Bitcoin Forum

Just received my SP20 a couple of days after the review campaign, delivered from Israel directly to Sweden without any custom charge!

Packaging is compact and sturdy, very well protected

http://photo.mystisland.org/sp20/sp20_1.jpg



 

Front end
http://photo.mystisland.org/sp20/sp20_2.jpg

Back end
http://photo.mystisland.org/sp20/sp20_3.jpg

* john cries from the amount of nice photos everyone else seems to be able to get*

PS: Oops, tell me if you want me to delete this post as I might be blocking your review chain.

The whole unit weights exactly 5.0 kg, very easy to carry around

http://photo.mystisland.org/sp20/sp20_4.jpg

The camera was used to capture pretty girls, now girls are gone and only miners left  :D

The connectors, one serial for testing purpose I guess, USB, SD, LAN, all described in SP 20 quick start manual (It seems I could not find the user manual from Spondoolies website)

http://photo.mystisland.org/sp20/sp20_5.jpg

I have only two spare 500W PSU left from old time BFL singles bundle (They were so generous at that time that they even delivered a PSU with their unit, I never used them ;D)

http://photo.mystisland.org/sp20/sp20_6.jpg


I hook them up, started miner and went to the Settings page, lowered all the voltage to minimum (0.58) and filled in 200 in all 4 field: "Max PSU Power Unit 1 (70W - 288W) ", and restarted miner. It turns out the miner only runs at 370GH and by looking at ASIC stats page, two of the ASIC loops are not working:

http://photo.mystisland.org/sp20/sp20_asic1.png

Naturally, I thought the PSU is not enough strong, so I took another 1300W PSU and tested, same fault

Then I saw on "Settings" page, I could select "Basic Voltage settings" and then pick the "slow fans, medium rate" option, restarted miner. Immediately it hashes at 1.4 GH!

So I think there is a bug in that advanced setting page, it does not recognize the PSU even I put 120 in all of the fields, and by using basic voltage settings, all those fields turned to 1100

Then I reduced all the voltage to 0.6 and the unit run at 940GH/S, ASIC stats looks like this

http://photo.mystisland.org/sp20/sp20_asic2.png

And the amazing thing with SP20:

each of the two PSU draws only 220W on wall! With only 440W I can get 940GH, e.g. 0.47J/GH on Wall. A 20nm very efficient Knc Neptune running 3.3TH at 1900W on wall, that is barely 0.58J/GH. And SP20 at 1.4GH draws around 800W, on par with Neptune, it is definitely a nice choice given the current market price of $659 per unit, very well engineered product!

http://photo.mystisland.org/sp20/sp20_7.jpg

That is the lowest I have seen.  Wow that is a impressive under clock.

those are the lowest numbers yet.  damn they are good.

please show a screen shot of the settings page.

Now after a restart it turns to 900GH, and ASIC stats shows lower frequency at 595Hz. Maybe due to rise in room temperature, anyway, settings are all set to 0.6

http://photo.mystisland.org/sp20/sp20_settings.png

And the only problem for home mining is its noise, just like a vacuum cleaner running, it is impossible to run at home. With even one door closed I can hear it clearly far away. So either you run it in a specialized garage/mining farm, or you have to take some modification for the cooling system. I will take it apart tomorrow to see what I can do with it

I suppose that the original fan deliver a huge amount of airflow like 200 CFM maximum, but when heavily under clocked, maybe a general low CFM/noise fan can deal with the heat

I'm personally thinking of buying a better fan and swapping it in, wonder if that'll make a huge difference?

Currently I can put it close to window, where 5c cold air are sucked into the room, maybe even a GT1850 (which is almost silent) can deal with the heat. But who knows when summer comes

Just checked the fan, it is more than 2A, definitely over 200 CFM, this is standard for server application, but not suitable at home

Thank you for the review.

Just upgraded the firmware to the latest version 2.5.33,  now I can drop the fan speed to 20 percent, still plenty of airflow (more than 100 CFM) but noise has come down quite a bit. I think if it is a pwm fan, even 10% could work, since now the back end temp is mere 46c degree, and I have seen some miners have a backend temp as high as 70c degree

http://photo.mystisland.org/sp20/sp20_temp.png

Thank you too for letting us review such a great product!

Before I started to operate SP20, I thought this kind of products are just like many other miners on the market: You stack them up in a mining farm and pay lots of electricity and cooling cost, and sell the coins to cover these cost, and bring down the bitcoin exchange rate during the process. I even planned to give it to friends who are interested in bitcoin after this review

But after seeing the stunning efficiency of this miner, I'm very impressed. I had a feeling that the old good mining time is back: With this unit running at 500W, the electricity cost is neglectable, so anyone can just hold on to all the mined coins without selling a bit

Now I'm very interested to run it for an extensive period of time with a modified cooling solution

Just heats up quicker.

The heat sink on SP20 is just entry level plain aluminum heat sink, they don't have huge surface area like those heat pipe heat sink with many thin aluminum fins, so the airflow must be very high to reduce the surface temperature

I just measured 20% fan speed noise at 1 meter away with my sound meter app, it is still 67db, and I personal feel is close to in car noise on highway (70+-db)

so far no  one has found a better fan solution.  I do think if you fully tear the machine down it could be cooled quieter.  

  Right now I have six.
  I wish I ordered more.  
  I would try lower downclock then my 1350gh and 750 watts.  
  your review means my 6 x 750 watt = 4.5kwatts and 6 x 1350 = 8100gh  could read
  6 x 440 watts = 2.6 kwatts and 6 x 940 = 5640gh

I can do more then the 6.  I will look forward to any idea you have to manage  noise.

I think the only way to cool it down better without lowering ambient is to take it out of the case and strap fans or point fans at the boards individually.

There is however one problem handling the boards. The heatsinks are only secured in two diagonally opposite locations, and it only takes a small bit of pressure to move the heatsink off the die which breaks the TIM seal between the die and the heatsink, it also may risk crushing or damaging the die, although that hasn't happened to me yet. But I think once you disturb the TIM seal it can affect temps.

Open the case
http://photo.mystisland.org/sp20/sp20_8.jpg

Controller board
http://photo.mystisland.org/sp20/sp20_9.jpg

It uses a channel design to cool both the heat sink and the VRM on the back of the board, so an open case cooling solution might not be as optimal

http://photo.mystisland.org/sp20/sp20_10.jpg

Left is a replacement fan I planned to put in. Unfortunately they don't have the same socket type and even the pin out is different. If I want to install other PWM fans, I would have to take the original plug and wire, solder on a PWM socket to make an adaptor

http://photo.mystisland.org/sp20/sp20_11.jpg

4 ASIC chips per board

http://photo.mystisland.org/sp20/sp20_12.jpg

The heat sink is hold tight not only by 2 screws on each corner, but also two drop of glue on each side. The glue is very hard, which makes the removal and refitting of the heat sink more difficult: you have to cut the glue first , and later when you install it back, you need some similar glue that can tolerate high temperature of the heat sink, a normal heat glue gun will not work

http://photo.mystisland.org/sp20/sp20_13.jpg

The heat sink uses a copper plate mounted together with aluminum heat sink that you could see on CPUs 10-20 years ago, they are joined together with some white thermal grease (which is not quality thermal compound). And the large contact surface of the copper plate does not really help since the contact surface with ASIC is small. You can see there are some thermal grease on the ASIC chip capacitors, and they are still wet as I touch them with a needle

If you re-condition the heat sink components and replace the white thermal grease with MX4 or even liquid metal on ASIC surfaces, that could bring down the temp 5-10 degrees. But due to the weak heat sink, it still can't carry out the heat enough fast unless you have huge amount of airflow passing through

It is a pitty that a modern 28nm processor only equipped with a heat sink that is decades ago, without heatpipe, without many thin finns. This in turn caused the annoying noise issue

http://photo.mystisland.org/sp20/sp20_14.jpg

Back of the ASIC board, clean layout

http://photo.mystisland.org/sp20/sp20_15.jpg

I only replaced the metal screws with rubber pins, it does not really make a lot of difference, noise at 1 meter (20% fan) dropped from 67 db to 66db. To replace with my own fan, an adapter is needed, do it later

http://photo.mystisland.org/sp20/sp20_16.jpg

To test different fans this is what I did:

http://s25.postimg.org/meu1nugsv/20141114_150328.jpg

Im curious, does the underside of the board (bottom of  picture) get warm at all? Looking at the design, I would think most air pressure will escape there,  following the path of least resistance, and relatively little air will flow particularly over the top heatsinks. Id try channeling the air by blocking most or all of the clearance to the underside and using a fan with perhaps lower CFM but high static pressure.

