LN2 is not viable, it's probably a good use for all the hardware made for the AMD CPU's, the old vapochill boxes, but the ASIC chips would probably struggle with the whole peltier setup needed, with no payout, and really hard maintenance/setup. like, 5 hours of your life you'll never get back.

a peltier cooler would be an easier sell, less "ahmagaawd, this is cooool".

maybe there's some benefit to a better cooler ...

let's see, you could use off the shelf copper heatpipes, fin radiators,

or, if you have a thermal cooling efficiency problem, e.g. too much heat, not enough space, e.g. rack mounting ASIC blades, or something,
water/fluid loops and 120mm radiators, much like the corsair / swiftech contained radiator loops.

it's relatively cheaper than a custom fin solution, and able to be chained in a bigger radiator solution, i.e. rack cooling with quick disconnects,

a combined pump off a 12v connector, so you can design a vent/radiator to rack-mount a bank of ASICs to cut down on ducting and fan noise, set each blade/rack to use rows, pump fluid in at the top and bottom, and parallel cool & pump 8-10 racks

.. for no particular reason.

one can dream.

how about a 36 port powered usb hub ... it's not upright, but it's certainly overkill, and that's half the fun.
http://www.amazon.com/Port-Powered-USB-2-0-Hub/dp/B008RXWM1M

here's 16 ports & upright , untested though.
http://usb.brando.com/industrial-usb-16-port-hub_p01936c039d15.html

if you want something official, http://elinux.org/RPi_VerifiedPeripherals#Powered_USB_Hubs is a tested list of hubs.

most 7 ports will have 2-3a, you should aim for the 4 amp versions.

This.
and you don't need to create an entirely new system of ASIC hardware that would stick out like a sore thumb.  

much easier and cheaper for covert agencies, hack people, hack the trust system.

i.e. let's talk destabilisation conspiracies.

step 1, create a screen and cover identity for one group, the hook, and the bait.
all you need to do is set up a large pool and invite people in, as to create a poisoned network node for confirming a larger pool, this needs to be "super secure" and well respected. think p2pool. this would be the security bait.

step 2, destabilise the other large pools and select a nominal target, so you can create a monopoly from the rubble, a big fish.

step 3, find an insecure server that hosts a large 40% pool, set up a tap/backdoor on the machine(s), insert your double-spend or false transaction record binary instructions into custom compiled pool daemon's and regular people will do the hard work of hiding data in the network.

step 4, establish trust in the big system and the little system at the same time, so that a general public will believe in following/propping up the big fish, while detail oriented people will move towards the security bait which has the hook.

step 5, you can keep the big fish away from the hook for as long as you need the system to work to do what you want it to.
after which, you can remove either the fish or the hook, and the system won't notice the loss. remove both, and tampering becomes more obvious.

The reality of doing this would be difficult to hide due to crypto being rooted within the protocol itself, it's the confirmations, or lack of, that matter in this kind of setup. The best analogy would be FOX news. you don't need to create or mock up news from scratch to insert a message, all you need to do is modify the content a little bit and then repeat the process, the other networks will propagate the false news as if it were accurate, forcing others to also confirm the "new" news.

i grabbed the "BF3" 11.10 catalyst preview last night, haven't rebooted post-install to get the CCC GUI installed (Win 7 x64) , but so far, restarted cgminer 2.0.5 on intel, and no single thread usage anymore, which is nice. will try 11.10 preview on another machine, but i think it will have the same effect, which should be great for thermal testing as it's a WC rig in a single loop.

so far, the CPU is pegging 4-9% for normal windows ops with chrome running, where CGminer used to have 13% by itself, it's down to 0/1% and 47 seconds CPU time for ~2 hours.
as for cgminer performance, it's ~388.6mh/s and 70'c on defaults, so it looks to be pretty consistently normal on HD6970.

haven't installed BF3 from the download, but it seems to load other games just fine. i usually leave cgminer running while gaming, and doesn't seem to affect FPS, but i'll test it later for other apps for any other performance issues.

