They seriously make 2x 8 pin 7990 cards? That's really cutting it close.
I've got a 3x 8 pin powercooler 7990 on (unpowered) risers running fine @ ~1.08MH/sec, haven't fined tuned it much. 2x would be very ugly
+1. We both have non malta'ed 7990s and can see the truth. |
|
|
|
|
Two PSUs. One plugged into the GPUs, one into the rest of the system and any remaining GPUs. On the 24 pin connector for the GPU PSU, use a paper clip to bridge the pins [google paperclip psu].
This forces the PSU into an always on position, in liu of being told to power on and off by the motherboard.
When you want to turn on, you turn on the GPU PSU at the cable, which will turn it on. Then power the system like normal.
|
|
|
|
 Just to confirm, you see a solid 178.00 shares/minute always? I have never seen that ever, but I think mostly its because of stales pulling down the accepted efficiency. Could you paste one of your outputs from the webpage? Total MHS: 13046 Received: 0000091550 Accepted: 0000091170 Per Minute: 178.29 Efficiency: 099.58% Up Time: 0d,08h,31m,21s Total MHS: 13022 Received: 0000091634 Accepted: 0000091080 Per Minute: 177.94 Efficiency: 099.39% Up Time: 0d,08h,31m,50s Damn lenny, that 99.98 is sexual. For some reason I haven't seen either blade go over 24 hours. It either says its restarted, or it is actually restarting roughly once a day. Not got the new cooling on yet, but its hardly hot and its typically restarting at cool times of the day/night. |
|
|
|
|
Your cheapest option is another 750-850W PSU + a paperclip. Its the most price efficient setup, even ignoring the fact you already have one.
|
|
|
|
Well, I just left it to settle for a bit..
1: 12928 MH/s 271456/ 275231 176.65/min 98.62% 0s up 025h 36:41
2: 12966 MH/s 272246/ 275240 177.18/min 98.91% 0s up 025h 36:33
3: 12899 MH/s 249589/ 253824 176.26/min 98.33% 5s up 023h 36:01
4: 12945 MH/s 11683/ 11808 176.92/min 98.94% 0s up 001h 06:02
I think there's quite a lot of factors at play to be honest:
a) latency of my Internet connection as it actually submits shares
b) blocks found lead to a drop, because these units don't do LP
c) pool load
d) stratum proxy (and PC its running on) load
e) local network layout as you say, but can't do much about that, I need to use a switch or I haven't got enough ports :p
They're all around 177 shares/min which isn't catastrophic at the end of the day.
Ah well. Yours are clearly broken so you might as well give them to me |
|
|
|
Replied in PM. It's just a bit of very ugly Ruby, but if anyone else is interested, http://telon.net/asic_status.rb.txtI have local DNS entries that make my asics reachable at asic1/2/3/4, if you don't you may need to replace with an IP where necessary. Sorry it's so ugly and basic, it was only a 5 minute hack to get information into a Terminal I can watch Investigate the position of the proxy rather than the platform. Another user had problems going through a switch to the blades, rather than straight to the router. but the fact its changed at ALL, suggests there is room for improvement. |
|
|
|
this is the output of a little script I wrote to get all the interesting data into one place..
1: 12949 MH/s 240824/ 243858 176.95/min 98.75% 4s up 022h 40:58
2: 12986 MH/s 241489/ 243866 177.45/min 99.02% 4s up 022h 40:53
3: 12886 MH/s 218411/ 222339 176.08/min 98.23% 4s up 020h 40:24
4: 12790 MH/s 14848/ 15200 174.79/min 97.69% 4s up 001h 24:56
That's fairly atypical for me, maybe a little lower than normal, I've seen anything up to 178 shares/minute.
Hitting 13000 exactly, 178 exactly. You might want to try moving your stratum proxy about, or testing 2 different proxies on 2 computers etc. Does look like your 4 is maxing out current arrangement. |
|
|
|
Found my first block yesterday, enjoy my coins  Also, my namecoin is staying as 0. I entered a wallet address but nothing is happening. |
|
|
|
Read through their business plan cover to cover, and bet 2.5 no. There is no way they're going to demonstrate this in 5 weeks. |
|
|
|
|
You guys are way way too schizophrenic. Reddit would have a field day here.
|
|
|
|
About 1 hour left!  What is the deadline for payment? Is it 48 hours like last time?
Not stated, but everyone's expected to pay ASAP - hopefully below the 24 hour point after auction end.  Seems to be very unclear seeing as we don't know if 1) there is anything to win, 2) we actually won. For those of us in the UK converting from Fiat, its hard enough as it is. |
|
|
|
|
FPGA doesn't mean mining.
|
|
|
|
|
I would say don't attempt this. Removing 800W of heat is hard enough, to get it quiet on air almost impossible.
|
|
|
|
found my first block! |
|
|
|
It is my understanding that the current generation of ASICs all get around 100MH/joule give or take a few million hashes. While that is upwards of 20x more efficient than GPU mining, how much more can be pushed out? Given we understand exactly the operation needed, does anybody know what the minimum amount of energy that could possibly be expended to perform SHA256?
