Author

Topic: [Review] SP20 Jackson in pictures (Read 1829 times)

legendary
Activity: 1274
Merit: 1004
December 22, 2014, 03:44:28 PM
#11
I'll be locking this one now, please continue discussion as needed in the main legendaries review thread.
https://bitcointalksearch.org/topic/closed-legendaries-review-party-of-spondoolies-techs-sp20-887807
legendary
Activity: 1274
Merit: 1004
December 22, 2014, 03:39:15 PM
#10
There was no duty on it, but I did end up paying GST, a $10 brokerage charge, and a COD fee. Pretty standard for anything you bring into the country.

For aclass, not really. On the die that I removed the heatsink I did notice that the spring force holding the heatsinks down with the copper shims in place is extremely weak. Originally I'd thought the glue was to keep the heatsinks attached during shipping, but it might also be necessary to maintain good contact with the die. Using just the included hardware but without the glue holding it down the copper plate would actually lose contact with the die when placed vertical.
I replaced the springs with some slightly longer and higher spring constant ones, and it worked quite well. I also just installed the aluminum heatsink and left off the copper shim off to see if there was any noticeable improvement/degradation vs having the copper shim but another air/grease gap, but it seems to be a wash.

Maybe next up we'll give this a go.
sr. member
Activity: 381
Merit: 251
December 22, 2014, 01:29:11 PM
#9
Mr.T: did you test it before and after getting it apart/together? Any differences?
sr. member
Activity: 272
Merit: 250
December 22, 2014, 01:28:27 PM
#8
Love the review.
Most informative from the ones posted already.
hero member
Activity: 572
Merit: 500
December 22, 2014, 01:24:32 PM
#7
Very informative technical review MrTeal, Thank you

It is a pleasure to see a review by someone with the expertise able to get into the heart of the beast and make intelligent comments with respect to the build and components used and still be able to reassemble to a functioning unit.

Just one small question from a fellow Canadian, did the unit clear customs duty free?

Trends

The Canadian customs charges are paid automatically by DHL. They collect them from you together with a small broker fee. You can pay that before or at pickup. They got from me customs duty and fees for every piece of hardware coming from SPT (SP10, SP30, RMA components). In order to get something back from them you need to speak with DHL customs and prove that the parts left the country. Then they'll credit the HST on those parts. They'll keep their fee though.  

Disclosure: I still run 3 Chili boards (100Gh/s total) designed by MrTeal around the BFL chip Cheesy Nice to see a SP20 review from someone who can actually re-engineer the miner if he puts his mind to it.
member
Activity: 119
Merit: 10
December 22, 2014, 09:06:05 AM
#6
Very informative technical review MrTeal, Thank you

It is a pleasure to see a review by someone with the expertise able to get into the heart of the beast and make intelligent comments with respect to the build and components used and still be able to reassemble to a functioning unit.

Just one small question from a fellow Canadian, did the unit clear customs duty free?

Trends
legendary
Activity: 1190
Merit: 1000
December 22, 2014, 12:13:31 AM
#5
Interesting review from technical point of view. Thanks  Smiley
legendary
Activity: 4256
Merit: 8551
'The right to privacy matters'
December 21, 2014, 11:58:12 PM
#4
some nice parts in there.
donator
Activity: 1414
Merit: 1051
Spondoolies, Beam & DAGlabs
December 21, 2014, 11:23:48 PM
#3
Well, I got my SP20 a little later than most and didn't get to the review as soon as I'd like, so this will be a bit different than most.
Most others have gone over the basic unboxing and review of the web interface, so I won't touch on that here. Instead, I'll go over a couple parts of the hashing board for people who might not know what all the parts are.
....
Thank you for an interesting, different, review.
legendary
Activity: 3416
Merit: 1912
The Concierge of Crypto
December 21, 2014, 11:15:46 PM
#2
Nice. The nice thing about review units, is, I'm pretty sure there is no warranty, therefore, we can open them up and not worry about if it breaks or not.

I'm waiting for someone to do a drop test. (just kidding.)
legendary
Activity: 1274
Merit: 1004
December 21, 2014, 10:40:02 PM
#1
Well, I got my SP20 a little later than most and didn't get to the review as soon as I'd like, so this will be a bit different than most.
Most others have gone over the basic unboxing and review of the web interface, so I won't touch on that here. Instead, I'll go over a couple parts of the hashing board for people who might not know what all the parts are.

Disassembly

Holding each heatsink on is a glob of extremely tough glue. They really don't want these things coming off. If you do get it off, you can remove the heatsink.

Under the heatsink is a copper shim, with some thermal grease at the interface. The shim had some strange imperfections though.

As you can see, there's a couple marks on the shim, right where it touches the die. Not really ideal, but it didn't seem to really make a big difference. Underneigth, well..

That's a LOT of thermal paste. Thankfully they use an extremely high flow compound, so it all just squeezes out the sides. Clean it off, and you get a much better look at the die.

That's much better. As you can see, this die was packaged in week 44, so end-Oct or the start of November. They don't have a huge lag between getting chips in and sending out products. You can also see SP is using an 8 layer board from the layer marker at the top of the image.

VRM Show and Tell
Now that we've got the board nicely take apart, let's see what is going on for each ASIC. First the top of the board.

There's a few supporting blocks in frame here. First, in green there is a DC/DC convertor that provides one of the minor voltage rails.
The blue squares are a pair of LM75 temperature sensors. These have a digital I2C interface, and they run to an I2C switch between the sensors and the minor voltage rail.
Up top in pink you have some caps on the 12V rail right near the output devices.
The center is dominated by the VRM controllers. SPTech went with two TPS40425 analog converters, in a master/slave relationship to provide 4 phases to the ASIC.
These are driverless convertors, they just provide the brains and don't drive any mosfets directly. To see the main outputs, we need to look at the other side.

Starting at the bottom, in red you can see an input inductor that filters the incoming 12V from the PSU. In pink you can see more caps on that filtered 12V rail. While the 12V rail doesn't have nearly current flow as the 0.7V output, the change in voltage wrt time (dv/dt) is extremely quick, so it's good to see lots of decoupling there.
Next in yellow you can see the main output devices, TI's CSD95372. These 60A devices are quite new parts, with some very nice looking specs. They provide the main switching that converts the 12V rail into pulses of 12V that will be filtered later.
Before the output inductors, there's two more components in blue that for a zobel network. The zobel helps to reduce some high frequency ringing on the switching waveform. It's a bit of a safety, you don't want too much ringing, but the zobel does dissipate energy so it can reduce efficiency a bit.
Next up is the main inductors. These guys are the main filtering elements, and are what change the pulsed 12V waveform that comes out of the mosfets into a very wavey 0.7V. To get the output nice and smooth we still need more filtering though.
At the top of the image in light blue you have the main output capacitors. There guys get charged up while the inductors are flowing current, and discharge it to the ASIC while the inductors are off. They also provide another often overlooked but very important protection. Because current through an inductor doesn't change instantly, if the load switches off suddenly (like the ASIC is no longer hashing) the high current will keep flowing and the voltage can spike. Often the minimum capacitance you need isn't set by what you need to keep ripple low, it's chosen to keep the maximum voltage low enough to prevent damage in the case that the ASIC suddenly stops hashing.
Lastly at the top of the image you can see the bottom side of the ASIC, with a bunch more small value decoupling caps as close to the chip as possible.


Anyway, I hope this has been a bit informative. I'm trying to fight off a killer head cold and probably have minutes left before the NyQuil finishes kicking in, so I'll check back in tomorrow to see if there's any questions.
Jump to: