Official Thread: AMT

[ISAWHIM]

Voltage changes are fine, as long as they are software controllable. Having to remove and replace resistors to adjust the voltage is beyond the ability of most customers, and cannot be considered after sales tuning. Adding additional modules (which was the whole point of that) and then underclocking the boards to improve efficiency is likewise not "tuning".

I will give you the 10%, but only once. You can either have the hashrate +/- 10% (minimum 1080GH/s) at 900W, or the hashrate at 1.2TH/s and the power at 990W.

I can agree with the top portion... but saying one or the other, when hash-rate and power are both +/-10%, still seems unfair to the bet. If I happen to get one with slow chips, thus the lower hash-rate... I will be forced to draw more power, and vice-versa... If I get one that hits 222GHs per board (assuming it is 8-chip boards, and 6 boards)... it would unfairly reduce load on the PSU, if I made it run the lower end of the voltage. That is like flipping a coin on production errors. Not enough play to even be within the updated estimates.

How about this... If I am horribly wrong, and thus they don't come within those specs at all, by the 10%. (Which is possible that they are still under-estimating...) Then I will put a link to that public admittance of defeat in my signature, and leave it there for 3 months... (Why do I feel like I am going to be right on the edge on this bet. lol.)

If I win... you just enjoy your miner. And if you are ever visiting Florida, you will owe me a beer... even though I don't drink.

[1714704674]

1714704674

[Syke]

I can agree with the top portion... but saying one or the other, when hash-rate and power are both +/-10%, still seems unfair to the bet.

It is fair. They claim +/- 10%, you want to have -10% on hashrate *and* +10% on power draw which is the absolute most extremely lenient way to deliver. You can have both and be fair.

[MrTeal]

Voltage changes are fine, as long as they are software controllable. Having to remove and replace resistors to adjust the voltage is beyond the ability of most customers, and cannot be considered after sales tuning. Adding additional modules (which was the whole point of that) and then underclocking the boards to improve efficiency is likewise not "tuning".

I will give you the 10%, but only once. You can either have the hashrate +/- 10% (minimum 1080GH/s) at 900W, or the hashrate at 1.2TH/s and the power at 990W.

I can agree with the top portion... but saying one or the other, when hash-rate and power are both +/-10%, still seems unfair to the bet. If I happen to get one with slow chips, thus the lower hash-rate... I will be forced to draw more power, and vice-versa... If I get one that hits 222GHs per board (assuming it is 8-chip boards, and 5 boards)... it would unfairly reduce load on the PSU, if I made it run the lower end of the voltage. That is like flipping a coin on production errors. Not enough play to even be within the updated estimates.

How about this... If I am horribly wrong, and thus they don't come within those specs at all, by the 10%. (Which is possible that they are still under-estimating...) Then I will put a link to that public admittance of defeat in my signature, and leave it there for 3 months... (Why do I feel like I am going to be right on the edge on this bet. lol.)

If I win... you just enjoy your miner.

You're not the one making the measurements, as stated in my original bet proposition. No offense, but you have too much on the line to trust that you'll document the results fairly. All that is required is any two people reaching these numbers of all the customers that AMT has for their 1.2TH/s miners. I suppose I should add a caveat that the account must have registered prior to Sept 1 / 2013 (the beginning of the month this thread was started) and have at least 30 posts in 2013.

I am using the claimed efficiency of 0.75J/GH, and giving 10% on that (0.825J/GH). I also restructured it so that you could have the hashrate 10% lower than advertised with the same max power draw, which is actually easier to hit. Those are the terms for the bet I proposed.
If you would really like a bet where the minimum hashrate is 1.08TH/s and maximum power draw is 990W, I would entertain that, but only at 1:1 odds.

[ISAWHIM]

I can agree with the top portion... but saying one or the other, when hash-rate and power are both +/-10%, still seems unfair to the bet.

It is fair. They claim +/- 10%, you want to have -10% on hashrate *and* +10% on power draw which is the absolute most extremely lenient way to deliver. You can have both and be fair.