Do you know what Moore's law is? Something that hardly applies to bitcoins. Makes sense for the microprocesser market worth tens of billions of dollars (maybe into the hundreds of billions) When your NRE for an ASIC runs several million or tens of millions of dollars, you have to make a financial decision if it's worth the investment, risk, and design/lead times for a marginal improvement over what already exists Look at it this way : Asic Miner is a 100% capitalist, maximize profit type of organization. They used, what, 110 or 130nm node for their fab? You could get much better power efficency and hashrates with a smaller process, but it's far more expensive and complicated. They have no plans to change node, they're bringing 262 THash/sec online with their 110nm in the next 3 months. A massive investment that shows they're dedicated to what they've got. You really believe they're making 26,200 blades? |
|
|
|
Now the new M cards have dual slot coolers, ouch. At stock clocks they were loading just below the 80s, horizontally mounted in an open air, single card system. This was on GPGPU computing. Due to the heatsink design, ZERO air is exhausted and its all left to be removed by the case. My PC cards heat the case ambient up to 40+ even with 3x250mm, 5x240mm of fans removing heat. You will have to open air, just have to.
Tldr: Mining is doable on the 3 slot PC cards, but will be almost impossible to dual M cards within a case. The M cards are 2 slot and don't exhaust air at all. WRONG. mining is quite doable on the dual slot reference cards. See next quote remarks. Power delivery:
This is the major problem. My PC cards have 3x8pin for a max of 525W. These new M cards only have 2x8 pins for 375W max. Even with my moderate overclocks (focusing on power consumption rather than MAX POWA), I'm going over/hitting the power delivery limits of the PC cards at about ~400W from the wall per card [1050 core, mem 625 and undervolted]. AnandTech saw this in Furmark, the card was power throttled HEAVILY. Now imagine trying to scypt mine with decent clocks - its not going to happen. WRONG. ...the M 7990s seems best at scrypt mining and for bitcoin mining they still perform well at 1150-1250mh/s I believe they can even be pushed to 1300+mh/s All at ~375 Watts Tldr: The power delivery on these cards LIMITS the performance of these cards and causes power throttling even on moderate computing load. Hence these cards are not appropriate for mining. WRONG. this quote is where dogie is assuming stuff again, it was a good assumption, however 375w is more than enough to power both cores on the new Malta 7990. AMD has had about 1 year to redesign power consumption of the new 7990. WHY WOULD I SAY ONE CARD COSTS $2000, are you kidding me. I was always comparing dual cards. I'd ask you to quote me but I know you'll just come up with some excuse again and then lock the thread again.... OH WAIT ITS MY THREAD. Dogie did say it costs $2000, at the time I didn't notice how bad he was trolling so I did not capture the quote. it seems he had gone back and edited his post because everyone was thinking he was a moron. So good job dogie, you were really right about nothing. Next time get both pieces of the hardware, then do a unbiased test of both pieces of hardware in the same environment. Spouting numbers and assumptions in every 7990 thread solely based on your self-qualified "masters of engineering" experience and not having the hardware has made you look like a jackass. Sup Troll!?!?!?! So much bullshit in here its not even worth it any more. And I edited my post? That's convenient isn't it, I edited a post that didn't exist  |
|
|
|
I'd like to vote to close the auction now. 1) its not an auction, 2) we need to verify of those first 50 who is actually going to pay, 3) we have plenty lined up for unofficial and unknown preorders, of which plenty more will not pay. We risk turning this into a farce if we're not careful.
Pretty simple to me. Bid and don't pay, be dealt with like any other auction bidder that doesn't pay. I believe that earns you a scammer tag if I'm not mistaken. Either more than 50 units will be available, or not. You're bidding on your place in a (possible) line, and just like any other auction you have no idea if you won until the auction closes. Very damned simple - don't bid unless you are prepared to take delivery. Why is this confusing? Because there is no way to verify who is following the rules until after everything has closed and things break. |
|
|
|
|
I'd like to vote to close the auction now. 1) its not an auction, 2) we need to verify of those first 50 who is actually going to pay, 3) we have plenty lined up for unofficial and unknown preorders, of which plenty more will not pay. We risk turning this into a farce if we're not careful.
|
|
|
|
|
The problem with this now is I'm pretty sure a good chunk of these bids can't be backed up.
|
|
|
|
|
I've asked John or another mod to pay a visit. If nothing happens in the next day or two I will remake, seems people are way way too schizophrenic.
|
|
|
|
|
Show Posts