Everyone that done review have hashrate dip to 0 in 1day hashspeed chart. In your case was the same? This units resets itself in 1 day periods?

That's a nice idea!  ;D You can even install a fan speed regulator outside the case

Have you tried GT1850 with the SP20 set to 0.6v?

Exactly, the construction reminds me of Avalon gen 1, you can narrow down the other channel to increase the air flow, but I think the effect will be minimal, since that original fan is a 2A high static pressure fan. The whole case is very cool, there is no direct contact between the heat sinks and the case

In my experience, the optimal output of 120x38mm fan is around 120CFM, where noises are still tolerable. Above that, you will dramatically increase the noise without too much improvement of airflow, a 200 CFM fan would sound 4x more noisy than a 120 CFM fan

Looks like a lot of cost cutting has been done to maximize profit.

I wish. Not true.

Yes, they just don't push enough air.

I've tried the GT1850 and up to the 4250

So why use crappy thermal paste and old inefficient heat sinks?

Not exactly .. I can see this a lot in most industrial setups, where you do not need all the bells and whistles for the consumer market. This thing is supposed to run somewhere outside of the living spaces.

They market it as a 1.7 thash home miner not industrial.

Nice pictures , Noice is the only reason i dont buy sp20 , if someone successfully hacked the fan noice please do share

The "home" doesn't necessarily mean the bedroom or den.

Both "facts" are untrue. Sure more expensive cooling stuff available. The solution we choose is adequate. It's certainly not cheap or crappy.
You can pour liquid nitrogen on this ASICs and probably get to over 2 THs, but the build cost will be over 0.6 $/GHs
Should we do it ?

Let him do it! BOC and AirLiquide can deliver Liquid Nitrogen to his house for cryogenic tests. The temperatures will be a little low though .. Canada is full of surprises during winter ;D

Announcing "Frosénn"-a new 2TH (and more!) miner from SPT  ;D

Agreed, the components are all of very high quality, very well designed and balanced, even the web page feels it is done by a professional design and integration team, just like branded: Carrier grade

Noise is not a big issue in mining farms. Especially when you are facing tough hashing war and your machine might be trashed after one year of operation

The size of Rockerbox chip is the same as GPU, and it is 28nm! Modern GPU typically employ a large heatpiped heat sink with lots of fins for silent operation, but that is aiming for at least 5 years life expectancy. When there are 4 such GPUs on one single board, no simple cooling solution except lots of air flow can dissipate those heat

It's more the layout that's the problem. The airflow passes over 1 chip and goes +40c, then it passes over the other chip and goes another +40c. The second chip is baked by the first chip.

Ok. I think the previous king of cfm/noise ratio delta AFB1212VHE can be applied for under clocked SP20, but the noise could not be much lower. I have a feeling that the noise is coming from the resonance of the box, the fan itself should not sound that much

I remember that the GPU mining efficient king HD 5970 had 2 GPUs, and the cooling element combined two copper core in one assembly

http://i2.rozetka.ua/goods/2159/arctic_cooling_accelero_xtreme_5970_2159447.jpg

Given the design of SP20's board, such design is not suitable, but at least a heatpipe element can be mounted on each of them, and I think the cost will also be low if mass produced

Maybe, those who prefer lower noise could order upgraded heat pipe heat sink to their SP20, and those heat sink can be reused on future spondoolies products (compatible mounting on future miners)

That heat pipe setup is nice. The regular 5970 passes the air across both GPUs:

http://images.bit-tech.net/content_images/2009/11/taking-apart-the-ati-radeon-hd-5970/12.jpg



A heat-piped SP20 would be great.

not worth it - as it is most users are underclocking anyways, so putting in $50 worth of hiher quality heatsinks serves minimal benefit.

Let's go with $50. I'm pretty sure a better cooler would get at least 100 GH/s better performance. That's $12/mo extra. You make up the difference in cost in 4 months. Better performance, better efficiency, lower noise, longer life. At $50 it's an easy win.

not really when you consider that pushing it for an extra 100GH will increase the overall power draw by 100W - you make the machine much louder and if unless you pay <$0.10/kwh its only amounts to a few extra dollars of profit. hence why a lot of people are running this 1.7TH/1.1kW device at 1.3TH/0.8kW

With better cooling you get better efficiency and lower noise. The problem with the SP20 is the unequal cooling of the chips. The second set of chips gets hit with air that's +40c hotter. Cool them with ambient air and you can lower their voltage and increase the efficiency.

The second set of chips get hit with warmer air, but not +40C hotter.

+1 better cooling would increase efficiency a bit, but not with much effect. A larger heatsink would allow slower-moving fans which would be nice, but IMO theres not much purpose in pushing them to the absolute limit unless you have cheap electricity, and even then the SP20 drives the chips quite high.

(8 chips in SP20 = 1.7TH) vs (30 chips in SP35 = 5.5 TH) If the SP3X series ran all 30 chips at the same power as the SP20 you would see it achieve around 6.3TH (similar to the original claims). However, that would use around 4.4kW at the wall. Instead, the SP35 does 5.5TH/3.6kW and the SP31 does 4.7TH/2.8kW. The hashrate gain is at about 1w/GH, which doesnt make sense for a lot of users right now

I can't speak for others, but the stats I posted in my review were accumulated while leasing my rig out because of the Paycoin frenzy.  Any significant drop in hashrate was likely the result of renters switching pools or accidentally entering invalid pool mining credentials. 

Ok, my intake is about 40c, and exhaust is about 80c, so the back chips are getting +20c air. It significant enough that they are downclocked 200 MHz as a result of the hot air from the front chips.

I have done two session (some modification and tweaking in between), the latest one is 20 hours but have not experienced any drop in performance

http://photo.mystisland.org/sp20/sp20_temp2.png

Every system has tradeoffs. In the SP3x series as well, the cooling solution isn't symmetric.

In case of the SP10, with the shared heatsink, the cooling was more symmetric.

The RockerBox ASICa with the FCBGA package aren't suitable for shared economocial heatsink. Mechanical stress, etc.

The system may be tweaked for better results. The entire software is open source as well.

Disclaimer: Modifying will cause a warranty lost.

A high-end solution would be vapor chamber based heat sink, but it is really a pain to recondition all of the 8 chips even if you have the new heat sink with the right size/mounting. I have not find a good way to remove those hardened glue without risking hurt the die under the heat sink. I will just leave the original heat sinks as they are

http://www.legitreviews.com/images/reviews/1461/gtx580_vapor_chamber.jpg

The difficulty trend is down, it is worth to do some tweaking when operating life expectancy is increasing

Maybe you can do some air channel to send some cold air into the back row of the heat sink

I just checked the ASIC stats page,  on my underclocked unit at 0.6V, the difference between different ASIC chips is maximum 10 degrees

Yeah, they're great when underclocked.

Success! Lowered noise to 60 db! Now it is not audible anymore when the door is closed. It is a comfortable level even when I'm sitting beside the rig, just feels like in a server hall, not industry warehouse with lots of machinery. The room's ambient noise is 33 db before running SP20

After 15 minutes of running, with a front temp of 25c degree, I get 60c degree for back temp, and none of the ASICs reached minimal  watching threshold of 85c degree

The test method learned from Biffa:
http://photo.mystisland.org/sp20/sp20_17.jpg

The fan speed regulator is running at 5V, but with such a large load, it gets very hot. I hang it up at the exhaust to cool it, but will try to make the adapter to utilize the PWM socket on board later


Hero of the day: Delta AFC1212DE.

