When mining centralizes sometime this year it will have an effect on BTC price since the consumer market for miners will be gone. Consumer circulation of Bitcoin to buy hardware is actually a large segment of the Bitcoin economy.
I doubt that. First of all, hardware manufacturers seem to instantly convert bitcoins to dollars, for obvious reasons (its not like TSMC accepts bitcoins). And do miners purchase coins to buy their gear with? Highly doubtful, since most hardware manufactures take dollars too, so I suspect almost all bitcoin purchases would be from existing stashes, so the net effect of mining hardware business, if any, is most likely that more bitcoins are being sold on exchanges, not the other way round. As for what will happen once mining becomes a large enterprise thing; there is no way to know whether or not these companies will liquidate everything they mine, or assemble a bitcoin stash. My guess is most will do a bit of both, it doesnt seem likely you invest millions in a large mine if you do no believe in the long term prospects of bitcoin. As for current small miners that will no longer be buying hardware; that could also mean they have more funds available to invest in bitcoins, something thats likely a better investment than mining gear anyway. Either way, the impact of mining on the BTC exchange rate is generally grossly overstated. There are nearly 13 million bitcoins. Each day, only ~3600 are mined. In two years, only half that.
|
|
|
I think it illustrates the exact opposite. Notice the diminishing returns at around 0.3w/gh? Other manufacturers may have hit that limit at 0.6w/gh (bitfury excluded).
Either you cant read a chart or you are assuming current asics run at a point to the extreme left of that chart, which would also imply those other asics can be overclocked by factor of 2-3x. Here is hint: almost none can. KnC, HF, CT, bitmain, Bitmine .. you'd be lucky to hit advertised speeds, let alone 2-3x more. Either way, welcome to my ignore list.
|
|
|
Caveat Emptor! Information asymmetry detected.
ITs not meant as promotion for BFL, Id be the last person to suggest you order anything from there. But the chart is interesting and relevant to earlier discussions about how you can scale power efficiency of every ASIC. Even if you take those numbers with a table spoon of salt, it nicely illustrates my point. There is about a factor 3x difference in power efficiency depending what voltage/frequency point you pick. That spread is not going to be vastly different for any other bitcoin asic.
|
|
|
Great discussion going on here, interesting to read.
Bitcoin mining is in a huge bubble right now. It's hard to get data to analyze that market, but how many $$$ have been injected in that market since 2012? Consider all the R&D, investments in mining business, and investments in ASIC manufacturers. I'm pretty sure the money injected in mining overshoot what the Bitcoin economy is able to support today.
Bitcoin mining can't go over what the Bitcoin economy is able to support. The better the Bitcoin economy is, the better it's network is going to be. Like any bubble, it usually ends with a crash. I think mining was temporarily saved by the Bitcoin bubble last November, but it can't grow like that eternally, not at that rate.
I'm pretty sure it's going to crash hard in the following months. The market has to correct itself, the sooner the better.
That doesnt make any sense. If people overinvested in mining, it isnt going to influence the exchange rate. Why would I sell my bitcoins lower because you (hypothetically) are out of a few tens of thousand dollars or dozens of bitcoins in unprofitable or undelivered mining gear? Getting screwed, whether its in BTC or fiat doesnt really influence btc rate. This is no different than when pirate's ponzi collapsed. If there is an impact, it will be tiny, and probably work the other way. Those that bought these unprofitable miners may hoard their mined coins hoping to eventually break even if the price goes up. But considering how few coins are mined compared to how many are traded, its not going to make a big difference.
|
|
|
FWIW, this is what BFL published recently: ![](https://ip.bitcointalk.org/?u=http%3A%2F%2Fcommunityhosting.net%2Fih%2Fimages%2Fmonarcmnm.gif&t=665&c=TrVQo4qxZ8HFlQ)
|
|
|
I do not understand why they are considering following the lead near bangladash. That is just the word of a marine search corporation. While the west australia lead was confirmed by several government search teams who picked up the blackbox ping.
