Klondike - 16 chip ASIC Open Source Board - Preliminary

[drewh]

Sorry to change the subject from Heatsinks.

I'm a little concerned about the plan for reading RESULT using the ESAUSRT. It seems like a nice hack but what happens if the serial port gets into a bad state - if an extra clock edge is caused by a reset, noise, or a result collision, and the framing comes out of sync - the serial port's bit count is wrong.

I agree that result collisions will be rare enough that they can be ignored - as long as the failure mode is simply a dropped share, but if the failure mode is that it stops being possible to get a valid result until a system reset, that would not be acceptable. It could also be very difficult to debug.

The solution might be as simple as a periodic process that detects that the serial port is in a bad state, has been for minimum amount of time, and does a firmware reset of the port to correct the situation.

I've seen this type of problem in other designs and just wanted to make sure you were aware of it.

[MbaseCNC]

It's all a big guess for now.
Also, there seems to be no schematics for the Klondike design, which makes it very hard to troubleshoot or check for the community.
I'm very surprised nobody asked for the schematics before (except for me, but no answer from the designer).
It doesn't give me a good feeling.

[wrenchmonkey]

It's all a big guess for now.
Also, there seems to be no schematics for the Klondike design, which makes it very hard to troubleshoot or check for the community.
I'm very surprised nobody asked for the schematics before (except for me, but no answer from the designer).
It doesn't give me a good feeling.

ARE YOU SERIOUS? The guy is putting INSANE amounts of work into this, and has stated that he doesn't want to release anything until it's tested. What are you implying by "It doesn't give me a good feeling"? If you aren't on board, then find another project. The guy's busy, is sticking with what he already stated he would do. And just because he doesn't spoon feed you, you're gonna start getting weird? 

[BkkCoins]

Sorry to change the subject from Heatsinks.

I'm a little concerned about the plan for reading RESULT using the ESAUSRT. It seems like a nice hack but what happens if the serial port gets into a bad state - if an extra clock edge is caused by a reset, noise, or a result collision, and the framing comes out of sync - the serial port's bit count is wrong.

I agree that result collisions will be rare enough that they can be ignored - as long as the failure mode is simply a dropped share, but if the failure mode is that it stops being possible to get a valid result until a system reset, that would not be acceptable. It could also be very difficult to debug.

The solution might be as simple as a periodic process that detects that the serial port is in a bad state, has been for minimum amount of time, and does a firmware reset of the port to correct the situation.

I've seen this type of problem in other designs and just wanted to make sure you were aware of it.

I would probably re-init the serial receiver at the end of pushing new work each time. That way it is always in a known state at the start of work.

It's all a big guess for now.
Also, there seems to be no schematics for the Klondike design, which makes it very hard to troubleshoot or check for the community.
I'm very surprised nobody asked for the schematics before (except for me, but no answer from the designer).
It doesn't give me a good feeling.

Many people have asked for schematics (every day) but I won't be providing them until I have a working product to release. I believe I stated that up thread somewhere. If not, well, I thought I did. I gave this much consideration before deciding that I'm not obliged to help everyone flood the market with clones before I even get the product of my own efforts released.

[MbaseCNC]

It's all a big guess for now.
Also, there seems to be no schematics for the Klondike design, which makes it very hard to troubleshoot or check for the community.
I'm very surprised nobody asked for the schematics before (except for me, but no answer from the designer).
It doesn't give me a good feeling.

ARE YOU SERIOUS? The guy is putting INSANE amounts of work into this, and has stated that he doesn't want to release anything until it's tested. What are you implying by "It doesn't give me a good feeling"? If you aren't on board, then find another project. The guy's busy, is sticking with what he already stated he would do. And just because he doesn't spoon feed you, you're gonna start getting weird? 

I know he's putting al his time into this, and I now understand he doesn't want to release the schematics before a working product.
It's not about spoon feeding, we all would like to contribute to this and I would like to do my share.
"It doesn't give me a good feeling", sorry, wrong choice of words from my side.
At least now I know why the schematics aren't released yet, but that they will be released in the future.

[TomKeddie]

I would probably re-init the serial receiver at the end of pushing new work each time. That way it is always in a known state at the start of work.

++ Exactly what I was going to suggest.

[wrenchmonkey]

It's all a big guess for now.
Also, there seems to be no schematics for the Klondike design, which makes it very hard to troubleshoot or check for the community.
I'm very surprised nobody asked for the schematics before (except for me, but no answer from the designer).
It doesn't give me a good feeling.

ARE YOU SERIOUS? The guy is putting INSANE amounts of work into this, and has stated that he doesn't want to release anything until it's tested. What are you implying by "It doesn't give me a good feeling"? If you aren't on board, then find another project. The guy's busy, is sticking with what he already stated he would do. And just because he doesn't spoon feed you, you're gonna start getting weird? 

I know he's putting al his time into this, and I now understand he doesn't want to release the schematics before a working product.
It's not about spoon feeding, we all would like to contribute to this and I would like to do my share.
"It doesn't give me a good feeling", sorry, wrong choice of words from my side.
At least now I know why the schematics aren't released yet, but that they will be released in the future.

It's nice that you want to help. I apologize if I came off harsh, but it seemed like you were implying something about the OP, that didn't sit well with me. This forum in general seems to be a giant troll magnet, with people looking for any excuse to sling accusations or passive-aggressive allegations.

