I've got a couple S9s here at the shop that have been sidelined. Might be worth looking at trying to fold a simple housing that can hold one board and two fans, maybe with the controller mounted on the back. I've got junk/spares of everything; will play around with that at the shop later.
Adding wedges internally to make sure your air stays in the heatsinks, especially at the far end, will be helpful. Don't forget that the middle position already has trouble cooling because the fan's motor reduces where air naturally gets moved. You might try a side position where the chipside heatsinks (responsible for the majority of the heat) get right in the main airstream, and then do a single baffle across the middle. You could fetch a small roll of metal flashing and cut and bend the right shape in a few minutes, then have enough material off the one roll to do a host of miners.
|
|
|
It'd take some custom enclosure but cooling would probably be more efficient if you had two fans side by side blowing directly into the heatsinks from the side of the board instead of from the end.
I wonder how well it'd work to use pull fans right against the heatsinks. Might work better for being able to aim your heat. 300W isn't too hard to clear either, considering my R606 can clear over 100W with a single 80mm 12v/160mA fan.
With some cleverness it might be possible to do away with the stock controller also, for just a single board. Hacking the voltage would require a bit of doing but a dummy USB2 adapter using the same connections as my Newpac/R606 broken out to the 18-pin cable and VH's cgminer driver, with a sufficiently strong controller, should be able to get ~2.7TH out of a single connection.
In any case, S9 quiet space heater is a great idea.
|
|
|
The title says NewPac, but everything below it references 2Pacs and BM1384 ASICs and all the performance data is 2Pac info.
Check the first post of this thread for NewPac-specific data.
|
|
|
Because you're looking at 2Pac tables. If the NewPac were capable of being set at 1.26V it would pretty much immediately burst into flames.
|
|
|
Yes, and yours is going out this week. Assuming a lost hardware order ever gets delivered.
|
|
|
Seems really unlikely. I've run 100KW of miners on under 500kbit bandwidth, which would take five months to download 800GB. The connection is shared with test stations that run up to 15x R606 for 24 hours at a time - maxing out that shared 3Mbit connection would take two weeks of 100% bidirectional use to move 800GB.
There's probably something you're missing, like a few orders of magnitude or someone stealing your wifi.
|
|
|
You need dead fans? I got dead fans. Only takes a minute to bust the motors out of 'em.
|
|
|
Going to a short wide board on the S15 was definitely an improvement. Anyone that ever ran Spondoolies rack gear understands exactly why. With long boards, the cooling air is already so hot by the time it reaches the last row of chips that it has very little effect. Short boards with short air paths don't heat the air as much so it cools the whole board better.
If you pull the boards out of an S9, you'll see the chipside heatsinks are missing on a wedge of chips at the fan end of the board. I assume that's to help open a void for air to come in from the periphery and cool the chips behind the motor. Not as effective as just, as Fuzzy recommended, adding a bit of actual gap behind the motor, but that's extra steps and a nickel off the profit margin. Who cares if it increases reliability; planned obsolescence is the name of the game so they have a vested interest in building gear that breaks.
|
|
|
The middle chips of the first few columns of the center board get almost zero airflow because they're right in front of the fan motor. Temp sensors are probably wired in at the other end of the board and don't reflect this.
|
|
|
I was just about to say something snarky about there being a table in the NewPac Support thread ( https://bitcointalk.org/index.php?topic=5053833.0) but it only goes up to about 300MHz so that's not directly helpful. However, since most sticks will run up to 400-500MHz on the stock voltage setting, you can probably extrapolate fairly accurately. At 300MHz it's listed as about 6.8W so 400MHz would be about 9W and 500MHz at 11.3W For a typical stick, 15W should be enough to get past 600MHz though you might have to up the Vcore a little for stability. The actual power draw will depend a lot on the efficacy of your cooling, as demonstrated by that the passive-cooled power draw at 125MHz is larger than the cooled draw at 200MHz. These sticks like to be kept cold.
|
|
|
Each USB connection is a separate device.
The pool sees a single miner because there's one instance of cgminer with a single connection to the pool as a single worker. If you used multiple instances pointed to the same worker it'd be the same. If you use different pool workers, one for each instance of cgminer, you can get per-connection stats.