I have no idea what georesonance found near bangladesh, the pictures really are intriguing, and it might be some plane, but whatever it is, it is not flight 370. There is no way a plane can rest on the bottom on that ocean, while giving engine monitoring satellite pings and underwater black box pings thousands of miles away. But if authorities dont follow up on it, conspiracy theorist will say its a coverup. May as well have a look whats on the bottom of the ocean there.
|
|
|
Well, let's estimate what the marginal cost of a completed TerraMiner IV is, not counting NRE costs. 1) According to this study, a wafer from a fab costs between $2000 and $4000: https://server1.tepper.cmu.edu/seminars/docs/BKM_semicon.pdf2) According to this picture( http://cointerra.com/wp-content/uploads/2013/12/photo5.jpg), Cointerra gets about (# of chips, length-wise)^2 * (Ratio of square area to circular area) = 28 * 28 * (0.78) = 611 chips per wafer Therefore, the marginal cost per chip = (Cost of wafer) / (# of chips per wafer) = $4000 / 611 = $6 per chip 3) That means the marginal cost of *just the chips* in a TerraMiner IV is: (#chips per TM IV) * (cost per chip) = (4*4) * ($6) = $96 4) Approximate the cost of all the components of the chip, plus the marginal cost of the circuit board creation: VRMs: Digikey has something quite like what they're probably using for $8: http://www.digikey.com/product-detail/en/IR3551MTRPBF/IR3551MTRPBFTR-ND/3770518Therefore, the cost is: (8*4) * ($8) = $256 Two power supplies @~$300 each: https://www.verical.com/#searchCriterion=mpnIDs&searchName=&landingPage=catalogItemView&searchTerm=295479&_i_=22*$300 = $600 Cooling components: 4 * (water blocks) + radiator + fans 4 * ($50) + $50 radiator + $20 fans = $270 Circuit board marginal cost: $100 * 2 ![Huh](https://bitcointalk.org/Smileys/default/huh.gif) (I don't have a good idea of this) = $200 Miscellaneous capacitors, resistors, and other components: $100? The beaglebone black: $100 The case $50? Total: 265 + 600 + 270 + 200 + 100 + 100 + 50 = $1585 5) Therefore a pretty rough guess on the cost of a unit below which Cointerra is actually losing money is $1600. Even if the cost estimate of the chips is an order of magnitude off, the price can still drop $1000 (i.e. go to $2600) and maintain a slim profit. Remember, these are *just* the material costs. The cost of doing business is *way* high than these costs - salaries, software, rent, etc. are huge for this kind of operation. The prices they've been charging have been recouping these costs in theory. i think you're light on some of the costs. the ones that stick out, I'm pretty sure wafers (in 28nm) cost more than the $4000 you suggested 'to buy' (maybe 'to make' they're cheaper as clearly the fabs aren't doing it for charity).. so you can assume the asics cost more than you're assuming. Actually, volume production of 300mm 28nm wafers is lower than estimated above. Down to sub $3000 range for high volume. Of course, bitcoin manufacturers are not yet touching the volumes of AMD/nVidia/Qualcomm etc, but there is also the fact that AMD/nVidia typically require more than a dozen metal layers, whereas Im guessing a bitcoin miner would need no more than 3. That also impacts processing time and cost considerably. If anything, $4000 per wafer sounds high to me if you are calculating the "end game". Many of the other estimates are also clearly too high. A naked PCB for a bitcoin rig isnt going to cost $100 to produce (you can buy far more complex fully populated ATX motherboards with many more layers, connectors and components for only a fraction of that in retail) and a beaglebone black retails for $49 on adafruit, so no way CT is paying $100. VRMS do cost a bit, but caps and resisters and all that, we all know basically cost nothing in volume, no where near $100 per board ALl in all, IMO the above BOM estimates are way, way high.
|
|
|
Sure it means something. It means if you post enough nonsense spam, you get paid more to add spam to your signature. Of course the thread about hot asian girls really could be the NSA ![Smiley](https://bitcointalk.org/Smileys/default/smiley.gif) .
|
|
|
I get back to the question of who's going to be left supplying infrastructure? Certainly at a point where things are saturated it would be very difficult to start a new company to do this (how would you fund it?).
Im sure some of these companies will continue existing as they branch out to different markets or products and most of them will probably end up mining themselves, either for their own account, like KnC, or reselling the hashrate, like BFL. But even those that will close their doors, will probably sell their IP before doing so (assuming its a reasonably competitive design), so anyone holding it can place an order with TSMC to produce another batch of wafers should demand increase again.
|
|
|
First of all, congrats davejh, you seem to be getting it now ![Smiley](https://bitcointalk.org/Smileys/default/smiley.gif) . What I'm unclear on is who is going to pay to run mining hardware that takes insane amounts of power and can never ROI?