This fan consumes 3A at its maximum output, so should be as powerful as the stock fan. However, it has 7 blades instead of 5 on the stock fan, so the noise performance is much better. In principle any 7-blades 120x38mm delta fan could dramatically reduce the noise

http://photo.mystisland.org/sp20/sp20_18.jpg

Record! Noise down to 48 db. Now even without door closed, I can barely hear it 10 meters away. Accomplished by previous low speed fan king GT1850

At ambient temperature of 14c, it can keep the back end temp at 60c. But now the difference between the lowest temp DC module and highest temp DC module is as high as 30 degree (33/66), the air flow is not fast enough to cool the second ASIC module in the channel.

At 22c degree ambient, back end reached 68c, now two of the ASICs reached 105c. It seems with this weaker fan, even at 0.6v you can not run at an ambient temp over 25c degree

So delta fan is still the better choice

Another issue: The metal ASIC board stopper under the fan is so high that it has a direct contact with the fan, that's the reason adding rubber pin for the fan did not remove the vibration noise from the case. I removed that stopper totally and the vibration noise is gone (When fully assembled, ASIC boards are fixed by other screws, even without this stopper they won't move)

http://photo.mystisland.org/sp20/sp20_19.jpg

Noted! ;D

Very nice drop in decibels!

I just tested with stopper moved out of the box, even without the vibration noise, stock fan at 20% speed still generate 68 db, maybe because of the lowest operating voltage for stock fan is 7 volt. But its cooling effect is strong, even at 27c at frond end, the back end temp never goes above 50c

And with GT1850, after a while I opened the case, some part of the case is extremely hot, especially the last ASIC heatsink, definitely over 100c, dangerous

why is your intake temperature so high? thats the problem. Ideal intake is about 15-30C and ideal outlet is 65-75C

I agree with you, after 3 days of testing, I found out that those 4 ASICs in the back row are much more stressed due to the uneven cooling design of the case, and since the front row of heatsinks totally block the air passage to the back row, there is no easy way to get cold air into the back row without passing through the front row heatsinks, especially the second heatsink near the bottom of the case, since the heat air from the first heatsink has nowhere to go but pass the second heatsink

Some kind of cold air intake duct can be constructed, but that will be very time consuming, maybe only suitable for one or two units

or set watts lower for them and set starting volts lower for them.

 I am running 5 stacked and 1 on a shelf.


I had  2 ways of heat+noise management.  heavy down clock to 1150-1200gh fans at 20-30%  since they are in an attached garage noise is not a big deal and I can even close the back door of the garage

   or a higher clock at 1300-1350 fans at  50-60% and  stagger the settings.  I have to open the back  door  of the garage for at least 3 hours each day.  and they are a little too loud.   

bottom line 7200gh vs 8100gh.   3900 watts  vs 4775   .54 watts a gh vs  .59 watts a gh 

 and a happy wife  which is very valuable in life indeed.

I have decided to go with the lower clockings for now.

Made an aluminum block to narrow down the lower channel

http://photo.mystisland.org/sp20/sp20_20.jpg

http://photo.mystisland.org/sp20/sp20_21.jpg

Unfortunately the result is worse, temps rose by a couple of degree under same ambient temp, maybe some turbulence made fan spin slower

excellent review and some great tinkering. well done on getting such low w/gh

I added some aluminum guides in the heat sink so that the warm air from the first heatsink does not go straight onto the second, but the effect is minimum

Then I noticed that there are 2 large holes before the last 2 ASICs, perfectly suitable for a test of cold air intake modification. I used some plastic foil to wrap the rest of the case, leave these 2 holes open widely by flipping the controller board to the other side

http://photo.mystisland.org/sp20/sp20_22.jpg

This immediately lowered the temp of the hottest ASICs. As a result, the hash rate jumped: The system's frequency management look at the highest temp ASIC to adjust the frequency of the other chips. Now the back row ASICs can be cooled more, the rest of the chips were tuned up accordingly. And you can see now all the temps are getting close (chip nr.4 is possibly faulty)

http://photo.mystisland.org/sp20/cold.png

It is also worth noting that the front temp on SP20's statistic page is not the ambient temp (I mistakenly thought so in previous posts), it is affected by the airflow speed. With original setup, the front temp reads 18c when ambient temp is 14c (measured by a thermometer in front of the main air intake of the SP20). And in my modified setup, the from temp reads 25c

nice catch !

Aluminum sheet guide installed on the heatsink. Since the system adjust the frequency depends on ASIC's temp, which is not visible in the ASIC stats page (only visible once it is above 85 degree), I don't know if this modification works. However, without this modification, the system runs at 5% lower frequency, so I suppose it has some effect

http://photo.mystisland.org/sp20/sp20_23.jpg

When I install those guides, I found out that one of the heatsink does not have those two drops of glue on its side, thus it can be easily tilt or shake by hands. I removed it, cleaned thermal grease, it turns out all of the chip's corner are broken. Without those 2 drops of glue, I can imagine it get knocked by each vibration during transportation

http://photo.mystisland.org/sp20/sp20_24.jpg

This is a weakness of the cooling system. Like Spondoolies-Tech pointed out, The RockerBox ASICs with the FCBGA package aren't suitable for mechanical stress, since its die directly exposed, and it is small: only 10x11.5mm in size.

The problem is, the heat sink is just too big for the small contact surface of the die, and it has only 2 point of tension. To make things worse, its tension is flexible because of the spring in each screw, this means it will just tilt to any direction and break the corner of the die with the least amount of incaution during install/transportation. So that's why they end up using some glue to permanently fix the heatsink on the board

And, to my surprise, the copper plate is just attached together with the aluminum heatsink with thermal compound without any screw.
They can slide against each other. This means most of the heat will be transferred to the heatsink by this layer of thermal compound, which has magnitudes lower thermal conductivity than either aluminum or copper. When installed, the pressure between them is decided by the spring of those 2 screws, and it is very weak (otherwise it will break the die)

http://photo.mystisland.org/sp20/sp20_25.jpg

Sexy!

The chipped corners reminds me of amd xp, durons... nice times!

It would be nice to have a miner with lga775(or newer) socket mounting holes! then i can be motivated to repair my cascade system for -80c cooling :D

Does broken corner affect performance? My broken chip runs at 500 Hz comparing with other chips running at 650Hz

if it can not go up to speed as the rest, then i guess 99% is because of the chipped corners >:(

The pictures you've posted shows a residue of the glue. Maybe not enough was applied to the specific ASIC.

Very unlikely.

Only residue, no sign of glue like other chips, no sign of glue on heatsink either  ::)

What kind of glue you are using? I have to get those glue if I'm going to recondition the heatsink, now I'm putting 2 heat pads on each corner to stabilize it

3M's DP460

Edit: The glue has no purpose other then holding the heatsink during shipment.

like what KfC didn't do, you mean?

 :D

Mine had none on any on the first SP20 I recieved, I commented on the instability earlier in the thread.

https://bitcointalk.org/index.php?topic=896161.msg9865070#msg9865070

Nice work johnyj,

but instead of putting that aluminum block inside the case, what about simply putting a stripe of tape on the holes on the intake side of the case to block air from entering that space?

I'm away from my unit or I'd made this test right now :)

spiccioli

careful with that aluminium foil! short-circuit...

I think some air is needed to cool the VRMs on the back of the center ASIC board

1st off: Why all these reviews spamming the hardware section?!

edit: OK, i see why now, Spondoolies-Tech doing Spondoolie things... making some smile, and some so bitter they never got one (people even calling it a scam, wow!)