The R606 is capable of higher speeds than six NewPacs even on a GS hub. The Vcore regulator on the R606 is good for 20A while the NewPac's main regulator can run 17A (at a bit lower efficiency), and NewPacs shouldn't be run at more than 3A input current, meaning more power is available to the chips in the R606 so the theoretical peak hashrate is higher.
VH can answer the question about mixing on hubs better, but overall it shouldn't have much effect. The main limits are power and USB throughput. Both sticks can draw the same power so that's not an issue. NewPacs run a higher hashrate so USB throughput is increased, but hubs use a bit more efficient packet structure so it shouldn't be possible to hit the throughput cap with only stickminers on a single hub, especially if you're using an ASICBoost-capable pool because that reduces USB traffic to AB-enabled devices by about 70%. As long as you keep everything cool there shouldn't be a problem.
|
|
|
Tell you what, boss. PM me your info and I'll hook you up with a NewPac so's you don't have to worry about missing out on the fun with that ditch gear you ended up with.
|
|
|
Hope you got good dust filtration or it'd get gross real fast.
|
|
|
More like there isn't much I can do to stop them, and I figure people will realize they threw their money on a crap product before too long anyway. At least those guys did their own board design, not like another 2Pac ripoff I saw for sale elsewhere and got the vendor to pull the listing.
|
|
|
(and Amazon systematically screws them over in every way possible but is too big and rich to effectively call them to task on any of their pretty horrible practices; it doesn't help that stuff like "undercut everyone forever in order to push them out of business and steal their markets" is literally the site's founding principles - which includes finding products that sell well from individual sellers, then buying out their suppliers and selling it directly, forcing those individuals out of business. And let's not forget withholding payouts to sellers, sometimes indefinitely, which means they get to draw interest on literally billions of dollars of other people's money)
But yeah, the guy selling on Amazon is one of the official sellers.
|
|
|
Is that the guy in the yellow case with a tiny built-in fan and BM1384 chips?
If so, I can't really recommend using it. They basically stole my 2Pac design and botched it into a fire hazard. The main regulator circuit is technically adjustable but there's no safe range on it which means you can overvolt the chips to like 2V quite easily. This is especially troublesome because they put diodes across the node-level capacitors to help balance on startup so if you give it more than ~1.4V those diodes start to conduct and then start to smoke. I think the chips are counterfeit (at least on the ones I looked at, the logos were lasered off) and not very well adhered to the heatsink. Both the ones I looked at had USB configs copied from one of my 2Pacs, including the same serial number on both sticks (all of my devices have unique serials).
But hey they got a nifty little built-in fan, so there's that.
|
|
|
For 2Pac advice, try the 2Pac support thread instead. Read the first post; it's got answers to about 90% of questions like this.
|
|
|
That conversation is probably more appropriate for a different thread. This'n is specific to stickminers and hubs that are good for them.
|
|
|
Glad it's going. There's been all sorts of issues with Win10 doing crazy crap and jacking up the USB throughput.
Here's what those fields mean:
0: the device number in the list GSI 10040069: device ID (type) and serial number BM1387:12+ type of chip (BM1387), number of chips detected (12) and ASICBoost is working on your pool (+) 550/550/547 Current speed (550MHz) / target speed (550MHz) / peak effective speed so far (547MHz)
|
|
|
Stock speed is 750GH; that's guaranteed from the factory. USB throughput limitations (because these chips don't buffer work very efficiently) start causing fits above about 800GH, but ASICBoost operates a bit differently (each work unit is a bit larger data packet but gets crunched four times in the chip, so traffic is reduced overall) and means you can get more hashrate out of the connection; the practical max (which is dependent on the machine and in no way guaranteed) is around 1.1TH and 950-1000GH is probably pretty stable based on user anecdotes.
We've pretty well caught up from last week's production disaster so everyone expecting a restock will be getting it this week. Good news is, while the chips we bought were lemons for the pod, they're working beautifully on NewPacs so we'll be full up on those soon too.
|
|
|
|