you can also use the new CGminer, set the fans/thermal setting to 50% / 80'c using --auto-fan/--auto-clock and it will move between 850-1000 / 300 depending on the temps you want to maintain or how good your cooling is. it will also handle GPU load/crashes, etc.

i've run 2x5770 with aoclbf at 950/300 and 850/300 in a regular case, no power tweaking was needed. but, i think using cgminer makes it way easier, as you can see more of what's happening as it's happening, and it handles soft-restarts of the GPU pretty well if you're trying to find temp/voltage limits and tolerances while under load.

whenever you have PSU questions, etc or are considering a new build, http://www.extreme.outervision.com/psucalculator.jsp
I also use http://whirlpool.net.au/wiki/psu_recommendations, but, yes, 1250w should cover most sins

750w will probably not be enough to cover GPU + CPU at peak, as the general peak load looks like 769w, and 1100VA on AMD 940, 741w/1060VA on i7 2600k

but the big issue is the 12v capacity, will need to be over 58-60A to handle the PCIE load on those cards. that's what to check for.

you should aim for 850w silver/gold or 1200w bronze/silver if you want more reliability or need specific cabling.

in some places, 1200w is cheaper than 850w silver rated PSUs, but, ultimately, brand names charge more. i.e. seasonic X series = Corsair AX series, same seasonic hardware, $50 difference.

as for the bronze/silver/gold, the 80+ gold rating applies to PSU conversion efficiency, soo, if you run the PSU at half-load capacity, i.e. 300w peak load on the big sticker 750w PSU , it will use less power and cost $50-100 more than the equivalent 750w. you're pulling ~740-780w, 60-63% of 1250w, so aim for the 1200w series on a cheaper PSU, or get a modular 850w-900w PSU, both should cover the same power load and the 12v rail

that said, silver rated at 50% load, will be roughly as good as a gold rated at 85% load. that gold standard is pretty damn good. i've run a 6970 x 2 on 750w gold for 3 days, no issues, and that was somewhere around 88-90% load.

sweet. this saves all the effort of getting a VPN just to buy steam games from the US/UK. always fun to see what http://www.steamprices.com/au/topripoffs looks like before i hit the buy button. saved $50 on DXHR that way. /bookmarked

as for MW3, it's just expensive in time/money/effort. for me it's $99 USD in AU Steam, but $59 USD, 6.6BTC for a game i likely won't be playing until 2012 (would be dealing with SWTOR, BF3, xmas, relatives, in random order), and also there's no preorder bonuses, no discounts, and more than likely i'd get bumped into waiting 2 days for the unlock (regional unlocks are staggered, it's worse elsewhere, i think JP still has to wait)

even if i buy it from the US store, MW has always been awful to buy through steam + for the same price, there's RAGE, skyrim, and other games.

i.e. http://www.steamprices.com/au/search?modern+warfare and http://www.steamprices.com/au/upcoming?p=2

just a bit of a blank ended question, and i've only read 1/10 of the thread, because, 6 hours time to read one thread ... (sigh.)

do any of the schedulers allow load-balanced/parallel pool usage ?

i.e. 2 pool types interleaved so that while you might be on mine_slush and their rotations/slice priorities, it will also interleave another thread with a mine_charity or mine_deepbit thread ?
it would probably be easier to work with the pool categories in parallel, so that each connected miner, would be handling various packets, and shared pools at the same time. it does complicate the proxy nature of bH, but this would work if each share were equal in payload or there were metrics to use on each share/packet. it would possibly require a more advanced metric measurement of the size/difficulty of the work, as well as a few metrics for the minimum rate for each pool type, and a way to rate-limit work.