2nd, nice review! now i want one.. time to sell my POS car

3rd:

the worst it would do is destroy a few calculation units, what it would most likely do is make the chip miscalculate a nonce, and give you ether hardware errors or just plain rejections. it wouldn’t slow it down in any way, just error out like mad.

thinking back to my AthlonsXP days too, i had one with a large chip that took out half of the L2 cache, it would run without an issue, but once i pushed it with data processing (moving files or calculation prime) it just die. i used it for net surfing, but i avoided large pages, it tend to kill it too.

i could go on about these cpus, like the old pencil trick, how i let out its magical blue smoke, etc, but that's for the offtopic sections.

The efficiency curve:
http://photo.mystisland.org/sp20/efficiency.png

The lowest setting 0.58V is very unstable, usually half of the loops are disabled. At 0.59V I can achieve 809 GH with barely 376W at wall (0.465W/GH), and at 0.61V, I get 1TH with 500W, also feels great. The efficiency drops almost linearly across the voltage range, I did not test above 0.68V, which consumes around 950W

Nice. What's the PSU efficiency you're using at 500W-1KW levels ?

2X EVGA 500B http://www.evga.com/Products/Product.aspx?pn=100-B1-0500-KR (http://www.evga.com/Products/Product.aspx?pn=100-B1-0500-KR)

What are the dimensions of the red outline area?

http://i61.tinypic.com/2qxm4co.jpg

10cm X 9.8 cm

You could get a 90cm fan there.

But I tried push/pull with 120CFM fans and the performance, albeit quieter was worse than the stock fan.

Depends what you want really, if you want high performance, you can't beat the stock fan but if you are willing to make do with 1.3-1.4TH then you can use a quieter fan.

I'd like to see a more 3/4pin stock fan header on the control board rather than the nano molex for us "home users" to experiment with.

That's what I thought at first (100mm). Now I'm thinking another 120mm fan mounted on 40mm or so spacers to allow for the connectors / cables. Might be the holes on the left just fit. On the right drill two holes. Then tape a shroud over everything. Maybe cut out some of that grille too (except the area that supports the PCB).

Maybe have the push fan a tad higher at say 150 CFM?

Noise is no issue. I have read a comment that the stock fan is not enough in hotter environments. Don't remember which review - there's so many...

The cooling is sufficient for the first two ASICs, but it gets worse and worse for each ASICs further back, so adding a fan from top of the box can help a lot, unfortunately that means you have to cut a hole on the top. The best solution would be getting individual intake channel for each ASIC, but almost impossible to do due to the compact design of the box

the curve was with what fan settings? 80-90 or lower?
fan settings would affect efficiency.
Obviously, you want fan settings to be as low as possible while maintaining working temp (<115C)

Just a reminder to everyone out there. Underclocking is not for everyone. If you are paying less than $0.18/Kw, then it will actually be profitable to run at stock speeds. Above the $0.18/Kw threshold you will actually start profiting (BTC-Electric) by running at underclocked speeds.

I pay $0.098/Kw, so I'm looking for the highest GH/s I can ran at while still keeping the fan speed at 40. I've been playing with all different voltages and this is what I've come up with. Note: I have this SP20 running along side an Antminer S4 and I have a 24" fan blowing air into the intake of both miners.

At the current rate of BTC income per month $0.014/TH/day I should make 0.61BTC/30days. At $330/BTC, I should bring in $201, spend $64 on electric and profit $136 every 30 days. Note that as Bitcoin difficulty increases the amount of BTC earned per month will decrease and there is always the price of BTC to factor in which changes on a daily basis.

At the nano farm price (free shipping) $1900=3x$633 and $140 for a 1050W 80 Plus Gold PSU, I'd have $773 invested per SP20. @ $136 profit, the SP20 nano farm would pay for its self in just under 6 months.

1050W 80 PLUS Gold pulling 957W at the wall
Mining Rate: 1430.13Ghs
Temp Front / Back T,B   21 °C / 67,58 °C
Fan Speed    40
Start Voltage 0.71 / 0.69 / 0.71 / 0.71
Max Voltage  0.71
Max Watts    220 / 220 / 220 / 220


Asic Stats
Uptime:2044 | FPGA ver:100
-----BOARD-0-----
PSU[UNKNOWN]: 0->217w[217 217 216] (->217w[217 217 216]) (lim=220) 0c cooling:0/0x0
-----BOARD-1-----
PSU[UNKNOWN]: 0->216w[216 216 216] (->216w[216 216 216]) (lim=220) 0c cooling:0/0x0
-----BOARD-2-----
PSU[UNKNOWN]: 0->209w[209 209 209] (->209w[209 209 209]) (lim=220) 0c cooling:0/0x0
-----BOARD-3-----
PSU[UNKNOWN]: 0->215w[215 215 215] (->215w[215 215 215]) (lim=220) 0c cooling:0/0x0
LOOP[0] ON TO:0
 0: DC2DC/1/:[vlt1:705 vlt2:709(DCl:794 Tl:794 Ul:709) 86W 122A  52c] ASIC:[ 85c (125c) 935hz(BL: 935)   82 (E:193) F:0 L:0]
 1: DC2DC/1/:[vlt1:705 vlt2:709(DCl:794 Tl:794 Ul:709) 88W 124A  71c] ASIC:[ 85c (125c) 935hz(BL: 935)   64 (E:193) F:0 L:0]
LOOP[1] ON TO:0
 2: DC2DC/1/:[vlt1:691 vlt2:695(DCl:794 Tl:794 Ul:709) 89W 128A  80c] ASIC:[100c (125c) 965hz(BL: 965)   81 (E:193) F:0 L:0]
 3: DC2DC/1/:[vlt1:693 vlt2:698(DCl:794 Tl:794 Ul:709) 85W 123A  88c] ASIC:[ 85c (125c) 930hz(BL: 930)   89 (E:193) F:0 L:0]
LOOP[2] ON TO:0
 4: DC2DC/1/:[vlt1:705 vlt2:709(DCl:794 Tl:794 Ul:709) 85W 120A  48c] ASIC:[ 85c (125c) 915hz(BL: 915)   78 (E:193) F:0 L:0]
 5: DC2DC/1/:[vlt1:705 vlt2:709(DCl:794 Tl:794 Ul:709) 83W 117A  62c] ASIC:[ 85c (125c) 885hz(BL: 885)   85 (E:193) F:0 L:0]
LOOP[3] ON TO:0
 6: DC2DC/1/:[vlt1:708 vlt2:709(DCl:794 Tl:794 Ul:709) 88W 125A  68c] ASIC:[ 85c (125c) 925hz(BL: 925)   88 (E:193) F:0 L:0]
 7: DC2DC/1/:[vlt1:703 vlt2:709(DCl:794 Tl:794 Ul:709) 85W 120A  75c] ASIC:[ 85c (125c) 915hz(BL: 915)   75 (E:193) F:0 L:0]

[H:HW:1429Gh,W:692,L:0,A:8,MMtmp:0 TMP:(21)=>=>=>(67,59)]
Pushed 28 jobs , in HW queue 4 jobs (sw:1, hw:1)!
min:43 wins:642[this/last min:23/26] bist-fail:27, hw-err:0
leading-zeroes:42 idle promils[s/m]:0/0, rate:1817gh/s asic-count:280 (wins:11+12)
Fan:40, conseq:200
AC2DC BAD: 0 0
R/NR: 312/0
RTF asics: 0
FET: 0:9 1:9
 0 restarted      0 reset          0 reset2         0 fake_wins
 0 stuck_bist     0 low_power      0 stuck_pll      0 runtime_dsble
 0 purge_queue    0 read_timeouts  0 dc2dc_i2c       0 read_tmout2    0 read_crptn
 0 purge_queue3   0 bad_idle
 0 err_murata
Adapter queues: rsp=3, req=20

To compare to what I would make in a month underclocked down to 0.650V @ 635W @ 1126GH/s @ $0.098/Kw (0.56W/GH). I would make $149 worth of BTC, spend $39 on electric, and profit $110. Compare that more efficient setup (0.56W/GH), to a less efficient setup 957W/1429GH = 0.66W/GH and I profit $26/month more on the less efficient setup.

