philipma1957
Legendary
Offline
Activity: 4214
Merit: 8343
'The right to privacy matters'
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 10, 2015, 08:32:16 PM |
|
At 0.4W/GH board-level you could be seeing closer to 15GH per chip and 110W per board. 8.2GH per chip you should expect about 0.3W/GH and under 50W per board. That's at what I hope to be a low estimate for regulator efficiency. If you ran them up to 15GH per chip you'd see 272GH per board, so almost 2.2TH and 900W from an 8-board waterblocked system. Which if those waterblocks can handle 4x S5 boards (~1200W) they can surely handle a 900W load.
Top end for the chips would be about the same as an S5, estimated 22GH (400MHz) at hopefully 10.5W board-level per chip. The overall system efficiency for top clock would actually be between 5% and 10% worse than an S5 (hopefully closer to 5%), but the overall benefit of having adjustable voltage gives you access to a lot lower efficiency ranges than the S5 is capable of doing - same as with undervolting an S1 to get from 2W/GH down to 0.8-1W/GH board-level - which greatly increases the service life of the machine without requiring hardware modification to reach the low end.
This all really looks excellent. It makes me wonder why Bitmain skipped the option of adjustable voltage in the original S5 design. They would have the same insight into the chip that sidehack does. Maybe they just assumed that a massively adjustable power supply would materialize (i.e the whole 9v thing that evaporated). yeah but a company like enermax make atx machines that allow for voltage tweaks http://www.newegg.com/Product/Product.aspx?Item=N82E16817194117&cm_re=enermax_fanless_psu-_-17-194-117-_-Productthis can go pretty low in volts and still be efficient .
|
|
|
|
aarons6
Legendary
Offline
Activity: 1736
Merit: 1006
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 10, 2015, 08:46:02 PM |
|
I'm still not sure what the holdup is on the S5 working below about 11V, which I think folks have gotten them down to 10.8V but that was pretty much the bottom end. I'll have to spend some more time with them to figure it out, but one of the things I suspect is the stringed clock line so I came up with a better way of doing it that should remove any inter-node voltage issues. It could also have to do with how their node-level LDOs are powered, which is another thing I've taken into consideration on our design.
i think the people that tested the s5 at low voltage couldnt get the unit to power up under 10v. maybe its because the controller board is powered from the main boards.. i wonder if you powered the controller board separate you might get 10v or maybe even 9v.
|
|
|
|
MCHouston
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 10, 2015, 08:47:58 PM |
|
At 0.4W/GH board-level you could be seeing closer to 15GH per chip and 110W per board. 8.2GH per chip you should expect about 0.3W/GH and under 50W per board. That's at what I hope to be a low estimate for regulator efficiency. If you ran them up to 15GH per chip you'd see 272GH per board, so almost 2.2TH and 900W from an 8-board waterblocked system. Which if those waterblocks can handle 4x S5 boards (~1200W) they can surely handle a 900W load.
Top end for the chips would be about the same as an S5, estimated 22GH (400MHz) at hopefully 10.5W board-level per chip. The overall system efficiency for top clock would actually be between 5% and 10% worse than an S5 (hopefully closer to 5%), but the overall benefit of having adjustable voltage gives you access to a lot lower efficiency ranges than the S5 is capable of doing - same as with undervolting an S1 to get from 2W/GH down to 0.8-1W/GH board-level - which greatly increases the service life of the machine without requiring hardware modification to reach the low end.
This all really looks excellent. It makes me wonder why Bitmain skipped the option of adjustable voltage in the original S5 design. They would have the same insight into the chip that sidehack does. Maybe they just assumed that a massively adjustable power supply would materialize (i.e the whole 9v thing that evaporated). yeah but a company like enermax make atx machines that allow for voltage tweaks http://www.newegg.com/Product/Product.aspx?Item=N82E16817194117&cm_re=enermax_fanless_psu-_-17-194-117-_-Productthis can go pretty low in volts and still be efficient . There are also breakout boards that have voltage adjustments on them for HP PSUs.
|
BTC 13WWomzkAoUsXtxANN9f1zRzKusgFWpngJ LTC LKXYdqRzRC8WciNDtiRwCeb8tZtioZA2Ks DOGE DMsTJidwkkv2nL7KwwkBbVPfjt3MhS4TZ9
|
|
|
sidehack (OP)
Legendary
Offline
Activity: 3360
Merit: 1859
Curmudgeonly hardware guy
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 10, 2015, 08:54:07 PM |
|
The controller board has I believe a 5V buck on it that's probably good for down to 6-7V input without any real problems.
