Slashdot Mirror
The Economics of Chips With Many Cores
meanonymous writes "HPCWire reports that a unique marketing model for 'manycore' processors is being proposed by University of Illinois at Urbana-Champaign researchers. The current economic model has customers purchasing systems containing processors that meet the average or worst-case computation needs of their applications. The researchers contend that the increasing number of cores complicates the matching of performance needs and applications and makes the cost of buying idle computing power increasingly prohibitive. They speculate that the customer will typically require fewer cores than are physically on the chip, but may want to use more of them in certain instances. They suggest that chips be developed in a manner that allows users to pay only for the computing power they need rather than the peak computing power that is physically present. By incorporating small pieces of logic into the processor, the vendor can enable and disable individual cores, and they offer five models that allow dynamic adjustment of the chip's available processing power."
IIRC this is done in mainframes for *ages* already...
In related news, an initiative of car manufacturers spearheaded by Ford has introduced an enabling 'cylinder per need' model. Car performance is wirelessly monitored in real time to give the customer the option to add in additional power according to his needs if he has signed to a plan designed to optimally fit his profile (composed on his overall lifestyle information). This also creates a new exciting opportunity to reduce individual carbon tyreprints for the consumer.
CC.
TaijiQuan (Huang, 5 loosenings)
I don't want to "rent" the processing power of my own computer, thank you. Nor do I want to "rent" my operating system, or my music, or movies. I buy those things, and I'm free to do with them as I wish.
Renting your own possessions back to you is the sweetest dream of all hardware, software and "entertainment" manufacturers. Never let them do it.
we discovered a new way to think.
You know what I still don't get? Why's everyone acting like dividing a CPU into several separate cores is a good thing?
Let me compare it to, say, a construction company having a number of teams and a number of resources, e.g., vehicles:
1. One team, 4 vehicles. That's classic single core. Downside, at a given moment it might only need 2 or 3 of those vehicles. (E.g., once you're done digging the foundation, you have a lot less need of the bulldozer.)
2. Two teams, can pick what they need from a common pool of 4 vehicles. That's classic "hyperthreading". Downside, you're not getting twice the work done. Upside, you still paid only for 4 vehicles, and you're likely to get more out of them.
3. Two teams, each with 4 vehicles of its own. They can't borrow one from each other. This is "dual core." Downside, now any waste from point 1 is doubled.
But the one I don't see is, say,
4. Two teams with a common pool of 8 vehicles. It's got to be more efficient than number 3.
Basically #4 is the logical extension of hyperthreading, and it seems to me more efficient any way you want to slice it. Even if you add HT to dual-core design, you end up with twice #2 instead of #4 with 4 teams and a common pool. There is no reason why splitting the pool of resources (be it construction vehicles or execution pipelines) should be more efficient than having them all in a larger dynamically-allocated pool.
So why _are_ we doing that stupidity? Just because AMD at one point couldn't get hyperthreading right and had its marketers convince everyone that worse is better, and up is down?
A polar bear is a cartesian bear after a coordinate transform.
In mainframes you have pretty much a single vendor (IBM). Even in the days of Amdahl and Hitachi, once you were committed to a single vendor they had a lot of market power over you. So the vendor can set its own price, and squeeze as much money out of each customer as possible by making variable prices that relate to your ability and willingness to pay, rather than to the cost of manufacturing the equipment.
In a competitive market where 100-core processors cost $100 to produce, a company selling 50-core crippled ones for $101 and 100-core processors for $200 would quickly be pushed out of business by a company making the 100-core processors for $100 and selling them, uncrippled, for $101. I expect the Intel-AMD duopoly leaves Intel some scope to cripple its processors to maintain price differentials (arguably they already do that by selling chips clocked at a lower rate than they are capable of). But they couldn't indulge in this game too much because customers would buy AMD instead (unless AMD agreed to also cripple its multicore chips in the same way, which would probably be illegal collusion).
Compare software where you have arbitrary limits on the number of seats, incoming connections, or even the maximum file size that can be handled. It costs the vendor nothing more to compile the program with MAX_SEATS = 100 instead of 10, but they charge more for the 'enterprise' version because they can. But only for programs that don't have effective competition willing to give the customer what he wants. Certainly any attempt to apply this kind of crippling to Linux has failed in the market because you can easily change to a different vendor (see Caldera).
-- Ed Avis ed@membled.com
1. Everybody gets the same chip, but it will be crippled unless you pay the highest price.
2. Everybody gets the same uncrippled chip, but there's a FLOPS meter on it that phones home, and you pay Intel according to the amount of numbercrunching your chip did for you.