Whoever happens to still have a miner or mining operation. Keep in mind that as long as its operationally profitable, it doesnt make sense to turn off those machines, even though they may never ROI. And you will most likely see tons of those, in fact you already see them now, miners who will never make a BTC denominated profit, but keep mining to minimize the loss. It will always be someone else's problem, whereas if the difficulty were to be allowed to drop significantly in order to offset the costs then the system is too easily exposed to 51% attacks (it's almost impossible to safely de-escalate an arms race).
As above, difficulty is not likely to ever drop (assuming constant BTC value), as that requires miners to actually be turned off. Thats never going to happen on a large scale. Once we approach the point of marginal profitability, ie, mining revenue is barely above electricity cost, sales/deployment will pretty much grind to a halt, but there is still an incentive to keep the existing infrastructure running. Before you mention reward halving, that will already be factored it when making the purchase/deployment decisions.
|
|
|
At this point I'm somewhat struggling to see what will actually allow anyone but electricity suppliers to see any sort of profitable returns in a few years time.
So what? Bitcoin isnt dependent on miners or its manufacturers being profitable. All bitcoin needs is that those miners actually run, and that is guaranteed. That said, i would expect the most power/cost efficient mining operations to be able to extract a small operational profit on average, but anyone who thinks this is a long term goldmine (excuse the pun) doesnt understand mining or is betting on BTC appreciation - which we all know can be profited from much more easily by just buying BTC.
|
|
|
Here's some proof, oh yeah and BTW our chip on 55nm achieved that "less than 0.5W/Ghash" back in june 2013 ![Smiley](https://bitcointalk.org/Smileys/default/smiley.gif) Seems like you even went below 0.4W/GH at the chip level. Not bad for your first ever asic design (IIRC) on an obsolete process and nearly one year ahead of the competition. Cant wait to see what you will do on a more advanced process. Havent kept up, is something upcoming ? That said, wouldnt waste my time trying to talk sense to "jimmothy". Logic is lost on him.
|
|
|
The implication is that the ASIC vendors planning to be in business in 2 years time have to work out what their costs are in the future and backtrack to set a price where the difficulty level is held under far more control than the technology alone would define. There's also a positive incentive for none of them to get too aggressive in pricing because the total market size is fixed; doing the least possible work wins both for them and existing miners. The latter is kinda obvious, since this mining is a zero sum game. As for the former, they figured that out looong ago. As did I. Before the first asic was even announced I predicted the dynamics of virtually free to produce GH (especially compared to FPGA's and GPU's of that time) could only result in plummeting prices and skyrocketing difficulty for years to come, and it would have to happen at a speed that would result in a BTC denominated bloodbath for nearly all miners, and it gave perverse incentives for overpromised preorders. I was ridiculed by miners drooling over the prospects of dozens of GH for only a few thousand dollar or a few hundred BTC. I kept clear, shut down my GPU farm never ordered as much as a Jalapeno and watched the onslaught. Several years later, perhaps confused by the BTC price spike, most people still dont seem to get it. I'll build a new version of the model that tracks difficulty to compute the hardware prices (essentially drop the hardware price on every difficulty change to keep new entrants able to ROI) - should be interesting! :-)
Yep, except if you want to make it realistic, new entrants will not be able to achieve a positive ROI. Almost no asic miner ever has (BTC denominated), and very few paying list prices ever will.
