jtoomim - Most of what you just said is stuff I've either looked at, am currently designing on or am currently prototyping.
|
|
|
Yeah I was right there in the first 20-odd pages of the S5 thread questioning the particulars of the design and reliability, how they handled certain likely issues and stuff. Prismas made a mediocre showing for string design, but the BE200 wasn't built from the ground up to be used in that topology like the BM1384 was (or the BE300 would be).
We've needed an economical home miner for two years. There's never really been much in that market sector that really suited every need (dB volume, price, efficiency, etc) exceptin' maybe AM Tubes, the S1, S3 and a couple others. Not enough, anyway. SP20 was really good gear, especially when it stopped costing $1000, so it's sad SPTech probably won't be doing any more machines like that but honestly it's never been their priority.
|
|
|
What I mean is, his revolutionary design was a string miner. I was being sarcastic about it being revolutionary, since folks had made BitFury string miners a year before. Strings are good because of fewer parts count and increased electrical efficiency, but the lack of voltage regulation and adjustability sucks. So there's still room for improvement.
|
|
|
Pretty sure Phasebird is an official AM spokesperson on the forum?
|
|
|
Not yet, they haven't even shipped chips yet. I did finally have some progress on the other thing I've got for you, and will continue to refine it today. Hopefully by next week I'm able to send a prototype.
It'll be a couple weeks before I'd have a miner, probably more like a month. I'm leaning on Novak for the firmware, but we both just got big jobs came in over the last week so we're both busy in different areas. I hope I don't have to do all the coding myself, which is possible but he's the better programmer.
Did you try to contact that legkodumov fella who designed the string board Bitmain used (they mentioned it themselves on some obscure board here). EDIT: found it https://bitcointalk.org/index.php?topic=728068.msg10435170#msg10435170I think that this guy was saying that he can give the design to anyone with chips (maybe for large volume, though, but it wouldn't hurt to ask). No I didn't. You may notice from reading that thread that I was cynical about his design, given that when he posted his revolutionary idea there had been several other string-based miners already come to market. Between Novak and I we have enough knowhow on the hardware and software to do what we want to do. Bitmain hasn't shipped chips yet because I forgot to send them payment yesterday. Regarding PSUs, if he's running 1400W from a PSU and then adding one fourth of an SP20, he's looking at adding probably about another 250W so that puts the supply at 83% rated capacity (or about 69% what it's actually capable of). Might be alright.
|
|
|
I bought a PSU-less stock machine for the museum, and it looks like the guy that was giving me one is actually giving me two (one of which may not work?) in exchange for some service so I might have a spare to jack with and see how far it'll push. Externally powered for an S6 would be nice, as long as the backplane traces are good for high current.
|
|
|
That article doesn't do a very good job explaining the upgrade kit. It states that the upgraded S2 will be able to hash at the same speed and power consumption as the S5.... which would mean the upgrade kit would offer a whole 150GH/s over the S2. Also, the S2 has slots for 10 boards - which are originally filled with what are essentially underclocked S1 boards. If the article is to be believed, then what are the other 8 slots going to be used for, if anything? I'll wait until something official from Bitmain actually appears instead of treating the linked article as any kind of source of truth, because that article smacks of just as much speculation as this thread ![Smiley](https://bitcointalk.org/Smileys/default/smiley.gif) . The article smacks of someone who doesn't actually know hardware attempting to write a qualitative piece about some hardware and doing a bad job with the details. Bitmain's given more info, still not a lot but enough to refine some of the speculation. At least we have the hardware config, and can use that to estimate specs. It looks to be pretty decent, as long as it's also not stupid expensive.
|
|
|
ADDITIONALLY - I have modded D750 PSUs (4 16AWG 18" cables hard-soldered, and an on/off jumper) for $32 for a limited time. I'm offering this with in mind the people who got burned by Bitmain's $99 price hike on S5s. I know it's not a lot but every bit helps, and that's what we at Gekkoscience try to do is help.
I have a limited quantity at present so it might take a day or two to ship. I think I can still ship the same as 750W kits, 3 PSUs to a $15 flat-rate box inside the US.
|
|
|
Not yet, they haven't even shipped chips yet. I did finally have some progress on the other thing I've got for you, and will continue to refine it today. Hopefully by next week I'm able to send a prototype.