Both of these models seem completely retarded to me, although the first is already sort of in use in the CPU/GPU market. Have modern processors overshot our needs by so much that our big worry now is to find innovative ways to cripple them? If so, maybe this processor war we're fighting is ultimately not even worth winning.
Someone already mentioned mainframes. Something similar is often done with calculators. Rather than design a new chip for each model, they design a single chip with all of the features. In mid-range and low-end models, it is crippled by the design of the keyboard and/or jumpers. It is often cheaper to dumb down a single hardware design than to produce unique designs for each segment of the market.
Mea navis aericumbens anguillis abundat
The cost of designing one core is the same as the cost of designing 10 or 100 cores, because copy and paste was invented several years ago. The cost of adding a core to the design is about 1%.
There might be a case for powering down unused processors to save energy, and there is a case for selling cheaper processors with reduced core counts where some cores don't work, but there is no case for disabling working processeors for economic reasons.
Sun's Niagra technology differs, cos it has "virtual cores" which gives you more virtual cores but slower. Its very good if you multi-thread (run apache) and p*ss- poor if you dont (run Windows).
Sent from my ASR33 using ASCII
Because it's dumb.
In 1999 I paid about AU$600 for a midrange Pentium Pro CPU. In 2008, I bought a midrange Xeon Dual-core for the massively increased price of... AU$600.
In 2000, I bought a shiny new Intergraph TDZ2000 with two PII 350s for the bargain cost of just $5,000. Now, Apple is prepared to sell me a Mac Pro with two 2.8GHz, quad core Xeons for the stupefying price of $2,799.00.
Now, explain to me again why it would be in my best economic interest to buy a computer with cores that could be disabled if I don't pay my rent?
"I've got more toys than Teruhisa Kitahara."
really. STOP IT!
This is crippleware, and a terrible idea for the average consumer...
Paying more for a product that costs the same to produce, or potentially even less because they don't have to disable the extra cores is a terrible rip off, and it happens already...
The same people who currently overclock, will buy the cheaper cpus with cores disabled and re-enable them... You will also get third parties who make a business out of doing the same, tho without the "exceeding design spec" risks of overclocking.
Personally, I will never pay more for a more expensive version of the same product, i will buy the cheapest available just as soon as people have worked out how to re-enable the disabled cores, and i will help my less technical friends do the same.
http://spamdecoy.net - free throwaway anonymous email - avoid spam!
I know that on Linux, I cannot immediately tell the difference between an SMP-enabled kernel on a single-core Hyperthreading system, and an SMP-enabled kernel on a dual-core system with no hyperthreading.
/proc/cpuinfo, I need an SMP kernel, etc. So if someone (Intel) suddenly decided to make a dual-core hyperthreaded design in which the "teams" actually shared a common pool, would I notice, short of Intel making an announcement?
In either case, I'm fairly sure I see at least two items in
As for your assertion, a quick scan of Wikipedia suggests that you're a bit naively wrong here. (But then, I'm the one pretending to know what I'm talking about from a quick scan of wikipedia; I suppose I'm being naive.) Wikipedia makes a distinction between Instruction level parallelism and Thread level parallelism, with advantages and disadvantages for each.
One of the advantages of thread-level parallelism is that it's software deciding what can be parallized and how. This is all the threading, locking, message-passing, and general insanity that you have to deal with when writing code to take advantage of more than one CPU. As I understand it, a pipelining processor essentially has to do this work for you, by watching instructions as they come in, and somehow making sure that if instruction A depends on instruction B, they are not executed together. One way of doing this is to delay the entire chain until instruction A finishes. Another is to reorder the instructions.
But even if you consider this a solved problem, it requires a bit of hardware to solve. I'm guessing at some point, it's easier to just throw more cores at the problem than to try to make each core a more efficient pipeline, just as it's easier to throw more cores at the problem than it is to try to make each core run faster.
There's also that user-level interface I talked about above. With multicore and no hyperthreading, the OS knows which core is which, and can distribute tasks appropriately -- idle tasks can take up half of one core, the gzip process (or whatever) can take up ALL of another core. With multicore and hyperthreading, the OS might not know -- it might simply see four cores. And with multicore, hyperthreading, and shared pipelines, it gets worse -- as I understand it, there's no longer any way, at that point, that an OS can specify which CPU a particular thread should be sent to. Threading itself may become irrelevant.
Well, anyway... What confuses me is that we still haven't adopted languages and practices that naturally scale to multiple cores. I'm not talking about complex threading models that make it easy to deadlock -- I'm talking about message-passing systems like Erlang, or wholly-functional systems like Haskell.