Wow, now that is more structured of a bet... (The post that followed this quoted one. Saw it as I tried to reply.)

But it says +/- 10% on all specs... Power-consumption is one of the two specs... Power and speed.

For the bet, if you accept my public humiliation as payment... I will take the power +/- 10% (At the wall), as that is what one would assume, for the device. (Since that is what we get charged for, by the power company.)

In light of "friendly" betting... Since "I am not building these units", and I still assume any "estimated" specs are just that, until a unit they built is spec-ed out once it is in their hands. I have no way of knowing what path they have taken to build them. Running minimum chips, it is not possible. Running similar qty of chips (48), it is borderline possible. Running more than "my estimates", (more than 48 chips), it is absolutely possible. However, that is not mentioned anywhere. They may have done estimations with 50+ chips, and decided to build with 48. That has not been determined yet. (I think minimum I calculated would have been like 34 operating at turbo, which put wattage over 1200W, obviously.)

Also, I chose the power one, because the PSU is obviously limited to those specs. Thus, that will determine, with the number of chips, what the maximum speed is. (Not the average, because it says nominal +/- 10%, for speed and also +/-10% for the other specs. That is two separate statements, not one, since both specs are mentioned. It does not say speed only +/- 10%. But I will take power.)

If AMT responds with the qty of chips used... I can make a more clear/confident bet. Without that knowledge, I am left only to believe that they will try to fill the estimated obligations they outlined. Using however-many chips is needed to reach that target.

I still believe they spec-ed the original estimates "by chip", and "at the boards total draw"... which was a mistake to do. (Unless that was specified as the specs specifics.) However, I do believe the updated specs are "at the wall", and "for the unit as a whole". They have not stated otherwise, so that is a reasonable assumption. Remember, they had to create estimates off what the chip-producer said were the estimates. They just failed to say, "specs may change at any time, and are just estimates". But again, we all knew this item was not even created yet, except on a test-lab board, with one chip.

If I am allowed to help those two people tune the machines... then I accept random peoples submissions.. but they could purposely make it do less, or draw more power, or just not have the ability to tune it. Also, you can't pick the two people...

and I changed my beer to a coffee...

[ISAWHIM]

I am using the claimed efficiency of 0.75J/GH, and giving 10% on that (0.825J/GH).

If you would really like a bet where the minimum hashrate is 1.08TH/s and maximum power draw is 990W, I would entertain that, but only at 1:1 odds.

1.2THs <- Spec #1
600-900W <- Spec #2

"Product specifications may differ from  (+/- 10% running variance)"
Two specs... (plural) two seporate +/-10% variations. If it was just speed, it would say only "Speed +/-10%".

So the range, of highs to possible lows are... (by specs)
1080Ghs lowest nominal speed you should ever get, within specs
1320Ghs highest nominal speed potential claimed, within specs

Both those speeds, potentially within...
810w lowest possible max the unit may be limited to, this is the -10%
990w highest possible wattage the unit is speced to, this is the +10%

So unless we have a dud PSU, that actually only delivers 990w for a minute, then 810w for an hour...
That places the estimates of worst power consumption at 990w, for obtaining the "nominal" of 1080GHs, and also the maximum draw possible from the unit, to reach 1320GHs.

That is 990w/1080GHs=0.92J/GH (worst case by specs max)*
That is 990w/1320GHs=0.75J/GH (best case by specs max)
* That would be a bum batch of chips, sucking lots of power from the PSU to stay in spec, over-volted.

You would not assume that hitting 1320GHs would be achievable at the PSU's lowest rating, or the highest low-rating of 810w. However, I would go so far as to say that the 1080GHs, should still be able to hit at 810w, as that would be -10% hash and -10% max-power, which should yield "nominal"...