|
|
|
I don't believe so - the problem is that everyone thinks that the operating costs should dominate, but at the moment equipment/capital costs currently swamp everything else. Thats only true for end users who are paying OEM's enormous gross margins. Those margins serve partly to pay for the substantial NRE, partly because they can get away with it because difficulty is still so "low". BUt dont let that confuse you. If it helps, look at KnC's datacenter. How much are they paying for their equipment? Its not going to be anywhere near $2/GH, I guarantee you that. If we clock slower in order to improve the W/GH/s ratio then we need more hardware. If we want to halve the W/GH/s ratio we drop the performance by about 30%. For the same amount of power we can achieve 1.4x the hashing capacity but now require 2 ASICs to do so. The hardware cost is what's totally dominating the current hash rate though. No, its not the cost. Its the producers gross margins. A year ago we paid $50/GH for preorders, today we are below $2 for immediate shipment and that is despite bitcoin appreciating ~5x. What changed ? Difficulty, and little else, considering I happen to be comparing 65nm (BFL) and 55nm (Bitmain) products here. If you calculate GH cost in function of BTC price / difficulty, you will see that barely budged at all. We are not anywhere near the bottom yet. Prices will keep dropping as difficulty skyrockets as long as gross margins are still phenomenal and manufacturers have therefore no incentive to scale down production. You cant predict whats going to happen without acknowledging this reality. So I definitely agree that hardware prices are high right now, but much of that has to be a factor of low production volumes. Nope. Its clear that cointerra or HF wont get the same pricing as AMD, but even if you double AMD's wafer price, you will find their margins are stunning. Lets do some math. Hashfast golden nonce has 4 81mm˛ dies in an MCM. That works out to (4x) 185 candidates from a 300mm wafer. How much does HF pay for a wafer? Industry standard volume pricing for 28nm today is in the ballpark of $2500 per wafer. Keep in mind that AMD/nVidia/etc chips regularly need 15 or more metal layers, whereas Im guessing a bitcoin miner would need no more than 3. That alone should pretty much close the price gap with AMD, but lets say despite that, HF pays $6000 per wafer. Thats $32 per GN candidate. Add a few dollar for packaging and yields (which should be very high given the redundant nature), and there is no way they should pay over $40 per chip. The PCB will cost too, lets say thats also $40. And lets add $20 in operating margin and to get a nice round high end ballpark of $100. Whats the market price for their EVO board? $1888. If you buy 5. And are gullible enough to beleve HF will ship that in June. Note that I didnt forget cooling, it comes without cooling, much less a PSU. Low production volume isnt a factor in current pricing. It will become a factor once prices drop another order of magnitude, at which point, bitcoin asic volume will have become non trivial even for fabs anyway. A 28 nm device of the same size will do more work but will also cost more than a larger geometry (this will change as 28 nm matures). MCMs are never particularly cheap and there's also a huge amount of variability based on the particular process being used too. 28nm has been mature for quite some time. MCMs do cost a few dollars, but think why they are using them. If it were cheaper to use 4 smaller dies, wouldnt they? MCM packaging apparently is cheaper than what it costs to mount the chips and coolers. That says it cant be much. Now, could someone decide to operate at cost rather than making a profit?
Thats not someone's decision that will drive that. ITs the market that will drive that. Who's going to pay $2/GH if difficulty is 10x what it is today? No one. And likewise, what bitcoin asic manufacturer is going to prefer not selling anything at all over selling a $100 PCB for $200? I think if someone really wanted to do this seriously they'd go warehouse scale and fab ASICs to meet their design.
![](https://ip.bitcointalk.org/?u=http%3A%2F%2Fpbs.twimg.com%2Fmedia%2FBmBf3-6CMAAT-lA.jpg%3Amedium&t=665&c=IXWDHNu--6vdWA) That would let them increase their volumes and reduce costs but there's a major cooling problem to be solved and that would have to be amortized in too. Cooling just increases your electricity costs.
|
|
|
You do realize bitmain s1 could achieve 1.2w/gh at the wall right? That's a tad bit more than half as efficient than an s2(1.1w/gh)..
You moron, S1 and S2 use the same chip. So of course S1 is capable of achieving similar efficiencies if you underclock and undervolt them just like in the S2. Thats the whole point. The difference is that S1 by default uses a clock and voltage setting that results in ~2W/GH at the wall while for S2 they picked a point that roughly cuts that by half. Got any better evidence of your claims? ![](https://ip.bitcointalk.org/?u=http%3A%2F%2Fupload.wikimedia.org%2Fwikipedia%2Fcommons%2Fthumb%2Ff%2Ffb%2FOhm%2527s_law_triangle.svg%2F500px-Ohm%2527s_law_triangle.svg.png&t=665&c=GCk19K5H8enrpA)
|
|
|
With 20c/kWh we hit a rate of 640 PH/s, 10c/kWh we hit 766 PH/s and 2c/kWh we hit 893 PH/s. Which only shows your assumptions cant be right. Ultimately mining is converting electricity in to bitcoins, you would expect a near linear correlation, not a 40% increase for a 1000% decrease in elecricity rates. The only thing preventing that from being truly linear is hardware amortization. YOu must have quite unrealistic assumptions for those. For W/GH/s then I agree that voltage and frequency scaling can probably take things down to 0.325W/GH/s at a loss of probably something like 30% of the performance (assume a square law - it's over-optimistic but it's somewhere to start). That would give us an improvement of about 1.4x but means we need more ASICs to hit the same hashing rates.