My fan is always set to 20, and the underclocked temp is also much lower. Underclock is to squeeze more hashing power into a limited electricity capacity, thus increase the amount of bitcoin harvested for a given power usage, not to make quick fiat money profit

Probably best to take them out of the box altogether and mount fans directly on top of the heatsinks. Sort of what Enterpoint did with the CM1.

Am I right in guessing that the PCB is approx. 120mmx240mm?

130x310, if you take it out, you should also cool the back side of the board, maybe a better way is to use two fans blow from both ends, and leave the cover open to let the heat out, but where to put the controller board could be tricky

Grill down!  ;D  I hung the cutted honey comb grill piece in front of the exhaust fan, and it is blown away by a quite large force, so remove it will improve the airflow for sure. And the hiss noise is gone

http://photo.mystisland.org/sp20/sp20_26.jpg

I can fit a fan wire grill to increase safety, but not a big deal, currently just leave it open (easy to clean the fan from dust ;D)

Thanks for the info.

I would think by leaving the cover open, you'd loose the channel effect to the inner two heat sinks. It's easier for the air to just rise in the big space than to go through the small space of the fins.

Do you think it's possible to turn one board around 180 degrees, so both heat sinks rows would face outward? Then I could blow from the side at each (after cutting holes) and the fan at the end would take care of the cooling of the back of the PCB.

Good effort on the grille! Get the other one too (except PCB support)  ;D

Made this adapter for using 3rd party PWM fans

http://photo.mystisland.org/sp20/sp20_27.jpg



However, the adapter might not be needed. After so many improvements made during latest several iterations, now a GT1850 can drive the SP20 at 0.6V without any dangerous temp, given 42 db of noise, which is as quiet as many home PC (My water cooled PC runs at 38db)

http://photo.mystisland.org/sp20/sp20_open_gt1850.png

Open middle hole, GT1850, 42db, 465W on wall,  975GH under ambient temp of 15c degree. Now it is really a green miner with a quiet fan  :D

I thought about adding two holes in the bottom of the case, just before the second bottom row of heatsinks, but that's too much work, and the increase in total intake surface will reduce the airflow speed for each intakes,  so I just use the original case design to cool the last set of ASICs, it is enough good in practice (Thanks god they have two holes originally designed to fit those cables, it suits the purpose well)

this is really interesting... does it improve the airflow at all? is there a way you can measure it?

There are reviews analyzed different types of grille and some of them said that a honey comb grill will give 10% less airflow. I can feel that honey comb grill obviously give resistance to air flow. While I don't feel any significant resistance if I hang a wire grill at the same place

I did think of that, it harks back to my watercooling days, but effect on resale value turned me off it.

I would not leave such a powerfull fan open, it can seriously injure a finger.

spiccioli

While it could hurt, it is an exhaust fan. Far less dangerous than an intake fan.

beat me to it.. the blades point inwards... i'd still make sure it was away from anywhere that the cat could get nosey about it.

Turns out I still have to pay 60 euro for the import tax and some extra charge to DHL  :-\

The PWM adapter did not work at first, since the wire on the stock fan socket is different: The power on that socket is negative 12V. In order to run a PWM fan, you have to switch the position of the power and ground line on the fan connector, which makes the connector wiring color looks different than normal PWM connectors. This first photo shows the difference

http://photo.mystisland.org/sp20/sp20_11.jpg

The Everflow fan on this photo has 9 blades, but it is actually noisier than delta fan due to it is only 28mm in depth. With a delta AFC1212DE installed, at 20% speed, the DC temp never goes above 60c degree, which is my comfortable zone. The noise goes up to 54db at one meter distance, still acceptable, and a great improvement over stock fan's 68 db at 20% speed

A new record: Ambient temp 5c degree, 0.6/0.605 setting, 897GH, consumes 345W at wall, means 0.385W/GH efficiency

At very low temp, the normal 0.59/0.6V setting can not start all the loops, must raise it to 0.6/0.605V to get all the loops running, but the result is quite respectable

http://photo.mystisland.org/sp20/sp20_lowest.png

Impressive. Got a screenshot of the ASIC Status page?

Indeed, impressive.

http://photo.mystisland.org/sp20/sp20_asic_low.png

But something is wrong here, with 345W, the webpage shows 897GH but mining pools shows constantly 700GH, I don't know if it is the latest firmware caused this problem, will restart it and try again

Edit: Something went wrong with previous settings, after several test, it seems 440W on wall for 930GH can be stable, still 0.47W/GH

0.47 J/GH is aligned with our undervolt results as well for 0.6V
There are differences between different batches of RockerBoxes, the last ones are more efficient.

Another observation: Too cold temperature will reduce the performance over time

For past several days, my outdoor temp dropped below -10c, and this affected the miner near the window, ASIC frequency keeps dropping over several days (together with the power usage), now I put in a slower GT1850 fan, and partly close the window to keep the intake temp around 10c degree, it has been stable at 970GH since then

Too bad :( 0.385W/GH would have been really nice.

That will be the dream of course. The efficiency is directly correlated to voltage, however the stable setting is around 0.6V, below that you might experience stability issues

I'm using 80+ bronze PSUs, if you run a 90+ platinum PSU, the efficiency can increase a bit

sounds like the S5 has the SP20 beat for use at ~1250GH/630W or ~0.51w/GH

well, very close, but S5 has this crazy fan-my amazon fan supply for mod (got both Noctua and Silverstone) will be here only on Friday-I even had to get the prime trial which I never wanted in order to get it faster..

someone said they had free shipping? I even rmemeber them using this as one of there reasons to order spondoolies and not bitmain...???? ???

they are trying to charge me for shipping or am i missing something?

order on ebay with free shipping...

Order through RoadStress GB link.

not long til it's over!
get in quick!
https://bitcointalk.org/index.php?topic=903286.msg9924866#msg9924866

Still waiting.... :(  My first set is suppose to show up today.  Hope they give me enough time to test them and decide between the SP20 or S5.  The S5 units that I received run very stable at stock and rather impressive overclock. Sound is going to be the determining factor I think.

try running the stock fan at 5V - its actually a very manageable volume at that speed (~2200rpm approx) and is sufficienct to keep the unit around 60-65C at stock speeds. With some cold air at the intake I can keep under 62C at 375MHz

I have a feeling the S5 will be a winner for 0.5w/GH efficiency and low noise (with 5V adapter for the fan or revised PWM controller settings), but the SP20 has a lot more flexibility to stack and be driven down to better efficiencies without needing a custom <11V PSU to do so.

Another note: The voltage that supplied to the fan on SP20 control board is -11V, so the stock fan is already a bit slower than normal, but still as loud as vacuum cleaner at 20% speed, maybe 10% can work but I don't have time yet to do the ssh mod

What you guys think about this ones ???

http://www.arctic.ac/eu_en/products/cooling/case-fan.html

Now quiet and stable: With GT1850 and cold air intake mod, 43db at 1 meter, ambient temp around 15c degree, 480W on wall with 80+ bronze PSU, extremely stable after 1 week of continuous running.

http://photo.mystisland.org/sp20/sp20_stable.png

Very satisfied with this 1TH rig with such low noise level (you can not hear it in the next room with door opened)

Bought this Fractal Design Newton R3 1000W 80+ platinum value PSU, suits SP20 perfectly

http://photo.mystisland.org/sp20/sp20_28.jpg

The platinum PSU draws 450W for 1TH, for 0.5W/GH, it can deliver 1.2TH at 600W (need more powerful fan on SP20 to keep it cool). Cheap 80+ platinum PSU is the way to efficiency

These fans are too weak to drive the SP20, actually most of the 120mm fans on PC store are low noise low rpm fans. 2500+ RPM and 100+ CFM PC fans are rare, most possibly you will need a 120x38mm 3000+ RPM fans to get a decent cooling on SP20 (if not aggresively underclocked) , but then noise will become a problem

Nice. Cost?

