM). I believe most cable/DSL modems use DSPs.
http://lkml.org/lkml/2005/8/20/95
This practice has been going on since the dawn of computing.
Back in the early 80's I was in charge of writing image processing (picture and movie) software using Data General minis to talk to the gamma cameras (this was in the Nuclear Medicine department of a famous American heart hospital). The Data Generals talked to a Perkin-Elmer mainframe (via HyperChannel) which was the machine on which most of my programs were run.
We ordered a processor upgrade from PE. A couple of days later, the tech from PE arrived to "install" it. I was curious, so I made sure to watch the whole operation over his shoulder. It was over and done with in about 30 seconds. He pulled a circuit board out of the machine, cut a tracer on the board, plugged it back in.
The bill was over $7,000.
One "Aw, Shit!" is worth 100 "Ata boys!"
No, it's like saying some consumers demand automobiles that will only go 150mph when they could go 300mph. Some people actually care about top speed. Some don't, at all. Some do, but care more about money. Welcome to the world of market segmentation.
The consumers aren't demanding a low-end CPU per se, they are demanding a cheap CPU. Others are demanding a faster CPU and are willing to pay. It's not just two groups either, it's a gradient.
It absolutely would cost Intel to enable the missing features. If their cheaper chips had the same capabilities as their high end chips, they have to cost the same amount of money, which eliminates market segmentation. Market segmentation generally benefits the company and the lowest-end consumers, while extracting more money from the highest-end consumers.
You can choose one price, equilibrium of all supply and demand, and then the people who want a high-end CPU and are willing to pay are winners because they get the same thing they wanted before, for cheaper. The people demanding the cheapest CPUs get no CPU, so they lose. There's a turning point somewhere in the middle between losers and winners.
I don't even understand why they would gain a lot in public relations. The number of people who care about this are few and this would be a one-shot. The manpower to disable the features seems pretty minimal too. Binning yields do mean that there will be a few lower-spec simply because the chip can't do better, which means that process already needs to exist.
Let me explain something to you.
No business sells a product at some fixed profit margin above the cost to manufacture. What companies do is they first determine their the lowest price they'd be willing to accept, their WTA, for a product. Due to economics of scale, the WTA goes down as the number of goods sold goes up. Imagine a graph where the x-axis is the number of units sold, and the y-axis is the lowest price they'd be willing to accept for that number of sales.
Meanwhile, they gauge what customers would be willing to pay (WTP) for the product. Obviously, the lower the price, the more people will be willing to pay. So you get another graph, where the x-axis is the number of buyers, and the y-axis is how much they're willing to pay.
It is mathematically provable that the maximum utility exists when you overlay the two graphs and set the price at the intersect point. However, this leaves some money on the table, as there are some customers whose WTP is less than the price point. You can't just lower your price, because you're already at the optimum price -- you'd lose out on all the additional money that most of your customers are willing to pay. So you have two options. You can have periodic sales, but this only gets those customers who pay close attention, and also loses money on customers with a higher WTP who happen to be lucky or thrifty. Or, you can create a product with reduced functionality and sell it at a reduced price.
You may wish with all your heart that this wasn't so. You may want companies to give away their best products at the lowest possible price, with no thought of securing funds to invest in future development. But this is how the world works. This is how prices are set. And, as I said, it is mathematically provable as the best method for everyone, consumers included. Incidentally, this system breaks down when a monopoly exists -- that's where consumers really start getting screwed, and that's what you should save your outrage for.
Intel has been doing that forever, from the 486SX, which just had a broken FPU
Some people here on Slashdot seem really upset about this software upgrade thing. But I was upset about the 487SX, and I still grimace when I think about it.
Before the 486, you had the 386 CPU chip, and the 387 FPU chip. A 386 motherboard would have a second socket for the FPU; probably the socket was empty when you bought a 386 system, but you could buy a 387 for a speed boost.