So...
That is 810w/1080GHs=0.75J/GH (worst/worst case by specs min)*
Unrealistic is 810w/1320GHs=0.62J/HG (super-ideal case by specs min)
* That would be a bum PSU with hungry chips, which would have to be under-volted to be 1080GHs

So I believe, by the claims, that they are saying, 0.75J/GH to 0.92J/GH, as the (+/-10% to each spec, 0.835J/GH would be the average) I left-out the spec for 0.62J/GH, since that is unrealistic to specs intent, and thus, might never be possible.

The lower wattage would be for power-saving operation, and is irrelevant to the 1.2THs "nominal limits". They would only apply to how low you could under-volt it, and still provide power. Which the 600w -10% should handle.

So, to win the bet... The only source to beat would be 0.92J/GH to 0.835J/GH, as that would be low to average, within spec. Obviously someone may get a super-PSU and super-chips, and that would be unfair to judge with. Right?

Honestly, I am more interested in the maximum potential... electric consumption is nearly irrelevant at these hash-rates and yields. 1200w is roughly about $4.61 a day at $0.16/KWh... (I pay about $0.08)... Earnings are between $500-$150 per day, at the moment, after paying electric. If I get 50% more hashing-power at twice the electric-costs... That is $750-$255 per day, at the moment, after paying 2x the electric consumption. Yes, I estimated as 1200W, just for the sake of argument. (1200GHs * 150% = 1800GHs or 1.8THs) By the time I actually cash-out, to reinvest again... that should be about $3000-$1000 per day again, as if I earned that the whole time. Since cashing-out as you earn is the worst payout idea ever. lol. (Shh, it is those peoples losses that become my gains.)

[AMT_miners]

AMT just "updated" their wattage on the site  to "900W -1200W." Can you believe it! What if they come in "below" expectations, how cool would that be!?

Seriously guys, have a field day with it. We're still selling the most economic chip on the market.

And the dancing cat was just cute and excited don't you think. Today was a good day guys, lets make the best of it and get excited for next weeks miners.

Lead time on the 1.2Th's is still end of march. Orders that get in soon will still ship by March 31st, just a heads up we'll take down that offer on Friday.

[ISAWHIM]

So how many chips are in the 1.2THs miners... Now that I seem to have a challenge ahead of me... (Wait, he didn't accept my bid! lol... saved by the bell!)

I wrote that long-winded explanation for nothing... I am not re-writing all that with 1200W as the specs...
Short version...
0.91J/GH - 1.11J/GH is the worst to expect now, from nominal...
0.72J/GH - 0.835J/GH is the best to expect now, from nominal...

I am guessing that 48 chips is the design... (6 boards with 8 chips each)

Lead time on the 1.2Th's is still end of march. Orders that get in soon will still ship by March 31st, just a heads up we'll take down that offer on Friday.

New purchases... not existing orders... (Just wanted to have you clarify that. )

[Syke]

I wrote that long-winded explanation for nothing... I am not re-writing all that with 1200W as the specs...

You don't need to rewrite it. The new specs are only valid for orders placed after today. All previous orders must receive the stated specs at the time the order was placed.

[ISAWHIM]

I wrote that long-winded explanation for nothing... I am not re-writing all that with 1200W as the specs...

You don't need to rewrite it. The new specs are only valid for orders placed after today. All previous orders must receive the stated specs at the time the order was placed.

Hush, I was looking for an advantage to win! lol...

Waiting for my 80-chip 1.2THs miner to arrive, that operates under 600w max. "At the wall". (lol)

Seriously though... I don't want to have to wait a whole two weeks to find-out how many chips my miner will have. What will I do for all my days? Then have to spend the next three months eating my words, if it only delivers 1.2THs running at exactly 1200W. Though I still doubt that will be the case.

I... I.. I just want a hug! Throw me some feels!

[clenell]

why is it AMT wont squash all this nonsense and just state how many chips they will be using? or did i miss this?

[RickJamesBTC]

why is it AMT wont squash all this nonsense and just state how many chips they will be using? or did i miss this?

You didn't miss anything. The only reason not to tell you, not to post pics of the boards they are having custom designed, is that they don't exist yet. Otherwise, they should be excited and proud to show off the upcoming product.

