Could have done something ultra-elegant but there's a trade-off between getting the damn things making me money, and playing around trying to be Jonathan Ive.
Very true  Parts of your effort have technical merit (eg. running fans on separate power rails), but the rest is eye-candy that I personally wouldn't want to spend time on (eg. making individual power switches for each FPGA). But congrats, it looks nice! At least, you enjoyed building it. My goal in designing an FPGA farm is to spend the less amount of time setting it up, and to make each part easily serviceable (fans, boards, cables). For example I don't understand at all the other guys stacking up boards with spacers screwed in. If they ever need to replace a board in the middle, they would have to shut the whole stack down, and to disassemble it (ugh!). |
|
|
|
There are already arguments upthread explaining why it's not a good idea and fallback procedures for DDoS attacks, etc.
Read above, I edited my post to explain why I think he will be unable to defend against DDoS attacks. |
|
|
|
I'm sure he'd be willing to make a version especially for you that donates 20%...  You are losing sight of my main argument: I am against the complexity of this centralized infrastructure that will get attacked, will get DDoS'd [1], making us all lose (eldentyrell wasting time effectively doing sysadmin work, and customers' miners being down). I am only willing to reward him more if he would distibute his bitstream the way I described.
[1] The anti-DDoS measures that eldentyrell describes in his FAQ seem trivial to bypass. He seems to only think about dumb attacks like SYN floods. For one, I can think of clients submitting many legitimate signcryption requests to effectively tax his CPUs, not network links. Let's be real, DDoS proofing a service is hard and costly. Eldentyrell certainly cannot do it given his very small profits (again, only $1300/month assuming 1000 FPGAs using is bitstream). For comparison, MtGox spends more than a few thousand USD per month for their anti-DDoS service. |
|
|
|
Well then ET is more magnanimous than you. Even more reason to support his protocol.
Technically this makes you a tightwad Whereas I am willing to compensate eldentyrell more, if he would distribute his bitstream the way I described. |
|
|
|
So Alice has plenty of incentive to defect. She doesn't have that much incentive not to. Not everybody uses this kind of FPGA, so the difficulty increase from propagating the data is minor, and hasn't much effect on Alice's own mining profits.
[...]
Even if this somehow works, it creates unnecessary risk in the valuation. Suppose someone develops an equivalent open-source bitstream tomorrow. Then the data will become worthless, and in retrospect, the data is almost worthless today. When Alice buys the data she doesn't know for how long it will have added value, so she's basically gambling. There's nothing inherently wrong with a risky investment but it's still inferior to a product where you pay exactly what it is worth. With a cut taken by the creator this is what happens - as long as his product creates value he receives a reward proportional to the created value, when it no longer does he's no longer rewarded.
Thanks. You make very good points in your rebuttal (Alice has incentives to defect, and this is basically a risky investment.) |
|
|
|
This entire "he's going to make too much money/btc" thought is a bunch of horseshit.
You don't understand me. I am saying the exact opposite. eldentyrell's method will make him too little money ($1300/mo with 1000 FPGAs). My method would likely compensate him more. |
|
|
|
Ask EldenTyrell how much time he spent on this bitstream, and how much he could get for that time say for design of something else like that under contract development.
Do the math yourself. You estimated there are 1000 LX150s in the wild mining (probably about right). Assuming all of them run his bitstream, eldentyrell would only make $1300/month. This is not fair compensation. The number of LX150s will increase over time, but it is likely that within 6-12 months, a free/open bitstream matching his performance will be released, reducing his income to zero. With my suggestion, I think eldentyrell could receive more, likely thousands of BTC (I would pay 100 BTC!) |
|
|
|
eldentyrell developed a complex non-self-sufficient bitstream interacting with a centralized infrastructure via encrypted messages to take a reasonable share of the profits of people mining with his fast bitstream. I think this complexity is unnecessary. There is another way to compensate eldentyrell that would (1) be simpler, and (2), allow him to make more money. I propose that miners pay a variable amount of BTC to be on a waiting list to receive the bitstream before others; each buyer is given a time slot when he will receive the bitstream; the more a buyer pays, the higher on the waiting list, the earlier he receives it. In other words, miners compete with each other to have the privilege of being the first ones to profit from the bitstream, before the rest do (or before the bitstream is leaked). Example: - Alice pays 100 BTC, eldentyrell will give her the bitstream on week #1 - Bob pays 80 BTC, gets it on week #2 - Charlie pays 60 BTC, gets it on week #3 If Alice is a large miner and wants no one else to get the bitstream for a few weeks, she can pay to reserve multiple time slots (yay competition!): - Alice pays 100 BTC -> receives the bitstream on week #1 - Alice pays 80.001 BTC -> week #2 - Alice pays 80.001 BTC -> week #3 - Alice pays 80.001 BTC -> week #4 - Bob pays 80 BTC -> week #5 - Charlie pays 60 BTC -> week #6 If a buyer fails to pay on the week of his time slot, eldentyrell simply skips him, and goes to the next on the list. Early buyers paid a lot of money to receive it early, and have little incentive to leak the bitstream. Buyers down the list are more likely to leak it, but by this time, eldentyrell should have received most of his compensation. I think eldentyrell could receive thousands of BTC this way. I, alone, would be ready to pay 100 BTC to reserve a slot! Comments? List interested buyers so far: - mrb 100 BTC
- needbmw 50 BTC
|
|
|
|
Another thing is what if bitfury gets the same idea and they release their 300 mh/s bitstream with 5% comission. No point in using a slower bitstream. Or even an improved open source bitstream that goes faster than 260 mh/s.