? The fact you would need more chips to achieve better efficiency doesnt again reduce the efficiency. IT does increase cost, but not compared to todays market prices, only compared to todays marginal production costs. The simulation allows that initial 1.4x to be factored in very easily. Let's assume that an ASIC vendor reduces the cost of their devices so that the cost/GH/s remains the same (they're effectively reducing prices 30%) so now 2 devices use the power that 1 previously did but we're now getting 1.4x the hashing capacity. Now the numbers work out as: 20c/kWh we hit 766 PH/s, 10c/kWh we hit 842 PH/s and 2c/kWh we hit 910 PH/s Im not sure you understand how mining hardware pricing works. Only two factors really matter; miners (as in end customers) breakeven point or perceived breakeven and manufacturer breakeven (on marginal costs). Currently these two are several orders of magnitude apart. As a result, miners are being sold for prices that miners perceive as potentially profitable. BUt that price point is linearly correlated with difficulty, so those prices will keep going down, way, way faster than Moore's law, at a speed limited only by manufacturing capability, until prices approach marginal production costs of the most efficient vendors. Now you can debate what those are, but asicminer claims $0.2/GH wafercost and my guess is that larger MCM 28nm devices are at least 2-4x cheaper per GH. I totally agree that the equipment cost is the biggest single factor. Actually, I dont agree ![Smiley](https://bitcointalk.org/Smileys/default/smiley.gif) Electricity cost will become the biggest single factor IMO For a ballpark guess, lets take $0.2/GH as marginal hardware production cost and 0.5W/GH efficiency. So each TH would cost $200 to produce and draw 500W. 500W @ 3 cents per KWH (including cooling, overhead) = ~$131/year. $200>$131, however we are looking at the "endgame" here, and therefore hardware prices will stabilize and their value loss can be written off over several years. If you just amortize over 2 years or assume a resale value of 50% of the purchase price, electricity has become the biggest factor. I'm not sure that anyone currently producing mining ASICs can get to $32 per device though
If by device you mean the asic itself, then obviously that can be done easily. If you mean a complete rackmountable miner, then obviously not, but in 6 or 9 months, no one is going to be selling miners in fancy enclosures with touch screen LCDs. Look at KnC's datacenter, or bigmegapower; naked PCB's on "ikea" shelves. Thats how it will look. they're still dramatically slowed down from the past 12 months.
Now thats something I and I presume everyone else will agree with. No one ever expected a monthly hashrate doubling to continue perpetually.
|
|
|
I'd been thinking about this a few weeks ago and finally got time to build a simulation of the hashing last night (it's a couple of hundred lines of C). Here's the extrapolation: ... The hashing rate here is in PH/s. The 3 predictions are for the price of BTC in 2 years time. The full write up is at: http://hashingit.com/13-megawatts-of-miningI like the chart, but I cant agree with your assumptions. IN particular electricity costs. Average electricity costs are irrelevant, what matters is lowest electricity cost (where mining is feasible). Because that is where mining will relocate, and places where electricity price is substantially higher, miners will have to shut down. Here is a read for you: http://www.spokesman.com/stories/2014/apr/26/northwests-cheap-power-drawing-bitcoin-miners/A map comparing energy rates led them to Central Washington, where hydroelectric dams churn out electricity that costs industrial customers less than 2 cents per kilowatt.You are assuming 10x that price. Secondly you are incorrectly factoring in asic "improvments". THere are two factors that matter, price per GH and watt per GH. The first one doesnt need Moore's law to improve dramatically from where we are today. ALl thats needed is difficulty going up further, and prices will come down proportionally. We are no where near asic manufacturing costs yet, in fact today marginal production cost of these chips is so low, that its irrelevant, they might as well cost nothing. Asic alone, I estimated HF golden nonce elsewhere at $0.04/GH in wafer production and packaging cost (assuming they can hit 800GH with their rev). As for watt/GH; also there we do not need Moore's law to get a doubling in 2 years. THe very same asics being sold today almost certainly have the possibility to be downclocked and downvolted to achieve that doubling (or more), be it at the expense of performance per chip. ALmost all miners sold today are being run and sold at the top end of the voltage/clock shmoo plot, which is evidenced by the fact that they typically barely overclock at all. To see what changing of voltage and clock can do, look at Bitmain S2, which is the exact same chip as in the S1, but is twice as power efficient. Im willing to bet KnC, HF, CT, etc can easily double their power efficiency too on their existing designs, once that trade off becomes worthwhile. But since electricity cost is only a tiny fraction of most miners cost, and pricing of hardware is far more dependant on hashrate than power efficiency, it doesnt make sense for 28nm designs today. Doesnt mean they cant.