[Mblackout]

what power supplies are you using.   Can i connect them to 208 v ?

[Chazaki]

Somebody posted a link to their flickr where they have a few more pictures. Here's the link in case anyone missed it from earlier:

http://www.flickr.com/photos/107201528@N06/10583836383/sizes/o/in/photostream/

No idea what to make of all of this. I agree with everyone AMT needs to state the number of chips and modules. And if units were supposed to ship this week it shouldn't be too hard to get a picture of the manufacturing facility (of which we know very little if not nothing about) and the miners on the assembly line.

I'm hopeful though, because like many, I am invested in this company.  If AMT would just spend 7$ an hour on some intern to respond to phone calls and emails I'm sure people would benefit tremendously.

Edit: noticed the pictures are old boards so they have no relevance to the 1.2 TH/s. Apologies.

[loshia]

...But it is not making your miner from 1t to 1.2t if you follow my thoughts.

When you solo-mine, your miner is not wasting "time", "bandwidth", "processing", sending your wallet or cgminer, the junk-shares. They just grab the next work-load and mine if they find no solution. Yes, it can drag a 1.2THs miner down to 1.0THs of average output. (Prior to the pool).

When you tell your miner to "give me the junk that is ____ diff", so you can submit it as a share... this is what it does...
1: Saves that hash for the diff you wanted (Has to check each result and compare it to your requested value)
2: Checks that hash, to make sure it is valid
3: Sends that hash through the stream (A stream which is shared with other miners sending this crap back also.)
4: Wraps it in a package to send to the pool
5: Sends it through a TCP boradcast, waiting for an "OK, got it".
6: Resends, if that fails... (skip this if it got back an OK, repeat this over and over until OK is gotten.)
7: Works on next load now (repeat from step 1 for next diff)

As opposed to solo-mining...
1: Check found block
2: Submit found block (Repeat from step 1)

Time lost "submitting junk", at the CPU/miner is enough to be noticed. Varies by each chip... depends how fast they puke-out info, and the CPU can direct that info in the threads, and process that info for packing, and sending.

The more "shares" (low diffs) you send, the more you lose in actual processing time. Not to mention the other-end, where you are right... higher diffs = more potential share loss, as they take longer to find. Thus, the need to tune the shares to the speed of the units. And... for 1THs, 128-256 is more "tuned" for less losses of both the processing at the miner and losses from the pool. (Seen more on faster blocks, as those do more "resetting workload", which is where the "workload-size" comes into play.)

I tune machines like crazy. Voltage is step 1, delivery and processing is step 2, cooling is step 3. Beyond that, there is nothing much that can be done. But it all matters, and 1-10% loss from dozens of machines matters a lot more than 1-10% loss from one machine. But I will go over all of that in a full review, once my miner comes.

You know what? Best case will be to mine at 2048 diff and pretend your miner is 2t instead of one. 1t miner is one 1t miner and it is staying 1t no matter what diff you are using. Simple as that. However you can earn more btc with same miner if you use it properly but it is not making it 2t
Do you make a diff between broadcast any cast multicast and UNICAST address at all You look pathetic with your conclusions dude. And your tech explanations do sound like .........You have to learn a little bit before spreading wisdom all over. PERIOD!
Gys, to all of you who listen and take ISAWHIM posts seriously. I am strongly suggesting to start mining at 10K diff immediately and your hash rate will go up 10K times GUARANTEED . Make sure you are using TCP !!!BROADCAST!!! ADDRESS waiting for an "OK, got it" as suggested/wisely explained by ISAWHIM   . Otherwise you are in a deep shit and you will earn 0 BTC

And finally make sure your home switch do support IGMP (multicast) snooping group your miners and go to cisco academy at the end. That will bring you 10 times over your hash rate guaranteed! BUT you need CCNP certification at least  

[athgaurd]

Well... this is something that I recently found... not sure how long ago it was added, but it wasn't readily easy to find.. however, you can at least put in your order # and email and it will tell you somewhat of a status.

http://advancedminers.com/shipping/

[athgaurd]

The breakers were installed in the existing breaker-box, the wires were run in conduit, by an electrician friend. (outside the walls) The outlets, that tie to the raw lines, were done by me. Anything at the "outlet-box", is allowed to be done by the home-owner, here in Florida. The rest, for insurance purposes, had to be done and also inspected by a certified electrician.