The 486 was the first Intel CPU with an integrated FPU. So, the 486SX was a way for Intel to sell a cheaper part, and to sell 486 chips whose FPU was defective. I get that. I'm cool with that.
The real 486 was called the "486DX". SX == no FPU, DX == FPU.
The 486SX and the 486DX were pin-compatible. If you wanted to upgrade a 486SX system, you could simply pull the 486SX out and pop in a 486DX.
But Intel tried to push a motherboard design where there were two sockets: the 486SX socket, and the 487SX socket. Instead of unplugging the 486SX and putting in a 486DX, you were supposed to leave the 486SX in place, and buy a 487SX, which was just a 486DX with an incompatible pinout (including one extra pin). You couldn't put a 487SX in a 486DX socket. When you put in a 487SX, the motherboard would disable the 486SX and it would just sit there, with the 487SX doing all the work, as it really was just a 486DX. (And an integrated FPU sharing cache with the rest of the CPU is better for performance.)
I found the whole 486SX/487SX thing to be breathtakingly obnoxious. It's one thing to provide multiple price points and find a way to sell CPUs with a defective FPU. It's quite another thing to engineer up a whole system that was cynically designed to lock up a perfectly good 486SX chip and trick a user into buying a special 487SX chip instead of just getting a 486 as an upgrade.
To make it even stupider, the 487SX cost more than a 486, because the 486 was being mass-produced. I found a Google Books scanned copy of InfoWorld that said the 487SX was 30% more expensive than an equivalent 486 chip! ($799 vs. $588 for a 25 MHz part) And a 25 MHz 486SX must have cost $258 because the cost of leaving the 486SX in place and adding a 487SX was $1057, vs. $588 for the 25 MHz 486DX plus having a spare 486SX you could sell or give away.
Nobody I knew ever bought a 487SX, and I don't think many companies even built computers with a 487SX socket. Even Intel can't push that kind of cynical "solution" and have wide success with it.
steveha
lf(1): it's like ls(1) but sorts filenames by extension, tersely
Except with CPUs it would be impractical to test every "location" on the wafer, so they have to make a map of which "chickens" will always make bigger eggs and which don't. That's where over clocking comes in because you might get one of the "smaller" chickens that only made Jumbo eggs 75% of the time so they had to exclude that spot. They also have daily shipments they need to make so they might need to fill some cartons with bigger eggs than on the label.
The NEWS here is that Intel has said in the "overclocking wars" that these chips were "burned in". Meaning they had firmware code that couldn't be changed after they left the factory. That's been the "party line" for the last DECADE of fixed multiplier chips.. now Intel is saying they might sell some upgrades?
because with IBM the machine is purchased under a contract with IBM. In the Desktop computing world, CPUs are a "thing" a resource in a box with a 1 year defect warranty. Intel has no "rights" to what the chip in my machine does only that I don't violate their copyright by reverse engineering it. IBM machines come with thick books guaranteeing CPU, uptimes, power consumption, etc. as well as warranties and service contracts for not meeting any of those specifications.
Intel is selling a widget that does "computer stuff", good luck with software. IBM is selling the actual service/utility the CPU provides, a system that will do so many credit card transactions per minute, 24x7.
Well.. I bought this 4-door car (its all they sell) but only the front 2 doors open. I can buy the upgrade package which they will send via OnStar which will unlock the back two doors, but its another $5k. If I change/modify the motor the back doors wont work any more.
Close, but it's more like you bought a two-door car that also had two unusable doors, then paid more to get all four working. You could have paid for the four-door version from the get-go but only needed two at the time and wanted to save some money. There's a subtle difference.
Some, including me, would argue there is value in increasing the longevity of a hardware platform by offering later upgrades. If you needed a 3GHz CPU why did you only buy a 2GHz CPU? I think what really pisses people off is that they value the increased performance but think they should get it for no cost. It's the same argument as mobile tethering. If it had no value, no one would give a shit. And because people do value it, only an idiot would provide it for no cost.