~230chf here in CH. https://www.digitec.ch/en/s1/product/fractal-newton-r3-1000w-power-supply-computer-631365?tagIds=76-524

This PSU is cheap because its fan is noisy (no PWM control, it does not run under light load, but spins up to top speed when the board is overheat). But for mining its noise is neglectable, so a perfect choice, you can find it for 200$ on Amazon

I'm running 2 underclocked SP20's on a Rosewill LIGHTNING 1300W 80 PLUS GOLD and I'm very impressed with the build quality. The modular cables screw on and there are a total of 8 6-pin PCI-Express connectors. I got mine at Newegg for $179.99, free shipping and a $20 rebate card. I can attest that this PSU really can run at 92% efficiency. If I go off what the SP20's say they are pulling and the power at the wall its running just shy of 93% at about an 85% load.

For those that are wondering I have a single 6-pin cable each on the top connections of each board and both lower connections share a double PCI express cable. So 6 cables total (2 doubles).

Here are some specs of the PSU
100-240V
80 PLUS Gold Certified
Modular Design
Efficiency up to 92%
140mm Silent LED Fan

Here are some stats of my 2 SP20s vs an Antminer S4.
WattsAtWall   GH/s    W/GH
1450             2000    0.725
1130             2159    0.523

Ok, so i decided to underclock some more. It appears that my newest SP20 underclocks and performs much better than my 1 month old SP20. I've now got my newest SP20 running a 0.597V and it shows a total of 353W on the stats, hashing at 819GH/s. My older SP20 is running at .610V (about as low as it will go) for 402W on the stats, hashing at 862GH/s. I've got both on the 1300W 80 Plus Gold Rosewill and its only pulling 781 Watts at the wall via a Kill-a-watt meter! That 1681GH/s @ 781 W for 0.464 W/GH. I'm not sure how accurate the stats page wattage is, but if its correct that means my PSU is running at 96.6% efficiency. 755W showing on stats combined and 781W at the wall.. 755/781=96.6%. Is this right?

New SP20 353W .597V 819GH ~0.44W/GH
Old SP20  402W .610V  862GH ~0.48W/GH

Combined @ Wall 781W 1681GH = 0.464W/GH

These 2 are running stacked @ 20C room temp.

That's some really impressive numbers!

I'm just curious what others think of this possible setup.

30amp 240V PDU (7200W max capacity) ($150)
4 x 1600W Rosewill 80 Silver w/ 16 PCI-E connectors ($1060)
16 x SP20 underclocked at 0.6V pulling 350W each ($7500) (15xMini-Farm + 1) (~800GH/s per underclocked SP20)

Run 4 SP20 on 1 1600W PSU using 1400W.

4 PSUs sucking 1400W each consumes 5600W total which puts only 77% load on the breaker.

Primary question is do you think 4 underclocked SP20s (350W ea.) running off 16 x PCI-E connectors on 1 PSU is too much for a PSU?

This would make a 12.8TH farm for $8710 w/ 0.47W/GH efficiency w/ a cost of $680/TH.

The PSU is the Rosewill HERCULES-1600 1600W 80 PLUS SILVER http://amzn.com/B00A940V1K

Another possibility would be to use the EVGA SUPERNOVA 1600W Platinum and connect 3 - SP20 with a slight bump on the overclock up to 500W each.

30amp 240V PDU (7200W max capacity) ($150)
5 x EVGA Supernova 1600W Platinum w/ 14 PCI-E connectors ($1665)
15 x SP20 underclocked at 0.621V pulling 500W each ($7000) (15xMini-Farm) (~1TH per SP20 underclocked)

Run 3 SP20 on 1 1600W PSU using 1500W.

5 PSUs sucking 1500W each consumes 7500W total which puts 83% load on the breaker.

This would make a 15TH farm for $8815 running at 0.50W/GH efficiency for a total farm cost of $587/TH

EVGA Supernova 1600W @ $333
http://amzn.com/B00NJG61JQ

Running the SP20s that slow is an expensive option.

I too think that running SP20 at its lowest settings might not be an optimal choice

Although the efficiency is impressive at 0.6V or less, it also reduce the hardware investment return. This is especially true when difficulty is starting to rise again. You race against the diminishing return of the mining rigs. After one year of operation, there will be future miners with newer process (16nm FinFet like knc claimed?) coming up, then all it matters are how much coins were digged out by then

I hope that SP20 can run at least 6 months before the next major technology advance wave hit the scene

How about this for an optional fan setup for 4 SP20s. Use a 14" 12V Radiator fan that pulls 2000cfm and mount it to a custom box made to hold 4 SP20 miners? The fan pulls up to 120W depending on how much voltage you send to it. The box would be made to hold miners 2 on top and 2 on bottom and just remove the stock fans completely? It would certainly pull enough air. The box would be made to only allow airflow through the miners? Could also be an option for 4 antiminer s5's as well.

PROCOMP 14" INCH ELECTRIC COOLING AUTOMOTIVE RADIATOR FAN 12 VOLT ($30 including shipping)
http://amzn.com/B003U7UR3A

operating three SP31 would be much easier...  ;D

It would certainly be easier but they would use 9000W vs only 7500W.

I also see my math was off on that one as 7500W would be too much for a 240V 30amp breaker...

Unfortunately my apartment does not have a garage that can allow me to install extra circuit capacities, modern apartment are designed to be green with less electricity capacity (and also higher electricity cost because the electricity source is green). It costs me $0.2 per kwh, the only thing I can do is to squeeze enough hash power into 2 KW maximum load. I must be able to cook/wash cloth/dry hair etc without worrying about triggering the circuit breaker. And constant high load on wires could cause fire  :P

My two sp20 came in today (really, really fast shipment, thank you!) and running fine.

I ran it at default settings to get a feeling for it and with stock settings some ASICs reach 120c with fan at 90%. Ambient temperature is around 15c.
Trying to undervolt now to keep all ASICs at 115c max. Isn´t it possible to operate at around 1600 GH/s without overheating?

Are you wanting to control the voltage to adjust fan speed in real time? If not, I have a few of these and they work great.
http://www.lowes.com/pd_416724-11292-SFC1-500B_0__?productId=4755301&Ntt=high+velocity+fan&pl=1&currentURL=%3FNtt%3Dhigh%2Bvelocity%2Bfan&facetInfo=
They move a ton of air, have a (manual) speed control switch, and it would be a lot simpler to interface than trying to PWM a big 10A fan.

Just received a couple of Gentle Typhoon AP-31 (5400RPM) for testing, this is a high-air-flow fan for 120x25 size, it performs almost identical to Delta AFC1212DE at 20% speed

 

waiting for the results as i have a few AFC1212DE and i was thinking about replacing the stock fan with one of them

It seems that AFC1212DE have two version, one is 1.6A and one is 3A, I'm running the 3A version and it is 54dB at 20% speed with grille off

I'm waiting for some AFB1212SHE delivery, might be a little bit quieter. I think 120mm fan's limitation is around 100CFM (50CFM under load), anything beyond that will sounds like a vacuum cleaner thus not suitable for home running

The strange thing with SP20 is that when the temperature is low (DC module below 50 degree and ASIC runs low frequency), it tends to drop the frequency after a while, means it has an optimal operating temperature of at least 60 degrees, this seems to be true also for knc's miners, it actually works better when the ambient temperature is higher. This is against my previous knowledge of electronics  ???

Not an electronic expert here, but batteries work better when warm as well. I think it may be that the materials which are used to make electronics have their specific characteristics, and an "temperature window" where they work best. Above and below not so much.

Why many guys seek extremely low temp just to raise the overclock potential, using even liquid nitrogen?