It'll be a couple weeks before I'd have a miner, probably more like a month. I'm leaning on Novak for the firmware, but we both just got big jobs came in over the last week so we're both busy in different areas. I hope I don't have to do all the coding myself, which is possible but he's the better programmer.
|
|
|
From a PSU standpoint, you might consider tying the current-share lines on those so they load-balance. This should keep the voltages in check and make common-railing multiple supplies less tricky. At least the pairs that power the same boards should be current-shared. But definitely defer to SP-Tech's advice on how to handle that.
|
|
|
Ah, guess my numbers were a bit off when I was looking over that config a couple days ago. Let's go back over that...
According to the chart in the first post of your S5 thread, 10GH per chip should be possible at 0.67V (12V/18 string) which should put up 360GH per ~100W board. Is that chart wrong, or am I overlooking something?
edit:
Assuming the W/GH is linear between 0.65V and 0.70V, at 0.667V you'd see right at 0.3W/GH so a 100W board with 36 chips at 0.667V would max at 333GH or 9.25GH per chip. This gets a 3.3TH miner at 10 boards 1000W but probably means underclocking to do it. If they're looking more like 12GH per chip at 6 boards, that'd be 430GH at 130W per board, 2.6TH at about 800W. That's a ballpark based on some assumptions. Bitmain, am I close?
|
|
|
Who knows. What they're doing could turn out to be pretty great. I just have a big loaf of cynicism and make toast every time someone mentions Internet of Things.
|
|
|
Dogie mentioned 3 PCBs when the assumption was they were the same configuration as the S5 boards (as stated in the previously linked article), which means a 2x15 string clocked high and drawing 250-300W per board. A 2x18 string clocked a bit lower will pull 100W per board, so a 10-board S2 is what you'll get, expecting around 3.6TH from the included 1KW PSU.
|
|
|
Posted this after pictures and additional details were added to Bitmain posting. Ah, so it is a 2x18 string and the news people that said it was the same configuration as an S5 (which is to say 2x15 string) don't actually know. A 2x18 string should work just fine. ...so an S2 kit would more likely be a 2x18 string clocked around 10GH/chip per blade at 3.6TH per machine, overclock potential about 3.9TH.
I like those numbers. Now I really want to play with one. Please may I have a set of blades to test the crap out of and post a review? Since the S2 controller is a BeagleBone at its core, I'm guessing the only requirement would be a firmware update?
|
|
|
Yeah I guess it could be worse. Certainly not the chipping I was expecting ( https://bitcointalk.org/index.php?topic=977292.msg10680225#msg10680225) especially comparing S1 and S2 chip densities, but the numbers are still approximately okay. If you want an independent third-party review (and by independent I mean I'll give an honest opinion even if it offends the manufacturer) it appears I have a windfall S2 chassis I wasn't expecting, and some 2KW PSUs and stout fans available, so I could build up a machine to test until it catches fire and see how good I think it is. Might even be able to voltmod and test lowerbound efficiencies and put up a forum review.
|
|
|
S2 Upgrade is certainly good news, but I still feel disappointed.
|
|
|
I'm working with a guy who'd been trying to get ahold of Bitmain for weeks about a bulk purchase of S5 without a single word in reply from them to either of us. Until I emailed them five minutes after seeing the S5 was back online at $320. When they got back to me 12 hours later (at $419) I was told there was no price break for up to 100 units. So basically if I wanted to lay down on a purchase like would have been very straightforward before the holiday, now we'd be giving them an extra $10K just for the opportunity. I'm sure they're not screwing their "If you need more than 500 S5 at once" customers over but I can't advise my guy to buy a huge lot from them if they're not willing to be reasonable about a test-the-waters batch.
|
|
|
I'd avoid anything built by Technobit, Rockerbox or not. Those guys sorta have a history of screwing customers.
I'm hoping more third parties will get back into the game of building miners off mainstream manufacturers' chips. I'm working on a home-friendly design with Bitmain chips, talked to Spondoolies about chip sourcing but their per-chip power requirements don't really fit with the topology I'm looking at. Maybe next time.
|
|
|
Yes, go right ahead. Also, Scrappy, I've been meaning to formally apologize for telling you to calm down about and be patient about the Technobit minion boards you ordered. Turns out I was quite wrong, and it only looked like they were actually working hard to get things taken care of. I never saw a single board of the 50-odd I was supposed to get for hosting for multiple customers. So yeah, you go right ahead and sell off those PSUs since I know that's what they were for.
|
|
|
When have conditions been bad enough for Bitmain to sell at a loss? They're usually the ones working hard to undercut everyone else and drive their prices down. Bitmain has always been an instigator in race-to-the-bottom price wars.
Additionallly, if they have to sell at a loss, why would they sell at all? For one, it's stupid. For two, shelf it in one of their massive data centers and self-mine on it.
|
|
|
|