I apologize for the shameless plug. Apparently some in the Mining subforum don't read the Marketplace subforum, so I would like to say:

There is a miner faster than Phoenix (+9%) on the HD 6990, it's called hdminer.
(It is not free, so do the math to determine whether its price is worth it or not for you.)

I lowered the price from 400 BTC to 350 BTC.
hdminer is currently the fastest miner for the HD 6990 at 708 Mhash/s at stock 830 MHz clock, compared to Phoenix's 650 Mhash/s, a difference of 9%.

Future reply to the random guy who will reply "but I get 7xx Mhash/s with $MINER when overclocking" -> I said at stock clocks !

Thanks. Unfortunately my personal pledge applies to shops with a page rank 3 or above.

That said, I sent you 5 BTC because I appreciate your effort. txid f715deb5fb689a0a9d069aa55876783ac451ef4e63e47c450fe936a4abd1c758

Mahkul, I would like to modify my pledge to 125 BTC (up from 100 BTC). The maximum remains 500 BTC (ie. I pledge for up to 4 shops instead of 5). And I would like it to only apply to shops shipping tangible physical goods (that's important).

Yes it is difficult. Don't get me wrong, I wrote a miner from scratch, code to parse the block chain, etc, I think I would personally understand Bitcoin if I listened to, say, gavin's econtalk for the first time. You have to learn to put yourself in other people's minds. Most non-tech people gloss over mentions of decentralization, assymetric crypto, etc. I have a friend who mine who still thought there were central servers somewhere distributing the work and did not know Bitcoin was distributed. That's because the Bitcoin presentations he watched did a poor job of re-iterating the core concepts of Bitcoin multiple times to engrain them in his brain :-) Mentioning the word "distributed" once in the intro is not sufficient, you have to build a presentation, a talk, that mentions multiple times the core concepts and why they are necessary for X, Y, and Z, to make sure you don't lose the audience.

gavinandresen: pause frequently during your talk to ask if the audience has any question. For technical talks, I favor this technique over reserving the end of a talk for Q&A time. Remember that the crypto guys in the audience will get it, even if the presentation is of average quality. Your goal is to make the non-technical guys understand it as best as they can. Quite frankly Bitcoin is very hard to explain correctly. I have never watched or listened to a Bitcoin presentation that did a good job at explaining it. The only reason I understand it is because I read the white paper and have a crypto background.

Place the cutting disk along the lateral axis to not damage the pins: Tip to Use a Dremel to Cut Open a PCIe x1 Slot
This technique is really safe and should take you less than 1 min to cut one.

It is an either/or choice I offer you:

1. Month-to-month contract: 331 BTC the first month, then (assuming a 40% difficulty increase every month) 236 BTC the 2nd month, 169 BTC the 3rd -> average 245 BTC/month
2. Or 4-month contract, fixed price of 270 BTC/month

Either way, I am less expensive than BitcoinRigs or vladimir (~280 BTC/month).

You pick #1 or #2 depending on your own prediction of the difficulty. Again I can't promise the exact price for option #1 as I modify the prices on a monthly basis, but you can cancel at the end of any month. And a 40% difficulty increase is really conservative: the average increase in the last 2 months was 60%, from 36k to 92k (36k*1.60*1.60 = 92k). That's why I am pretty certain I can say even option #1 will end up cheaper than my competitors.

Your math is correct (more precisely you generate ~330 BTC/month at current difficulty). My offer is not pointless: see posts from Raulo and jgarzik in this thread who explain how it can be useful.

Yes some people value new coins more than old ones: here is someone asking to pay in Bitcoins to get Bitcoins.

You can also buy my services as a charitable act if only to make the Bitcoin network stronger, or to rank high in public mining pools statistics and impress your friends :-)

Whatever the customer demand is, I am here to satisfy it!

Yes, I do offer the cheaper deal ("my price will be an average of 245 BTC per Ghash/s/month over the next 3 months" -- a conservative estimate.)
As I said, both BitcoinRigs and vladimir lock you in for 3-4 months (you will generate much fewer than 326.7 BTC/month by then.)

Edit: if you prefer their kind of flat pricing over 3-4 months, I can match that! 1 Ghash/s for 4 months at a price of 270 BTC/month. What do you think?

