Initial testing on the Molex adapter board looks good. It's currently drawing about 18A off the 5V rail which is good since I intended to give it a 15A rating. A little airflow goes a long way too, just saying. Tomorrow I'll test how well it handles the heat of that much 5VDC conversion while drawing about 20A of 12V through the PCB.
I'm glad it's working well right off the bat. Maybe I'll put up pictures tomorrow.
|
|
|
Not really, no. Each board has 21 nodes of 3 chips. Comms go from chip to chip at the same node, then from the third chip of one node to the first chip of the node above in a continuous data path of all 63 chips. You take out one chip, you take out the whole board. On S1 and S3 you could take out one chip and all chips before it would keep hashing, but you know if you interrupt the current in a string topology the whole thing shuts down.
|
|
|
Not sure what you mean. The whole card is an individual string, and each card is individually addressed at the control board.
|
|
|
Other reasons include cooling both sides without having to mill around components, and cooling both sides without risking shorting across the different node-level ground potentials.
Both the dead boards in my care have string power.
|
|
|
I was getting a bit discouraged after a few weeks with no communication but I got a phone call this morning that things are kinda still on track. I'm still shooting for an S1 refit. Let's see if it'll happen before the end of the year.
|
|
|
I'm still hoping to get my hands on some of those sweet new chips. Guy I was working with is like 5 weeks behind estimated initial schedule so that's irritating, but we'll see what happens.
|
|
|
I'd say that anything with a breakeven point two generations in the future even with free electric is a terrible investment, quiet or not.
|
|
|
If that's their focus, they're shooting for a pretty limited market - and it's further limited to people within that group who are approximately idiots given the pricing and return estimates.
Also if it's the same power base as S9, which appears to be similar to S7, it's probably undervoltable but not without some hackery.
|
|
|
I need to check on stock, but I can put together some $70 server PSU (DPS800) kits to power these. Won't be whisper quiet of course but it'd work fine on 120V. Looks like I might end up with one or two hosted so it'll be fun to actually see one in action soon.
|
|
|
A bit, yeah. PM me with details.
|
|
|
The boards are a different size and have no mounting holes. You'd have to remove the chipside heatsinks without damaging the chips, and probably do some milling on the waterblocks to provide space for any chipside components with clearance problems, and then figure out how to mount the boards without shorting out or damaging anything on the PCB backside. Overall, it's not impossible but prohibitively difficult and generally a bad idea.
|
|
|
It has no problem cooling because your 3KW of PSU are burning about 300W of heat in themselves (assuming ~90% efficiency) so the power density is a fair bit less. The heat-generating parts in a PSU are probably also either solid metal or a big beefy silicon rated for 150C max internal temperature, rather than high-speed data paths on the most ornately etched crystal in the world, making the hardware a bit less finicky overall.
But yeah, one of the S9 boards that failed within 2 days in my hosting I don't think ever saw over 90F ambient temps, and were probably between 75F and 85F for most of that time. It also started to hash but died out quickly when the frequency was turned below 300MHz. Kinda strange.
|
|
|
Prototype PCBs for the Molex adapter have finally arrived. I'll be assembling and testing in the next few days and hopefully they work well and reliably.
I have already tested the new Dell 750W PSU board and it's in need of slight revision (including adding PWM fan control as suggested in this thread; it's been tested but not yet integrated), so I'll probably have another prototype PCB order going out soon. I intend to have a test version ATX24 adapter on that order - nothing fancy, directly powered with unswitched 12V EPS rail and probably two jacks for Molex cabling (compatible with the Molex adapter board, gotta love interchangeable parts).
I'm also playing around with a 1500W PSU that works with my edge-connector board (compatible with DPS800 and HP Common Slot family supplies), which might be interesting to some of y'all with six ~200W cards - depending on how all the jacks line out.
MarkAz is kinda the partner in crime for the cabling and adapter solutions so he'll probably have first access for actual in-rig testing. But that's okay because he likes posting pictures a lot more than I do and he's better at it anyway.
|
|
|
I'm living on less than one S9 worth of money in a given month, and it's only that much because I have an old credit card and student loans to pay down. Actual living expenses (rent, utilities, groceries) I could probably get by on about half an S9 per month, and that's right here in the US. Depending how and where you live, it's not even hard. I had a neighbor for a while with a wife and baby doing fine on a household income of under $300 per week.
The last miner I bought new was an S3, near the end of their run when prices were like $280. S9 prices would have to be about half where they are now for me to think about buying, and I've probably got cheaper power than most of y'all. And even then I probably wouldn't buy one because Bitmain hasn't done much in a really long time to demonstrate a desire to not actively screw people over.
|
|
|
Looks like a DPS800, which is rated for 800W at 120V. That's output power, so if Bitmain's estimate of 850W wall is accurate it should be fine.
|
|
|
And if it's pulling about the same power per board as the stock S9 boards (which it probably will given the same chip count and hashrate and efficiency and whatnot) but they like claiming warranty void if you only use 2 jacks on the S9 (pretty sure they did for S7, is that spec on S9 also?) then I would expect 3 jacks per hashboard as well. I mean, it'd be consistent anyway.
|
|
|
Last I heard CK's opinion on it, BFG started as a fork of cgminer and stole a lot from it. Maybe that's changed, I dunno, but he wasn't happy.
Kano has said a lot about bmminer, and some folks have run string tests on the compiled program and found its contents were nearly identical to cgminer. Given Bitmain's penchant for doing as little work as possible to produce a functioning product I'd be surprised if they wrote their own thing from the ground up, especially since it was built for basically the same hardware they've already had a crappy fork of cgminer on for about two years.
|
|
|
Just gotta hope the guys at corporate don't look too closely at bmminer and realize it's a potential liability being a blatant closed-source fork of a widely-known open source project.
|
|
|
Sweet, a 2600W PSU with 14 jacks. I sure hope they're 18AWG so people use a buttload of splitters to run more than one miner and end up melting down the wire. For less than that $300 I could get you two 1500W PSUs with 12 16AWG cables each. I mean 2600W in a single PSU is pretty great but I still expect to be disappointed.
|
|
|
Sweet, hadn't seen that photo yet. I was just going off logic. Looks like the boards are laid out pretty much exactly how I figured they were.
|
|
|
|