I will not do that as mine bitstream would not work with any board around there... not enough power for it and future version... So for existing spartan board I am not competitor to his bitstream niche. At least Icarus and Enterpoint boards can deliver 12A per FPGA. This should be enough for running your bitstream. (You said your FPGAs consume 12W at 1.25V core, that's only 9.6A). |
|
|
|
Can't edit post above but it seems there is this writing on the chips this time around :
NDBAN20242
4PDHATWOA
Can't find anything about these but they look very strange indeed for a FPGA chip.
These are Altera's Lot Number and Trace Code. https://bitcointalk.org/index.php?topic=84651.0 |
|
|
|
Edit: you changed the picture from your original post. It does look more similar now...
Yup. The one I had linked to was a much smaller Arria II with only 45k LE. |
|
|
|
Compare this Altera Arria II EP2AGX260 FPGA against the 2 chips on BFL's Mini Rig card shown in https://bitcointalk.org/index.php?topic=75764.msg906496#msg906496 :   The marking near the bottom edge of the package (blurred in the reference pic) is obviously Altera's Lot Number and Trace Code. The chip's package is identical. Same 4 asymmetric golden dots in the corners. Same PCB traces under the green soldermask on the 4 borders. Same arrow in top left corner. This is an Arria II for sure. Which one though? Well we know the Single is based on the Stratix III EP3SL150 (142k LEs at 65nm) and mines at 832 Mh/s. BFL claims a Mini Rig card mines at 1.5 Gh/s, so its FPGAs must have 1.8x more LEs, or about 260k... And there is one model matching exactly this prediction: the Arria II EP2AGX260 (257k LEs at 40nm), see http://www.altera.com/devices/fpga/arria-fpgas/arria-ii-gx/overview/aiigx-overview.htmlArria II being 40nm also makes it close to the 45nm Spartan 6 in terms of efficiency (about 20 Mh/Joule), which is exactly what BFL claims (25 Gh/s at 1250 W for the whole rig). |
|
|
|
yo,man,you just skip my post!
Sorry, do you have an email where I can contact you? |
|
|
|
Come on guys, I ask you to make offers, but you never do and always ask me to quote a price... that's no fun! Alright David_Benz... let's say I can sell you 15 cards for 1190 BTC. I will save you negotiation time, this is my rock bottom price. Okay and that would be about 12 Ghash, right? How big of PSU do I need to run 2 of those on one machine? Thanks. At stock clocks, 10.5 Gh/s. Overclocked, sure, many people do 770 Mh/s per card, so 11.6 Gh/s. 2 non-overclocked GPUs (but mem downclocked), on a super-lean config draws only 670W (730W at the wall), so a 700W PSU would be sufficient. If you overclock, prefer at least a 900W PSU. Personally I run 3 cards with downclocked memory on a 1200W PSU (1110W at the wall). |
|
|
|
Come on guys, I ask you to make offers, but you never do and always ask me to quote a price... that's no fun! Alright David_Benz... let's say I can sell you 15 cards for 1190 BTC. I will save you negotiation time, this is my rock bottom price. |
|
|
|
... a stack of boards sitting ready to go ... Can you give any rough number/quantity to that "stack of boards"? "tens" leaves me as a less than happy camper. (Not unhappy. Just less than happy. ) Hundred or more leaves me with hope that they might quickly catch up with order backlog. tens would make me a happy camper. In fact, only ten (singular) would make me happy... as I am next in line |
|
|
|
Notify is not the same as subscribe. Notify sucks giant balls.
Yep. And the bookmark mod would emulate subscribe. |
|
|
|
|
If I have not replied to someone's email/message, it is because the offer was not interesting enough.
I still have 6990s available for sale. Contact me, people, and make offers!
|
|
|
|
|
Show Posts