Only cost about $350 for everything to be done. Friends work for beer.

(As for the bet above... change that to peak, and not average, and it would be a reasonable bet. But not one I would be willing to submit to, as "shit happens". Since "out of the box", is also not specified, and "optimization" isn't specified.. I would take you up on that offer, if you allowed it to be "judged" after a suitable "optimization period". Also, once AMT says how many chips will be in my unit. Without that, it is a blind bet. You have a contradiction too... you say "hit 1.2THs", then say "average 1.2THs"... that is a big contradiction... "Hit 200MPH", and "Stay running at 200MPH average", is not the same. Do you even lift bro?)

Also note: I think I am getting screwed by the power company though... we put all lines on one half/leg of the power... So it seems like I am drawing twice the amps at the meter. If you do setup new breakers, have them balance them to the two separate legs. If you draw 100a on one leg, the meter here seems to see 200a because it seems to measure the greatest draw across both legs, as amps, but charges for watts. It seems to assume if 100a is drawn on one, that 100a is also being drawn on the other. If I had that balanced, it would be 50a and 50a on each leg, and the meter would read 100a total, not 200a total. That is my next thing to fix. (That, or I need a PF-correction unit.) Electric meters are such horrible technology. (Again, this is what seems to be happening, when I measure the timing outside. It does not match my actual draw of the units. Measured directly. Chime-in if you know the solution to that one. That solution was told to me by the guy who installed the breakers. He didn't think it would be that dramatic of an offset.)

Yes, I had the power company come to reset the meter and "test it", they said it was working fine. Sure... it only charges me double, it is supposed to do that! lol. So much for paying half price. (It isn't quite double, but it is more than measured draw.)

Balance your breakers. You should have two busses, and one neutral bus (usually physically two since it's also the ground, but they're tied together). Make sure you stagger the breakers so that if you have an even number, they are 1:1 on the two busses. Meters are actually fairly accurate, but they do count parasitic losses and unbalanced loads. The meter would have to be a much more complex and expensive device to do as you suggest, whereas balancing the load across the two busses is easy. Just make sure the main is off and/or you have high rated insulated gloves

Never heard of this for a residential application... however makes sense on a commercial application for sure.

[ISAWHIM]

You know what? Best case will be to mine at 2048 diff and pretend your miner is 2t instead of one. 1t miner is one 1t miner and it is staying 1t no matter what diff you are using. Simple as that...

Funny thing to notice... on the video, where it is running with a 8-diff to start at 780-999GHs... near the end, the pool auto-adapts to 512-diff, and the miner instantly averages 1.002THs... Let it run a little longer, it would have gone to 1024-diff and ended with an average of about 1.1THs, I am sure... (Missed that last time, it was only like a few seconds of the last frames.)
http://bitmine.ch/?p=5176

I am sure they cut the video off on purpose, to hide the fact that it would do even more. That average shot-up way too fast, after having to also calculate the 2 minutes of 780GHs, from the cold-boot, in the beginning of the video, where it took forever to climb to 999GHs from 998GHs average.

But you are right... Even watching cgminer choke, live, in that video... While the 8-diff is there, still there once the 512-diff comes. with the bottom card stalling below average as 180GHs while the others all ran smooth as butter at 222GHs... Yet, I am wrong. lol.

I can make you a video too. One that shows my miners pulling about 710KHs with diff-1 submission, another with the sane units pulling 730KHs with diff-32, another with them running 762KHs running solo, another with them running 781KHs solo and workload-64 size... then tell me again how my 700KHs miner is not a 781KHs miner. I could do some port-snooping and packet sniffing on the network, and accurately measure the output, along with all the repeating failed buffering of cgminer's horrible http/tcp protocol sorta-standard communication.