I have tested several times, at 0.61V, with a high speed delta fan, the highest achievable frequency is 700hz. With a slower fan GT1850 , that speed will jump to 740HZ. Since the efficiency is only related to voltage, I would like to keep the voltage as low as possible while pushing the frequency as high as possible. It seems that I can only achieve that with a slow and quiet fan GT1850  ;D

In traditional overclock, at a given voltage, to achieve higher frequency, you need better cooling. In GPU mining, ambient temp can go as low as -20c without any drop in performance. But SP20 will just die if the ambient temp drops below 0c degree

How about this?

Say an electronic part performs at best when it's at 60 degrees. An overclocker without Nitrogen would otherwise run it at 100, but the Nitrogen brings it back to 60.

What you are trying is to run at 40 and expect better results. But the part is not in its preferred range, so the efficiency is less.

I've got the 4250rpm version on my SP20 for a few weeks now. Also tried the Scythe Kaze Ultra, which works quite well, but is also hard to find.

Also from the same range the scythe kaze jyuni 1900rpm I've also tried, at 110CFM

None let you run at the same temps as the stock fan, all are significantly quieter even at full speed. With a slight underclock they are fine.

For me noise is not an issue but heat,  so I would like to hear if someone came up with a better fan solution than stock fan to increase cfm. I was wondering if this delta model could work. PFB1212UHE-F00. Its rated 252 cfm vs stock 216 cfm.
http://www.amazon.com/Delta-Electronics-PFB1212UHE-F00-120x120x38mm-connector/dp/B004Y15ALK (http://www.amazon.com/Delta-Electronics-PFB1212UHE-F00-120x120x38mm-connector/dp/B004Y15ALK)

I got Scythe and would like to try. How did you connect it-to internal connector or to PSU through adapter? Thanks
i don't think that it has PWM anyway, so it does not matter, but Scythe is just 46dB (3000rpm, 133 CFM). Surely, it beats the SP20 stock fan by being quieter.
i am running SP20 at ~650W/1200GH anyway, so less sound would be great.

Since the stock fan uses a special connector, you have to run your fan with dedicated power supply outside of the box, thus can not utilize the auto-fan feature of the latest firmware. You either run it at full speed or use a voltage regulator to control the speed (regulator will overheat under high load, cool it with exhaust fan)

http://photo.mystisland.org/sp20/sp20_17.jpg

Thanks. I will try auto feature tomorrow am, not sure that it would be enough-so far Sp20 fan was persistently noisy (even at 20). If it goes down to 50ish in dB, then I don't have to bother. There is such thing as the physiological adaptation. A persistent sound at a reasonable dB will be ignored by the brain after a while. It is going up and down in dB and pitch that makes sound more noticeable.

Hey Guys, just found this fan from Enermax:
http://www.enermax.de/luefter/twisterstorm/

154cfm at 38db.

I´m planning to use the SP20 in my living room, so I want to replace the stock fan and underclock to get about 1.0 - 1.2 GH/s. Priority is low noise, not W/GH.
You think this could work oder do I need more cfm output?

I read to whole thread but still not sure if its worth a try  :-\

try new auto feature and underclock first, i suggest.

Impossible to get 154 cfm at 38db for a 120mm fan, the best 120x38mm fan with 130CFM is already at 70db range, those figures from manufacturer are all marketing tricks

For underclock use, a 60CFM GT1850 with noise around 45db at 1 meter distance is the sweet spot, these are true numbers

let me try a fan / Noise summary:

simple exchange is not possible as SP20 uses a very special and tiny PWM Connector.
This connector seems to be a 1.5mm JST ZH 1.5mm 4-pin
(http://www.ebay.com/itm/Micro-Mini-JST-Connector-1-5mm-4-pin-w-Wire-x-10-sets-/161350482183) not 100% sure but 99%

to use the new smarter fan control you either try to order theses connectors or disassemble the stock fans one and solder etc.

as Sp20 comes with a high power 120x38mm fan that is specified up to 200 cfm and this amout of air may be needed in the next summer, I think we need to differ the CFM classes and target GH/s.

-1 safe as stock up to 1700 GH

-2 safe for medium hash rates and conditions

-3 safe for green underclocking (below 1100 GH)

class3 will be an easy one, many low flow 120x25mm fans will suffice and be nicely quiet and can be driven from external PSU.

tested replacement fans class 1-2:
Delta ? 1212 (afaik well known server fan?) high CFM
Scythe Kaze Ultra (38mm depth, medium-high flow) PWM?
are both really quieter at same flow?

Did anyone tried to stack fans in the fan-space of the sp20? there is a lot of room for bigger ones?!

please feel free to add your experiences like the firmware wish list!!!

I don't think 1700GH is practical unless you have free electricity, you could just buy coins instead of exchange electricity, heat and noise for it  ;)

Delta fans are very good for server usage, previous king of high CFM fan
http://www.xtremesystems.org/forums/showthread.php?261778-120mm-Fan-Testing-on-an-MCR120-Radiator-Round-6
http://img641.imageshack.us/img641/2346/ra7deltavhevsall.png

I found out that many users have their DC temp more than 70c degree and still hashing happily, so I decided to raise the temp by partly close the window of the room. As a result, the maximum achievable hash rate increased at 0.61V

Now 485W on wall, 1051GH ------> 0.46W/GH  :D See how long it lasts

In general, I completely ignore any specs published by manufacturers of computer fans other than size and RPM.
90CFM@1400RPM with 18dBA of noise? (http://www.newegg.ca/Product/ProductList.aspx?Submit=ENE&IsNodeId=1&N=100007997%20600035655%20600035590%20600035661) Sure buddy, whatever you say. When Delta (or Nidec, Comair Rotron, etc) publish a real datasheet, I generally will go by the numbers presented without questioning them too much. Computer fans, not so much. I'd be extremely hesitant to replace a 150CFM server/industrial fan like the AFB1212SHE (3700RPM/53dBA) or the 200CFM one in the SP20 with one marketed towards desktop users based on specs alone.

which one of those SLOTS labeled as 1,2,3,4 are the slots that power the back chips ?
Thank you !

Agree 100% with MrTeal.

Scythe/Nidec seem to be pretty close to published specs, I've had some good luck with those.

However, and this is a bit caveat.

No "quiet" fan, even an undervolted Delta is going to allow you to run at max speed on a SP20

To get to 1.6+TH you need 180+CFM and air moving at that speed needs a certain type of fan and both the fan and the movement of air produce a shitload of noise.

The target for me has been much quieter than the stock fan @40-50% at circa 1.4TH which is perfectly doable with something like a 130CFM Scythe

johnyj: thanks for the link.

my last test was: with http://www.noctua.at/main.php?show=productview&products_id=80&lng=en
works with limited CFM okay up to 1100 GH/s and when PWM below 50% it is really much quieter and my PSU is getting louder. I speak of round about 44db(a) from 1-2m distance. 44db are completely killed by shutting the door of a room.

Judging by the ASIC stats page the bottom two sockets. But for 100% certainty you should ask Zvi in the main thread. Or just test yourself by unplugging them

yeah just plug in 3 of the sockets and see what loop does not fire up.