I am correct:

* BitcoinRigs: 1337 (USD for 3 months for 1 Gh/s) / 3 (months) / 1.60 (USD/BTC) = 278 BTC
* vladimir: 4444 (GBP for 4 months for 4 Gh/s) / 4 (months) / 4 (Gh/s) * 1.65 (USD/GBP) / 1.60 (USD/BTC) = 286 BTC

(1.60 USD per BTC was the average these last few days - the run up to 1.90 USD is unlikely to last.)

I have lowered the price to 331 BTC per Ghash/s per month.

Don't forget that TOP500 refer to double precision numbers. So that 4.7 PFLOPS double precision is 9.4 PFLOPS single precision (these numbers x 2). The latter number is what should be compared to 8 TFLOPS (single precision) in your excercise.

Well, from another viewpoint, the 2 biggest pool operators, slush (180 Ghash/s) and Tycho (160 Ghash/s), effectively control 49% of the global network hashrate (the last 3-day average was ~700 Ghash/s). Based on historical growth, in the near future they will assuredly indirectly control more than half of the global hashrate. If there is collusion between these 2, they can in theory start doing evil things.

Of course I trust them today, but a pool duopoly (or monopoly) is not something I am comfortable with in the long term.

I guess this thread is an opportunity to summarize the top miners (>10 Ghash/s) who have publicly discussed their setup:

o vladimir: ~40 Ghash/s (source: his comment). Has said publicly he uses a bunch of Radeon HD 5970, hosted in UK datacenter(s). He sells mining contracts.

o ArtForz: 38 Ghash/s (source: IRC #bitcoin-dev #bitcoin-mining). Half of his mining power (~19 Ghash/s) is provided by 28 x HD 5970 + few other ATI GPUs. The other half (19.2 Ghash/s) is provided by 96 structured ASICs, from his own design. Claims 200 Mhash/s per chip at about 10W each. Pictures of his GPUs: http://bayimg.com/eABDfaAdd http://bayimg.com/KAAeaaAdp

o Myself: 15 Ghash/s: bunch of HD 5970, plus a few HD 6990. Open-air motherboards on rack shelves (kind of like Facebook's OpenCompute servers). Most of my nodes look like this: http://blog.zorinaq.com/?e=47

I don't know who the 4th guy with >10 Ghash/s could be.

Edit: this is gutsi with 14 Ghash/s with 5850, 5870, and 5970 cards.

More shaders that account for a 2x-3x advantage and presence of a native integer rotate instruction that accounts for another 1.7x advantage, see:
Why are AMD GPUs faster than Nvidia GPUs?

The reasons games are not 2x-3x faster too on AMD GPUs is because they have a harder time fully utilizing the superior number of shaders (AMD's VLIW microarch is hard to exploit), whereas Bitcoin, being what we call an embarrassingly parallel workload, has no problem exploiting them.

I replied to you who stated "ATI has not released the Linux drivers yet" which is untrue:
- 11.3 was fixed to support the HD 6990 (previously aticonfig displayed "no supported device")
- There are people mining successfully, with poclbm and Diablo's miner (not only my miner), on Linux, on the HD 6990.

A glitch on the website doesn't mean it is unsupported. Like I said, select "HD 6xxx". You and I can argue about the meaning of "fully supported", but what I want to convey is that it actually works. Again, people are mining with it. You may want to try poclbm which runs faster than Dialbo's miner on the 6990.

Yes, please put a link to the whitepaper prominently on the front page. I did not take Bitcoin seriously until I stumbled upon the paper.

As stated numerous times, this is untrue. The 6990 is fully supported under Linux by the Catalyst 11.3 driver. I am mining with it!

The confusion comes from AMD's website that only lists Windows 7 drivers when selecting "HD 6990". Select "HD 6xxx" instead.

Yeah, as documented in the wiki, these speeds are only achievable with my hdminer.

I measured power efficiency of the 5970 vs 6990 here, my miner is able to close the perf/Watt gap between the 2 cards to within 5% (compared to your 850/700 = 21%). It is likely that, combined with memory downclocking, one can reduce that gap to 0.

(BTW please pm me or email me m.bevand@gmail.com when/if someone claims my bounty. I don't always monitor this thread.)