That being said, he's stated several times in the thread that he's not going to release schematics until he's absolutely certain that it's ready and fully tested. If anything, that personally gives ME much more confidence, and a "good feeling". I think a lot of people are coming late to the party, and not bothering to read the whole thread. There's a decent amount of fluff in the thread, but if you even just read all the posts by OP (BKKCoins) you will get a pretty good understanding of what's going on, and why.  

[steamboat]

Just to put a bad diagram to your plans for Klego's.... which orientation are you thinking again a vertical or horizontal arrangement?

Fans at the end of either end of the tunnel would work in either arrangement. The horizontal version can fit in side a rack mount 19" wide I believe. The horizontal version would also less prone to heating I think given the density.

As BkkCoins mentioned, There needs to be more spacing on the front side of the boards. I am working on a stacking design with the boards laying flat, heatsinks facing each other, with standoffs to provide spacing.

Figure at minimum 12mm on the front of the board for clearance, double that if you want to use vertical pcie connectors. Keep in mind the wiring needed for the board, you don't want to design anything that will kink the wires or make them difficult to replace. I'll have some initial designs up in the next couple days.


For the SATA debate. I think it's a very bad idea to use SATA. As mentioned earlier, they aren't designed to pull that much power. Additionally, there is more waste from power supplies that are only going to be powering the boards. Most power from the PSU is on the 12v rails. The best density would be to daisy chain pcie connectors to each set of boards, using as much of the 12v rail as possible.

[danattacker]

I'm not sure if this has come up yet, but seeing the recent thread about someone "reconditioning" their Avalon and mentioning that the pads underneath each chip are lower than the soldermask, it got me thinking...

Would it be better to leave the whole bottom of the board without soldermask?

and...

What kind of surface finish do you plan on using? I believe any type of plating, such as gold, will leave a flat surface. But, HASL will not leave a flat surface, possibly degrading the thermal bond with the heatsink.

I don't think any of these issues are show stoppers but I was wondering what you guys thought.

Edit: Maybe these are not issues if you plan to use a thermal pad.

[sensei]

A thermal pad, tape or paste would be required IMHO.

[turtle83]

A thermal pad, tape or paste would be required IMHO.

https://bitcointalk.org/index.php?topic=212090.0 <-- Avalon got it wrong imho. Perhaps a soft thick pad? or the blob of paste under each vias like OP.

[TomKeddie]

https://bitcointalk.org/index.php?topic=212090.0 <-- Avalon got it wrong imho. Perhaps a soft thick pad? or the blob of paste under each vias like OP.

Quite stunning photos, you can see the marks from the applicator used to apply the paste.  It is also interesting to notice that it worked just fine with these issues and the temperature reduction wasn't massive.

I'll be using heat pads, I have a small amount lying around from previous projects.

[BkkCoins]

I'm not sure if this has come up yet, but seeing the recent thread about someone "reconditioning" their Avalon and mentioning that the pads underneath each chip are lower than the soldermask, it got me thinking...

Would it be better to leave the whole bottom of the board without soldermask?

and...

What kind of surface finish do you plan on using? I believe any type of plating, such as gold, will leave a flat surface. But, HASL will not leave a flat surface, possibly degrading the thermal bond with the heatsink.

I don't think any of these issues are show stoppers but I was wondering what you guys thought.

Edit: Maybe these are not issues if you plan to use a thermal pad.

In order to keep the board reasonably cheap I won't be able to expose full copper on the bottom. This is because it would require using blind/buried vias so that inner layers don't short out on the bottom. This is why I went with exposed pads under each ASIC, and the tracks running from each ASIC have to clear the pad area to go via.

I'll likely just use HASL for the test boards so I can see how well that works, but I agree with it not being flat and it's likely that final boards would be ENIG (more costly)(depending on just how much trouble the HASL turns out to be). I'm really thinking there will be some 1x1cm pad under each ASIC, at least enough to account for the solder mask thickness.

I'm interested to know exactly what was meant by the person who took apart the Avalon and inspected the heat sink mating. I don't quite have a good image from the comment above. Did he mean it was just heat sink against mask except some exposed areas, or there was solder mask but thin pads that held the heat sink slightly off the mask, or something else?

edit: never mind - just went to see the photos. This makes me feel better...
(because if that works then almost anything will)

[drewh]

I would probably re-init the serial receiver at the end of pushing new work each time. That way it is always in a known state at the start of work.

That seems reasonable.

[glendall]

Really getting excited about this design . Keep up the good work BKKCoins.. !

[sensei]

A thermal pad, tape or paste would be required IMHO.

https://bitcointalk.org/index.php?topic=212090.0 <-- Avalon got it wrong imho. Perhaps a soft thick pad? or the blob of paste under each vias like OP.

What the heck were they thinking? Let's spread it on and then wipe it all off? Didn't they do any production process testing and tear it down to see how well it works?

I like thermal paste. We use it on all our production product. Only prototypes get thermal tape.

BTW: That's one big heat sink.

[dunand]

Is there a group or company interrested in manifacturing the klondike in Canada?

[peonminer]

sub'd

[limit]

Great Job. Watching

[sensei]

Is there a group or company interrested in manifacturing the klondike in Canada?

I will be, and I'm just north of Canada.

Windsor, Canada that is...