|
|
|
|
toptek
Legendary
Offline
Activity: 1274
Merit: 1000
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 10, 2015, 08:56:09 PM Last edit: June 10, 2015, 09:06:34 PM by toptek |
|
I don't have a torn-down S3 to verify, but the S5 board and S1 board have the same holes. Actually the S5 has one fewer hole (top center) but it's present on the heatsink. As far as I can tell, the heatsinks from an S1 and the S5 I'm looking at are identical.
i have a few s1's 3's and 5's. i can tear em down tonight and look at them. edit: actually they are the same because there was an s1 to s3 upgrade kit. They are i upgraded three S1 and still have one i use . and according to this these S3 water blocks will work with the S5 http://shop.syscooling.com/goods.php?id=47
|
|
|
|
aarons6
Legendary
Offline
Activity: 1736
Merit: 1006
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 10, 2015, 08:58:40 PM |
|
The controller board has I believe a 5V buck on it that's probably good for down to 6-7V input without any real problems.
hmm, i read the posts about the guy that tried to lower the voltage and he was saying at lower volts it wouldnt boot up, but if he booted it up with normal volts he could lower the volts while it was running down to 10v and be ok. that led me to think it was something to do with the controller board not getting enough power from the main.
|
|
|
|
sidehack (OP)
Legendary
Offline
Activity: 3360
Merit: 1859
Curmudgeonly hardware guy
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 10, 2015, 09:30:45 PM |
|
Could be, if it in fact didn't light up the controller. I can probably hook up a controller to my bench supply and see where it stops taking in juice.
Yep, the controller (unplugged from hash boards) still booted with 7V input. Not sure if it's trying to pull something from the board that isn't getting read properly at a lower voltage or not.
|
|
|
|
Xian01
Legendary
Offline
Activity: 1652
Merit: 1067
Christian Antkow
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 11, 2015, 12:59:16 AM |
|
... almost 2.2TH and 900W from an 8-board waterblocked system.
This is quite relevant to my interests. How can we make this happen ? ![Smiley](https://bitcointalk.org/Smileys/default/smiley.gif)
|
|
|
|
sidehack (OP)
Legendary
Offline
Activity: 3360
Merit: 1859
Curmudgeonly hardware guy
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 11, 2015, 02:07:41 AM |
|
Well, the first step will be a successful test of my 18-chip board. The next step will be a successful test of the regulator circuitry. The next step will be programming a microcontroller for voltage adjustment, fan control and temperature sensing and writing driver software to make it all work right. Then integrating all of that hardware onto a finalized 18-chip board, prototyping and testing. And probably by that point it'll be time to redesign it for a different chip because we're into the next generation already. And then test that new design. And then start gathering money to get a big fat batch of chips and a couple thousand PCBs. After that, it's pretty much coasting.
|
|
|
|
Digitalmocking
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 11, 2015, 02:13:33 AM |
|
This is pretty interesting stuff, I too would love to get in on this.
|
|
|
|
PlanetCrypto
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 11, 2015, 03:14:58 AM |
|
In our experience, the heat absorption capability of these waterblocks is dependent on flow rate and coolant composition. We like and use a 50/50 mix of water and automotive green antifreeze (Walmart brand). But pure (de-ionized) water has better thermal capabilities. I might add that all the tubing we use is of the oxygen barrier type (metal, PEX, etc. . .).
50/50 in your miner. Go pure de-ionized water with a corrosion inhibitor. I use a product (super coolant don't remeber the name) 1/20 mix to prevent corrosion. We never string more than 2 waterblocks in series, as the coolant temps can become excessive feed to the last waterblock in the chain. This is due to the flow capability of a 3/8" pipe at less than 5psi differential pressure.
I run 2 C1 in series (4 waterblocks - 8 boards). Each waterblock is about 3C hotter than the previous one. So if the first waterblock is at 30C the fourth will run at 39C. - Our system capacity is larger than a few ounces, so the cost of plant fluids and chemicals is a consideration
- Our heat dump loops are 100' of 3/4 buried PEX, it gets to -30F here in the winter
- automotive antifreeze has corrosion inhibitors and a bezillion other useful additives
- automotive antifreeze helps lubricate the circulators
- a colored fluid makes hunting down leaks easier
- a scented fluid makes hunting down leaks easier
We like closer temperature tolerances from machine to machine and set alarms to notify us when things start to get outta' hand. Am familiar with the advantages of de-ionized water w/plant chemicals from my days in the nuclear power generation. Hope that clarifies things a bit and why I made the statements I did.