|
|
|
Here is the specs https://bitcointalk.org/index.php?topic=495357.0. Again you need evidence that lowering voltage to increase efficiency is possible and because it works with cpu/gpus doesn't necessarily mean it has to work with bitcoin asics. I have yet to see bitmain/hashfast claiming anything below 0.6w/gh which they would happily do if it were possible. I assume they have already tested the chips to find out the maximum efficiency so they can advertise such. Why would they not? Maybe if you tried reading a bit more carefully, I wouldnt have to repeat myself over and over. How many times did I explain that to increase power efficiency you have to lower the voltage? Didnt I specifically say "Got a link showing that power effiency does not increase quadratic with voltage?" Now, where in your first link does it show they changed the vcore? Nowhere. for whatever reason, that poster only changed the clock. Did he not have access to the vcore settings, did he not bother trying, is it a firmware or PCB issue preventing him from changing it, I dont know nor do I care. In no way does it refute my "theory". As for Bitmain, I never said they could achieve <0.6W on a 55nm design. Given that they already claim 0.68W/GH at the chip level, they probably can, but only I gave them as an example of doubling power efficiency without as much as a chip revision, simply by lowering clocks and voltages from near the top of the schmoo plot to somewhere lower. The same will work for your GPU, for your CPU (both of which will in fact do this automatically when mostly idle) and for pretty much any asic with programmable clock ever created because its a direct result of effects inherent to CMOS technology combined with Ohms Law.. If you dont believe me, see if I care. I understand very clearly the difference. What I don't understand is why KNC would RUSH to the newest node size (spending more than necessary simply to be first) when they could simply lower voltage and save millions? Wouldn't it make sense to wait until 20nm is cheaper since production cost is nowhere near a limiting factor as of now? Only reason I can think of for doing this would be that they are limited to 0.6w/gh (at cost effective $/gh).
My god you are dense. You are the only one who ever claimed that the only reason KnC is moving to 20nm is to achieve <0.6W/GH. Everyone else understands that 20nm, if ever they get it working and yields become reasonable, should provide lower production cost per GH as well as better power efficiency.
|
|
|
What makes you so sure that current gen chips are underclockable to better than advertised efficiency? Like how bitmine advertised low power mode at 0.35w/gh (according to your understanding of physics it should have worked) yet underclocking did diddly squat. Underclocking alone indeed does "diddly squat" to power efficiency. The key is lowering the core voltage. If core voltage can not be raised/lowered to achieve higher speed/better efficiency, that generally hints at a design flaw. I have no idea whats going on with bitmine specifically. Got a link showing that power effiency does not increase quadratic with voltage? You need some evidence before you can claim so confidently that all current gen chips are underclockable to below 0.4w/gh (at a reasonable $/gh)
I dont, because I never made that claim. How exactly is spending 10 million rushing to the smallest node size cheaper? If you dont understand the difference between NRE and per GH production cost, there is not much I can do. Not that $10M sounds realistic to me, for a chip as simple as a bitcoin miner. Its not going to have 15 metal layers like a highend CPU or GPU. Id be surprised if it has more than 3, maybe 4.
|
|
|
I'm pretty sure it's not that simple or they would all be advertising/doing it.
Your ignorance of laws of physics doesnt change them. Every one with some experience over/underclocking CPU's and GPU's would be well aware of the range and impact of core voltages. As for why they arent doing it yet; it doesnt make financial sense yet. Hardware prices are still far too high, electricity cost is still utterly marginal for most large customers. Pricing is done per GH, cutting that in half to get better power efficiency doesnt pay off now. Fast forward 6 months and you will see. If knc could simply lower their voltage why would they spend 10 million on 20nm nre just to get 0.4w/gh at the chip level? Because 20nm should also be cheaper to produce per GH, due to the increased transistor density. Not that I (ever) expect(ed) a 20nm part before late fall, but thats another story.
|
|
|
|