China airmail arrived.
I can confirm these connectors do fit.
you only need to cut away some tiny outer plastic "noses" with a kitchen knife, then it plugs perfectly into the sp20 controller board.
then solder some standard 4-pin PWM connector at the cable and you can check every quieter 120mm PWM fan you like without losing warranty.

i have lot similar whit this fan
http://partner.delta-corp.com/Products/FANUploads/Specification/EFB1512HHG(REV01).pdf

my is delta EFB1512he dc 9-12 v 2,43 a  279 cfm
at 7-8 v is totaly silent.posible mount outside of sp-20 only need one screw adapter
https://www.dropbox.com/s/rzzt1tiso7q2cng/IMG_20150208_142317.jpg?dl=0
https://www.dropbox.com/s/645zt3ymqhltoe2/2015-02-08%2014.23.18.jpg?dl=0

https://www.dropbox.com/s/6ny7qfng9tcy71d/2015-02-08%2014.23.36.jpg?dl=0

https://www.dropbox.com/s/2wigeplb0hpj4wu/2015-02-08%2014.26.31.jpg?dl=0
https://www.dropbox.com/s/021nny00ow7yk87/2015-02-08%2014.27.27.jpg?dl=0

also have https://www.dropbox.com/s/3x1ad7gwuzqdifm/hladnjak.jpg?dl=0 full coper heatsink 0,6 kg    7-8 $
or use this https://www.dropbox.com/s/adm3904jfbrpo6t/hladnjak%202.jpg?dl=0  low profile 0,65 kg     8-9$
please all sugestion is welcome.
now i dont have time because i work in attic on cleean room whit military nuclear grade filter whit isolation rockwoll 20-30 cm
http://www.ebay.com/itm/NEW-ASTROCEL-I-1000CFM-CLEAN-ROOM-HEPA-FILTER-24X24X12/360960905857?_trksid=p2047675.c100005.m1851&_trkparms=aid%3D222007%26algo%3DSIC.MBE%26ao%3D1%26asc%3D28808%26meid%3D1adc69b801a84b0a9d0b66c9a5cd9ba7%26pid%3D100005%26rk%3D5%26rkt%3D6%26sd%3D301492912443&rt=nc

also sugestion for hepa filter is welcome

Ok so i read the posts in this thread and maybe i missed it but i didnt see ne one posting about doind something as simple as flipping the fan 180 degrees. The temps reported on the main status page showed:

Before = 66/58

After flipping the fan to blow air instead of pull air = 42/41

Asic temps all show 85 when before the first half were 85 and the second half were anywhere from 105-115. After the swap the very last asic in the chain hovers between 85 and 95 but thats acceptable. The other asics do not go up or down so im not sure what their temps are. Im thinkin 85 is the lowest it will go because when the miner first starts ik its not 85 degrees but the stats page say 85 on all the chips so idk what the temps are but they r either exactly at 85 degrees or lower. Thats seems like a pretty damn good decrease in temps that involves removing 4 screws and putting them back in....keep it simple peoples lol.

Also my settings are- .62/.62./.62/.62
                                .67
                                 180/180/180/180
                                  1322gh/s after running for about an hour.

For pulling air through the miners u need a sealed tube because any non sealed areas make the airflow into the miner less powerful. After removing the lid to flip the fan u will see a big gap between the fan and the top plate that the controller card mounts to which makes a bunch of air run over the top of the controller card instead of where its needed which is the asics itself. Some might worry that makes the controller vulnerable to overheatting but after flipping the fan theres so much more preasure in the miner that air hits all over the damn place which cools everything a hell of alot better. I was gonna turn my fans up to help cool the chips better but the flip made such a huge diff i dont need to raise them above 40 not at my current voltages. Hope this helps.

Temp Front / Back T,B
27 °C / 78,74 °C
Fan Speed
40
Start Voltage
0.67 / 0.66 / 0.67 / 0.66
Max Voltage
0.71
Max Watts
288 / 288 / 288 / 288


 0: DC2DC/1/:[vlt1:724 vlt2:727(DCl:794 Tl:794 Ul:727) 102W 141A  67c] ASIC:[ 85c (125c) 1000hz(BL:1000) 14832 (E:193) F:0 L:0]
 1: DC2DC/1/:[vlt1:724 vlt2:727(DCl:794 Tl:794 Ul:727) 105W 144A  86c] ASIC:[ 85c (125c) 1020hz(BL:1020) 14906 (E:193) F:0 L:0]
LOOP[1] ON TO:0 (w:27661)
 2: DC2DC/1/:[vlt1:697 vlt2:701(DCl:794 Tl:701 Ul:727) 99W 141A 103c] ASIC:[120c (125c) 1010hz(BL:1010) 14426 (E:193) F:0 L:0]
 3: DC2DC/1/:[vlt1:677 vlt2:682(DCl:794 Tl:682 Ul:727) 91W 134A 103c] ASIC:[120c (125c) 950hz(BL: 950) 13235 (E:193) F:0 L:0]
LOOP[2] ON TO:0 (w:29846)
 4: DC2DC/1/:[vlt1:722 vlt2:727(DCl:794 Tl:794 Ul:727) 99W 136A  63c] ASIC:[ 85c (125c) 980hz(BL: 980) 14349 (E:192) F:0 L:0]
 5: DC2DC/1/:[vlt1:724 vlt2:727(DCl:794 Tl:794 Ul:727) 106W 146A  88c] ASIC:[100c (125c) 1040hz(BL:1040) 15497 (E:193) F:0 L:0]
LOOP[3] ON TO:0 (w:26576)
 6: DC2DC/1/:[vlt1:693 vlt2:698(DCl:794 Tl:698 Ul:727) 98W 141A  92c] ASIC:[120c (125c) 1000hz(BL:1000) 14330 (E:193) F:0 L:0]
 7: DC2DC/1/:[vlt1:646 vlt2:653(DCl:794 Tl:653 Ul:727) 70W 108A  85c] ASIC:[120c (125c) 840hz(BL: 840) 12246 (E:193) F:0 L:0]

[H:HW:1513Gh (500),DC-W:773,L:0,A:8,MMtmp:0 TMP:(27/27)=>=>=>(79/79 , 69/69)]
Pushed 10 jobs , in HW queue 4 jobs (sw:2, hw:2)!
min:59 wins:53412[this/last min:17/15] bist-fail:1758, hw-err:0
leading-zeroes:42 idle promils[s/m]:0/0, rate:1048gh/s asic-count:29392 (wins:8+9)
wall watts:1048
Fan:40, conseq:200

Any mods?

Ya im wonderin that myself cause he/she quoted me but if he did what i did to 3 miners so far he would have better numbers then that.

EDIT: oops didnt notice his voltage settings so his clock is higher then mine but the same questions still stand though.

Interesting, I guess that will really help for the ASIC chips, since air direct blow on the heat sinks. The question is about the rest of the components, for example the PCI-E connectors, they will receive hot air blow out of the case, if they also have a high load, will they burn?

Currently I modified open case have all the ASICs running below 85c at 0.61V, 43db, but I found out that ASICs works better when they reach 100c, I can try this inverted fan setup when I have time

any news on that?

I think I'll need to flip the fans too.
What about the noise?

U should flip the fan. Idk y most dont. Just make sure u dont over tighten the fan screws otherwise the blades might not spin. One of the 3 sp20's i have had the fan grill slightly bent inward...i couldnt tell that it was and still cant but once the fan was reattached after being flipped the blades didnt spin. I had to losen one screw more then the other 3 and it spun freely again. I boticed it when it powered on and didnt start with the fans at full speed for the first few seconds like it normally would not when it started mining so be sure to not assume they spin freely just spend the few seconds and spin it by hand to make sure before putting the top bracket and cover back on.

My pci connectors havent mented or discolored in ne way. I needed to run them with the fan flipped because i couldnt have the huge noise they generated when running in the original direction.

I flipped my fan and eventually got melted connectors. Ambient temperature is pretty high at that location though.

Thanks for sharing this, Newar!

How hard were you pushing your SP20 in terms of voltage/power/fan settings?

I have not flipped the fan yet, I just bought bunch of GT2150 fans from chinese online dealer and they provide better airflow than GT1850, at 50dB 1 meter away, not noisy at all when I'm sitting 7 meters away in another room

It is summer time and intake can go up to 30c degree, but the connectors are all quite cool in a pull fan configuration

The only limit I had in place was max. 250 Watt instead of 288 Watt. IIRC, 250 is offcial Molex spec limit, but Spondoolies were told 288 is ok by their supplier (this info I found in one of the threads, but don't have a link handy). Considering the temps at that location I went with the official spec.

Perhaps unsurprisingly, the top connector in the centre of the grille got the worst of it. Got to love basic physics! ;)  Maybe to flip the unit so the connectors are on the bottom would be something to consider.