The more i think about it, the less it seems practical, except in the cases where you want to maintain goodwill with pool operators by contributing 1 out of every 4 shares, instead of 1 out of every 100 or 500 shares, or limit it down to 1 : 10, etc.

it would also reduce the 'surge' phenomenon, where combined rates halve or decimate once the minimum share dividend has been reached, it also allows for acceleration and deceleration over time for PPS/Proportional 'surges' where a new block will reset the current shares. on the flipside, if aggressive pools implemented duration-based timeouts or a last-man-standing proportional system, to prevent surges, which is not as difficult to implement as a routing metric, this approach would be designed to stop that kind of ban taking place, as long as enough shares were being sent to keep that 'window' open.

without a real metric routing protocol, modifying an existing scheduler could probably do this on a 1:10 share basis, maintaining a minimum work rate of say, 80mh/s on one, or two "non-hoppable" pools and still continue with the PPS/Prop/last-man-standing pool hopping to maintain a shared dividend.

or is this idea just silly for reasons i can't fathom, namely that routing metrics are ridiculously difficult and complicate things ?

if i remember, the ATI temperature/fan monitors i downloaded, have cards 0 to 5 in win7 x64 and cat 11.7.

this could be the reason why it can't talk to cards 7 and 8.

i'll see if i can find a link to confirm it's not a driver/interface problem ... perhaps it's the reason why GUIminer only sees cards 1-6.

i always get nervous with WC,  just due to the extravagance.

i do have a corsair H70, baby steps, but the big triple radiator kits look fierce enough to a) weigh more than the case b) seem to be just as noisy with 3 fans to cool 2x6970's as the air-cooling rigs. one day, i'd probably like to get something like the swiftech 320 edge kit, with the blocks to cool things. but it's nearly the cost of a 6970 for the WC kit and the extras. sigh.

in theory, if the regular ~50db gets annoying i might spend the money, but it would have to be amazing. and you'd also likely need a UPS that can handle 1kva minimum as well, just to make sure the heat load is not staying around

i was thinking video cards, you'd get free attention from anyone and everyone mining, for any of numerous reasons.

but really, who can beat porn.

i was really impressed with this program, until i tried to save the mem/core clock speeds and it caused the PC to lock up, as it applied the speed, but not the fan, and then not the voltage, etc. and then only to one card. it might have been an older version but it was annoying at the time.

i did end up editing the ini files, but the clock rates didn't apply properly, and i had to reboot into safe mode to tweak overclock rates. i think the main problem was that the fans didnt kick in or allow easy granular control via the sliders, you could set 40 and hit fan speed, and then you'd move the slider to 55 and it wouldn't set unless you did it quickly. again, older version.

that said, it's a great program for unattended use, and had it running for 5 hours, with each card having different rates and performance settings. it works great.

at the moment, i find it's easier to use barelyclocked and cgminer in a batch script and put a shortcut to the start.bat file in the startup folder, it also makes sure to return clock/fan/voltage rates back to normal when you close the window or quit cgminer, which can be a help when you need to shut down,  be in the same room as 4 GPU fans at 70-80%, or play games for a few hours and return to mining, etc.

the recent NVIDIA geForce drivers have OpenCL 1.0, the latest GeForce 280.xx Driver have OpenCL 1.1 which should be more compatible. it's a recent addition to the beta drivers, so it should be in the WHQL's

1. uninstall the current nvidia driver,
2. download and install the newer 280.xx WHQL version and
3. reboot,
4. see if GUIMiner starts up by reading the logs. if it does work, you should notice the GPU fan starting to get louder.

mostly i expect you won't get very far, as GUIminer isn't built for NVIDIA's OpenCL and it will be far below 100Mhash/s, perhaps 30-50MHash.

Have you considered buying a new video card or are just testing the waters ?

i tried GUIminer with NVIDIA previous to changing my video card, i was running 275.33 and ATI AMD SDK 2.4  (due to not running beyond 30MH/s) and found that i had stability issues with OpenGL applications that i needed to use, as well as PhysX and other applications. there may be better options for Nvidia mining, but i'd recommend not using the AMD SDK if you're going to use a GeForce GTX card, you only end up getting 30-50Mhash/s, and an unstable gaming system