|
|
|
|
PlanetCrypto
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 11, 2015, 03:26:31 AM |
|
Well, the first step will be a successful test of my 18-chip board. The next step will be a successful test of the regulator circuitry. The next step will be programming a microcontroller for voltage adjustment, fan control and temperature sensing and writing driver software to make it all work right. Then integrating all of that hardware onto a finalized 18-chip board, prototyping and testing. And probably by that point it'll be time to redesign it for a different chip because we're into the next generation already. And then test that new design. And then start gathering money to get a big fat batch of chips and a couple thousand PCBs. After that, it's pretty much coasting.
LOL!!! Ahhhh, The Readers Digest condensed version of board design.
|
|
|
|
sidehack (OP)
Legendary
Offline
Activity: 3360
Merit: 1859
Curmudgeonly hardware guy
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 11, 2015, 03:32:21 AM |
|
Yeah maybe one of these days some of these projects can be team efforts. Between Novak and I, we've basically got three or four active design projects going, and all the manufacturing tasks, and all the hosting and maintenance and packing and inventory and consulting. So things don't always get done as quickly as folks might like.
|
|
|
|
PlanetCrypto
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 11, 2015, 03:49:52 AM |
|
Yeah maybe one of these days some of these projects can be team efforts. Between Novak and I, we've basically got three or four active design projects going, and all the manufacturing tasks, and all the hosting and maintenance and packing and inventory and consulting. So things don't always get done as quickly as folks might like.
... and posting in the thread and answering PM's and rarely having a life and eating a burger occasionally and . . . What you get accomplished in as short a time as you do. The 2 of you guys are amazing.
|
|
|
|
sidehack (OP)
Legendary
Offline
Activity: 3360
Merit: 1859
Curmudgeonly hardware guy
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 11, 2015, 03:57:35 AM |
|
I've actually been slacking lately. Novak's putting in some long hours, but today we cut out early to play baseball instead of build stuff. As for rarely having a life, that's been my general state for years now and I'm pretty used to it.
Don't compliment how much I get done in a short time in a thread for a project that's over a month behind expectations. I was hoping to have a functional TypeZero board by now, and I don't even have Compac final-version PCBs in hand yet. Though really if I had passed on doing a stickminer and went straight into dev for the TypeZero it probably would be done already; I lost a buttload of time making those sticks work. Never would have guessed the regulator would have been that much problem, but to solve it I ended up finding a pretty sexy buck driver that I'll be using in a couple other things - including a new project we may be talking about next week if I can get enough materials together to get a plan ironed out, codenamed "aardwolf" for no real reason other than it's a pretty cool name.
|
|
|
|
Mikestang
Legendary
Offline
Activity: 1274
Merit: 1000
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 11, 2015, 03:58:32 AM |
|
I've actually been slacking lately.
Well get back to work, damn it! ![Grin](https://bitcointalk.org/Smileys/default/grin.gif)
|
|
|
|
Digitalmocking
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 11, 2015, 04:13:48 AM |
|
You're in Portland OR right sidehack?
I'm local to the area and would be willing to help, I'm in a weird position to have a lot of free time for the next few months.
|
|
|
|
aarons6
Legendary
Offline
Activity: 1736
Merit: 1006
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 11, 2015, 04:40:42 AM |
|
i live in washington state.. about 3 hours from oregon.
|
|
|
|
philipma1957
Legendary
Offline
Activity: 4214
Merit: 8343
'The right to privacy matters'
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 11, 2015, 05:12:41 AM |
|
You're in Portland OR right sidehack?
I'm local to the area and would be willing to help, I'm in a weird position to have a lot of free time for the next few months.
I think Missouri Rolla is the town….. But my memory is meh.
|
|
|
|
alh
Legendary
Offline
Activity: 1845
Merit: 1052
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
June 11, 2015, 05:50:20 AM |
|
You're in Portland OR right sidehack?
I'm local to the area and would be willing to help, I'm in a weird position to have a lot of free time for the next few months.
I think Missouri Rolla is the town….. But my memory is meh. While my memory may be equally bad, I too think Rolla Missouri. When I attended college back in the 70's, it was University of Missouri, Rolla (one campus of the University of Missouri system at the time). The acronym was UMR. It's since renamed itself to something more glitzy. At the time it was only about 4000-5000 students, though it graduated about 800 engineers a year.
|
|
|
|
|