Eagles may soar, but weasels don't get sucked into jet engines.
Well maybe the employees that work for Intel can decide that if they're not getting wages that seem fair in proportion to what managers are getting, they can just move a little slower. No need to have different employees, they can just adjust their productivity to match the price. How efficient that would be. Management has already set a precedent, so they shouldn't have any ground to complain, right?
Burger stands could just use some slightly foul dressing to offer lower priced options without having to cook differently otherwise. I wonder if Intel is violating some prior art, like spit in the soup for customers that don't tip well?
If chips have a back-door to control one feature, what else is in there? Can they be really secure if they've got hidden controls or debug modes? People were upset when Intel was going to digitally serialize their chips. Whatever happened with that? Of course if chips can be uniquely upgraded it seems we know.
I hope Intel products get more serious competition. Also, the fuss about power consumption should be just for laptops. Feel the top of a recent iMac sometime. Hopefully Steve pressuring them will help. Did Intel ever come up with some answer to small geometry leakage currents besides lowering the voltage? Shutting down sections helped too, but a process that isn't prone to the problem is needed. The Core series was a huge leap from the Pentium 4, but it doesn't really seem like we've seen that much since considering how long it has been. It could be worse. At least CPUs aren't licensed by the year, waiting to expire after some freshness date. (the way it feels with Apple expiring old apps by omitting Rosetta in Lion)
Exactly and Intel as in TFA has a history of "pre crippling" the hardware you buy from them. it was a real PITA for me when XP Mode first came out as it was hell trying to tell which Intel chips supported virtualization and which didn't.
This is one of the reasons I recommend folks switch to AMD. with the AMD chips ALL the features are supported by ALL the CPUs, even the newer Semprons have virtualization and with most of the new boards you can take your chances and see if the dual you got is a triple or quad that they just turned off a couple of cores to make quota or if they were bad cores because unlike Intel they don't cut the traces. in some chips there have been reports of even turning on the L3 cache that was turned off to make a Phenom into an Athlon.
After the virtualization bullshit, the bribery, and finally the compiler scandal I decided to put my money where my mouth was and after being a lifelong Intel man switched my home and shop to AMD exclusively and frankly myself and my customers couldn't be happier. Unless you are in one of those rare niches (granted more geeks here may fit than most places) where you literally squeeze every chip for every MHz of power the AMD chips are frankly insanely overpowered for most tasks and dirt cheap to boot. No wondering if chip X supports feature Y, they all support everything, and you can build a quite nice quad for less than $400.
So don't support the pre crippled bullshit go AMD. I've found so far in my own personal tests the only thing AMD does is turn off cores to fit price points and they are often trivial to turn back on for $0, making them an even better deal if you don't mind taking the chance that the dual or triple you bought is actually a dual or triple because of a bad yield. I've found at least in my shop about 1/3rd of the ones turned off are done so because it had a bad core. But frankly the new Athlon and Phenom quads are so cheap and so overpowered for what most folks do that I don't even bother with less anymore.
But it is certainly better IMHO to give the customer a chance at a free upgrade than it is to nickel and dime them like Intel. You should have seen the look on my last triple core customer's face when I told him he got a free upgrade to a quad. Talk about a happy customer!
ACs don't waste your time replying, your posts are never seen by me.
Yup, back in the mid-80s I worked for a firm that wrote EPoS software for petrol filling stations (gas stations). There was a whole extra feature set that could be enabled simply by programming a special character (might just have been an "@" sign, I forget) into one of the programmable setup fields, and we charged quite a bit for it.
Our field-service engineers got so embarrassed at this (as did those of us in the software department with a conscience), that if time allowed they'd often open the box up and pretend to fiddle inside, maybe faking an EPROM change, to do it.
Eventually one or two site managers got wise, and the word spread as to what the secret was, and everyone was getting it for free, so we had to make it so it really WAS an EPROM change...