In the end, for those who do mine in pools, it does matter what actually gets accepted as work, for earnings.

[loshia]

You know what? Best case will be to mine at 2048 diff and pretend your miner is 2t instead of one. 1t miner is one 1t miner and it is staying 1t no matter what diff you are using. Simple as that...

Funny thing to notice... on the video, where it is running with a 8-diff to start at 999GHs... near the end, the pool auto-adapts to 512-diff, and the miner instantly averages 1.002THs... Let it run a little longer, it would have gone to 1024-diff and ended with an average of about 1.1THs, I am sure... (Missed that last time, it was only like a few seconds of the last frames.)
http://bitmine.ch/?p=5176

But you are right... Even watching cgminer choke, live, in that video... While the 8-diff is there, still there once the 512-diff comes. with the bottom card stalling below average as 180GHs while the others all ran smooth as butter at 222GHs... Yet, I am wrong. lol.

I can make you a video too. Ont that shows my miners pulling about 710KHs with diff-1 submission, another with the sane units pulling 730KHs with diff-32, another with them running 762KHs running solo, another with them running 781KHs solo and workload-64 size... then tell me again how my 700KHs miner is not a 781KHs miner. I could do some port-snooping and packet sniffing on the network, and accurately measure the output, along with all the repeating failed buffering of cgminer's horrible http/tcp protocol sorta-standard communication.

In the end, for those who do mine in pools, it does matter what actually gets accepted as work, for earnings.

There is no need for video to be made dude. Make sure you are using TCP Broadcast adress waiting for pool ok that is enough for me If you made me a video of that i will appreciate it very much
check cgminer code it is diff independent counting shares. It returns all shares diff one. cgminer checks it against diff1 if it is HW error and counts it as good SPEED COUNTERS but submits only the ones matching diff. But i suggest you to focus on Broadcast address and movies ywhere you are good at . I am doing you a favour and saving some of your valuable brain time.
To make it more mysterious - diff can be passed to chips and they can return only shares matching diff but that was not IN THE VIDEO!

[ISAWHIM]

TCP Broadcast adress waiting for pool ok

Wow, you are funny, that was a short answer, obviously.
There is a "TCP broadcast", to the "address your miner is connected to", which waits for an "OK, work accepted". That is how CGminer knows to display, "Rejected shares" and how it detects "New block" so you can dump your work, to start loading and processing new work.

Should I have put all that, or just said...
TCP boradcast address waiting for pool OK.

TCP is a slow protocol that is bloated. It makes solid connections to ensure packet data is broadcast "completely". If not, it resends it, or starts a new connection if the OK is not received. UDP would have been better for speed, but is often blocked by ISPs and those fragmented packets are never guaranteed to even be delivered. That requires you to use separate UDP tracking, or just use horrible TCP standards. HTTP, the thing broadcast over TCP, by cgminer, is even more bloated. there is an additional "OK" status for HTTP, in the bloated header, so it knows the request was a success.

I guess anyone can mine. Must be why it is growing in popularity. all this OK broadcasting from your address. You are just inviting them in.

[loshia]

TCP Broadcast adress waiting for pool ok

Wow, you are funny, that was a short answer, obviously.
There is a "TCP broadcast", to the "address your miner is connected to", which waits for an "OK, work accepted". That is how CGminer knows to display, "Rejected shares" and how it detects "New block" so you can dump your work, to start loading and processing new work.

Should I have put all that, or just said...
TCP boradcast address waiting for pool OK.

Con, Kano help me with this please
You are embarrassing your self dude..Rely. I am not joking here...

 

Funny part is that you believe in the bullisht you spread. That is called COMPLETE lack of brain i have no other LOGICAL explanation.
The final comment for your broadcast shit is that any TCP broadcast adders can be used at layer 2 only.  got it? 7 layer osi model. Your posts are well quoted so do no twist it please.