The sticks are tested for speed before shipping out, but one test run may not always be indicative of constant performance and environmental variables do change things. If a stick's running slowly, adjust the voltage up a little bit by turning the pot screw in the bottom corner clockwise just a little bit. These sticks are built to overclock; they're built to be tweaked.

Last time I helped out a mine guy, we set him up with an older but still quite solid Dell server loaded with PFSense tied into a 24-port Cisco managed switch with a subnet per port, so he could map IPs to physical locations really easily. We put some multi-head gigabit NICs in for future expansions, but that initial configuration could have fed over five thousand miners and only cost me around five hundred in materials. I based the config off my own hosting setup, which peaked around 70 miners on about 18 subnets.

From me to the resellers, or from the resellers to you?

In any case - maybe!

Production has started.

I know those chip positions are the case on T17 boards, but is it the same on T17+? That's a 44-chip board, not a 30-chip, so it's more likely chip9, chip12, chip33, chip36 if they kept the same distribution pattern as T17 and T17e boards.

From what I know of other *17 boards, there are probably four chips with temperature sensors attached, two on the air-inlet side and two on the outlet side. They'll be in pairs, opposite ends of the same row, second or third row down from the top and up from the bottom.

I'm not sure how to tell from looking without lifting heatsinks because the parts are probably hidden up close to the chip. I could probably tell you from a UART sniff of the return traffic from the hashboard.

Make sure it was the heatsink that fell off the chip, not the chip that fell off the board. Is the chip surface undamaged?

There's a S17/T17 hack and repair thread here with a pretty decent writeup on how to re-solder the heatsinks but it'll require a bit of equipment to do it right. Bitmain used a low-temperature-process solder which is good for cooling and repair work but bad for operating especially in T-series miners which tend to push high power levels and therefore high temperatures, close enough to melting that the solder gets pretty weak.

Most thermal pastes are not designed as adhesives, so if they "stick" two things together in a permanent bond it's probably because something very bad happened. If a product is designed to glue heatsinks on, it'll say so up-front and words like "grease" and "paste" will probably not be in the title.

A Compac F with multiple BM1397 would not be that great, considering USB isn't good for more than about 15 watts and one BM1397 will handle that on its own. When doing a multi-chip setup, one also has to consider that the surface of the IC is copper-plated and electrically grounded, so you can't share heatsinks across multiple chips unless they're at the same ground potential. BM1391 doesn't have that limitation, but as you say, they're also not as fast. BM1387 can clock down to around 60W/TH (but very low speeds, ~50GH per chip at best), 1391 around 45W/TH, 1397 around 35W/TH.

Don't tell anybody, but I'm working on BM1391 next. Supposed to have some M30 chips inbound to play with too but so far I have no info on how to use them. I (GekkoScience) tend to not publish a lot of info either because I got it under NDA or because the manufacture and sale of my miners is the primary income for multiple households and I don't want to risk undercutting my own business or the employees that rely on it.

Another reason to make y'all wait a minute for hardware is driver verification. Kano made some fairly hefty changes so we need to test backward-compatibility with old hardware before calling it ready. Wouldn't want you guys to update cgminer and find out it only works on the Compac F now would we? That's a little too Apple for my comfort.

We are working on a larger-scale miner (finally!) but it won't be 3KW and it won't be Bitmain chips. Not gonna say much more until we have some real stuff to show off, as is my habit.

Everyone's "end of September" delivery target is based on that we're right now gathering up everything we need to start production, and then we start manufacture and finish assembly and testing before we ship, so nobody will have stock for several weeks. That's pretty much how all my new product announcements have gone - the first batch is a preorder with specs based on working prototypes, in order to raise capital to manufacture in bulk.

No, this stick can't be pushed to 600-700GH. There's only the one chip on it, and you're still limited by power and cooling. Software limits you to 538GH, if the hardware doesn't catch fire first. Fire voids the warranty, by the way.

Oh yeah um, the official announcement with pictures and real-world performance stats.

https://bitcointalk.org/index.php?topic=5355470.0

EDIT: Kano's setup instructions can be found HERE: https://bitcointalk.org/index.php?topic=5355470.msg58200639#msg58200639

---

Hey guys.

So some of you already noticed because of reseller presale announcements, but we're dropping a new stick built around the S17 BM1397 chip.

We call it the Compac F. It is red. Hot-rod red.



It is also made for overclocking. Hot-rod overclocking.



The two photos above were taken from the same test. I haven't stressed my tester sticks past 3 Amps from the USB bus, but if you are able to keep everything cool, in theory, the hardware should be good for more like 500GH. Which is a little absurd. Kano is working on the driver for this guy, and we agreed to cap the speed to 800MHz (538GH theoretical) for safety and sanity's sake. Because of a new IC package design by Bitmain, the chips cool more efficiently, which when coupled with our proven robust Compac power systems makes this stick quite a machine.

This isn't going to be marketed as a baseline miner like the Compac, 2Pac, NewPac. It's physically compatible with our previous USB sticks and built off the same proven platform, but this one is a limited release with only one planned batch. If enough people want it, we may make more.

Resellers are all the usual suspects. 419Mining, ASICPuppy and BitcoinMerch are stocking in the US. Bitshopper.de is selling in Germany, and Eyeboot.com will have stock in Hong Kong. MineFarmBuy is also interested but are a broker, not a reseller, so you'll have to talk to them to figure it out.

So we're still ironing out some kinks, but I got two sticks on my testbench right now, each putting up 330GH at 13.5 watts from the USB bus.

You know, in case y'all wanted a little more to get excited about.

When I did direct sales, pre-orders were common because I needed to raise money for production. The only difference here is I told my resellers we had a new thing before I announced it on the forum.

But yes, GekkoScience is dropping a new stick. We're still tuning the driver so a full top-end test hasn't been done yet but the testing specs are already a significant improvement on the NewPac.

And this one is Red.

Hey, I told you guys we were working on exciting stuff.

I don't have any solid news right now, but we're working on some stuff that should be pretty exciting here shortly. The rest of this year, if plans stay on track, is going to be a lot of fun.

S9 boards regulate internally. It's very unlikely input voltage is a factor. But given you're asking a machine that's already not the best design for cooling to run a 15% overclock it's far more likely a heat issue. The ASICs will actually draw more power as they get hot, which makes them hotter, which draws more power, and because of the string topology they all run in a sort of easily-perturbed power balance, so when overclocking, temperatures become a big factor for stability. A 10C increase in input air could make a big difference.

Did you tune the voltage? 130 watts for only 1TH across 10 sticks isn't that great, can probably do better.

Also are you sure it's a NewPac and not a 2Pac?

Well I've never tried, but it's entirely possible the driver is auto-installing for both devices. There might be a way to force it to recognize them as distinct but having no MacOS machines or Futurebit hardware to play with, I can't really say anything for sure.

Anyone else messed around with this?