Hint: Erlang programs can easily be ported from single-core to multi-core to a multi-machine cluster. Haskell programs require extra work at the source code level to be made single-threaded, and can (like Make) use an arbitrary number of threads, specifiable at the commandline. They're not perfect, by far; Haskell's garbage collector is single-threaded, I think. But that's an implementation detail; most programs in C and friends, even Perl/Python/Ruby, will not be written with multiple cores in mind, and, in fact, have single-threaded implementations (or stupid things like the GIL).
Don't thank God, thank a doctor!
And moreover, they apparently forgot which problem they're trying to solve between paragraphs 4 and 5. They start talking about the real problem of many cores creating a very large space of core/memory architectures that would be difficult to choose between and support. Then they veer off into the rent-your-own-hardware-back-to-you idea and never finish reasoning out just how it would work before they come back. A few minor things they ignored:
- How do they turn cores on? Difficult level: No, you can NOT have a privileged link through my firewall onto my network.
- How do they stop me from hacking it and enabling it all myself? Difficulty level: Mathematically impossible since you can't stop Eve from listening if Eve and Bob are the same person.
- How do they propose to bill me? Difficulty: No, I will NOT let my CPU spy on me.
- Why should I hand you everything you need to force me to upgrade against my will?
- What happens if you go out of business and leave me stranded?
- Even if you don't see what's wrong with charging me continually to access my own hardware, do you actually think I won't?
In conclusion, Profs. Sloan & Kumar of the University of Illinois, I believe the premises and reasoning behind your proposal to be flawed, and the proposal itself to be unworkable and contradictory to openness in computing. Or, as we say on the Internet, wtf r u doin???That was my first thought. Then my second thought was having to go through the "Intel Genuine Advantage" activation process every 45 minutes.
CheShA: Manchester Breakcore / Drill and Bass Yes I'm a s
I've done this kind of thing. nVidia 6800LE with half it's shader processors disabled (had 4 blocks of 4, 2 blocks disabled), which could have half of those (1 block of 4) re-enabled without issue. Athlon XP 2500+ that could have the FSB changed to 200MHz instead of 166 and it would BECOME a Athlon XP 3200+ (name and all).
And the best one: Two Athlon XP 2400+ cpus that I unlocked with a conductive pen to be Athlon MP 2400+s, and I still use in a dual-cpu board now.
Generally, unlocked or overclocked pc parts burn out faster than if they'd been left alone (e.g. the 6800LE I mentioned died a horrible death, and now doesn't work at all). However if the chip was DESIGNED to be able to be unlocked, it would be perfectly safe.
The cost of designing one core is the same as the cost of designing 10 or 100 cores, because copy and paste was invented several years ago.
Copy-pasting a hundred cores will cost almost ten times as much as copy-pasting 10 cores because you have to pay the patent holder who invented copy-paste.
-
- - You can't take something off the Internet! That's like trying to take pee out of a swimming pool.
Generally, unlocked or overclocked pc parts burn out faster than if they'd been left alone (e.g. the 6800LE I mentioned died a horrible death, and now doesn't work at all). However if the chip was DESIGNED to be able to be unlocked, it would be perfectly safe.
Design is one. Manufacturing is two. Chip manufacturing is not perfect. It is more likely that the disabled parts failed full test, but that parts were still working (and thus make it sellable as a downgraded chip). All you did was enable the defective parts. And then it blew. No surprise there.
Having worked at nvidia, there is a reason those extra TPCs were disabled and its not because of a cripple ware model but because of yield. We cannot produce chips that are perfect all the time. So we settle for chips that are perfect a small percentage of the time, mostly perfect an ok percentage of the time, and half working a good percentage of the time. We then make 3 or 4 different series (GS/GT/GTX/GTS/Ultra) with different TPCs in each series, disable the TPCs in each chip that doesn't work or fails to pass QA and then ship them. If you unlock them, you are frying you working card because some of the faults could be things like "Oops, there was a short in the TPC because the transistors cooked too close to each other" or "Oops, the clock passes too close to the +12V in this module -- if it hits 50 Celcius, it could turn into a short". This model helps products from being prohibitively expensive for a fabless company because we are billed on "silicon wafers used" on not on "number of fault free chips produced".
... for CPUs, there are effectively ZERO variable costs to the producer once you've purchased the chip and it's in your hands.
... make something once, get money forever.
Dedicated circuitry to create artificial scarcity and control actually adds unnecessary costs.
This might be useful in very specific scenarios where somebody, say, owns a supercomputer and rents it out, but even there, I'm sure there are far better solutions that don't involve the CPU hardware.
This is, like you suggest, just a BS wet dream of the manufacturers
Right now we probably have few enough major chip vendors that with a little bit of collusion, if they decided not to compete, they could probably pull something like this on us. This doesn't look likely right now, but it seems possible. Hopefully some other (possibly foreign) company would enter the market if that happened. Competition is healthy for a market.