Klondike - 16 chip ASIC Open Source Board - Preliminary

[ecliptic]

It might not be that hard to switch out the regulator for a higher current one as well.  Not sure what the package is and if there is another one that fits though.

The problem is there are no higher current regs. That was the highest I could find at 16A, and I didn't want to start playing with ganged up regs. An alternate choice was putting 4 regs per board at 12A each, or going controller + discrete like on the Avalon, but other than the ASIC they are the most costly part. Well, not counting the heat sink either.

I see.  I edited my post but perhaps you could add the component pads for a third regulator of the same type, but don't populate it by default.  If people want to OC they can add it manually to increase the current available on the 1.2V plane.  It would also very likely reduce the current draw on the original two, (unless you were to OC so much you max out the new third regulator)

Similarly it might make sense to add locations for additional decoupling caps on the 1.2V lines of the chips themselves, again not populating them.  often it is the ESR and ESL of these lines that determines the decoupling efficency, so having more caps could allow you to push them higher, if voltage is the problem (as opposed to heat)

[1715281204]

1715281204

[1715281204]

1715281204

[1715281204]

1715281204

[BkkCoins]

It might not be that hard to switch out the regulator for a higher current one as well.  Not sure what the package is and if there is another one that fits though.

The problem is there are no higher current regs. That was the highest I could find at 16A, and I didn't want to start playing with ganged up regs. An alternate choice was putting 4 regs per board at 12A each, or going controller + discrete like on the Avalon, but other than the ASIC they are the most costly part. Well, not counting the heat sink either.

I see.  I edited my post but perhaps you could add the component pads for a third regulator of the same type, but don't populate it by default.  If people want to OC they can add it manually to increase the current available on the 1.2V plane.  It would also very likely reduce the current draw on the original two, (unless you were to OC so much you max out the new third regulator)

Similarly it might make sense to add locations for additional decoupling caps on the 1.2V lines of the chips themselves, again not populating them.  often it is the ESR and ESL of these lines that determines the decoupling efficency, so having more caps could allow you to push them higher, if voltage is the problem (as opposed to heat)

I'm not going to worry about it for now. I suspect that there isn't much head room above 300 or we'd have heard something about it. If people test with 14 chips and find you can do, say, 350 MHZ at X amps power, then I'd think about a revised board aimed at overclockers with programmable core voltage as well - but IMO that's getting ahead of the task right now, which is to make a working board at all.

[ryepdx]

...IMO that's getting ahead of the task right now, which is to make a working board at all.

+1

[erk]

What's the eta on the sample Avalon chips, to see if this whole thing works as intended?

[wrenchmonkey]

It might not be that hard to switch out the regulator for a higher current one as well.  Not sure what the package is and if there is another one that fits though.

The problem is there are no higher current regs. That was the highest I could find at 16A, and I didn't want to start playing with ganged up regs. An alternate choice was putting 4 regs per board at 12A each, or going controller + discrete like on the Avalon, but other than the ASIC they are the most costly part. Well, not counting the heat sink either.

I see.  I edited my post but perhaps you could add the component pads for a third regulator of the same type, but don't populate it by default.  If people want to OC they can add it manually to increase the current available on the 1.2V plane.  It would also very likely reduce the current draw on the original two, (unless you were to OC so much you max out the new third regulator)

Similarly it might make sense to add locations for additional decoupling caps on the 1.2V lines of the chips themselves, again not populating them.  often it is the ESR and ESL of these lines that determines the decoupling efficency, so having more caps could allow you to push them higher, if voltage is the problem (as opposed to heat)

I'm not going to worry about it for now. I suspect that there isn't much head room above 300 or we'd have heard something about it. If people test with 14 chips and find you can do, say, 350 MHZ at X amps power, then I'd think about a revised board aimed at overclockers with programmable core voltage as well - but IMO that's getting ahead of the task right now, which is to make a working board at all.

I agree. OC will be nice, but ultimately, the primary goal should be getting stable boards produced and functioning. There will be plenty of time to tweak. OC/OV can be in version 1.1.

[Bicknellski]

...IMO that's getting ahead of the task right now, which is to make a working board at all.

+1

+1

If that is all you can get out of her Scotty that is fine... had to ask though cause you know everyone is thinking it.

[BkkCoins]

I'm just leaving for my 36 hour Bangkok round trip. I won't be checking in again until I get back. So don't worry that I'm not responding here - I'll be back with PIC chips and parts and proto-boards etc. No test boards yet but maybe next week. I hope when I get back FC4B will have some money in PP so I can sell some btc for that scope.

[turtle83]

I'm just leaving for my 36 hour Bangkok round trip. I won't be checking in again until I get back. So don't worry that I'm not responding here - I'll be back with PIC chips and parts and proto-boards etc. No test boards yet but maybe next week. I hope when I get back FC4B will have some money in PP so I can sell some btc for that scope.

Let me know if u got some time to spare... ill buy you a beer or 2...

[Damme]

I'm following this thread with a lot of excitement, I do alot of hardware developing myself, and I would in future want to order pcb's for soldering and programming myself.
But my thoughts are about the chips. There are tons of group buy, but, how do we know those are legit and not scams? Which to choose ? there are alot too choose from. (Sweden here)

I'm a bit worried to order chips actually! Any advice?

[erk]

I'm following this thread with a lot of excitement, I do alot of hardware developing myself, and I would in future want to order pcb's for soldering and programming myself.
But my thoughts are about the chips. There are tons of group buy, but, how do we know those are legit and not scams? Which to choose ? there are alot too choose from. (Sweden here)

I'm a bit worried to order chips actually! Any advice?

If in doubt, wait until completed boards are available from an online site, and they have been reviewed by several forum users. That's what I will be doing. I think the process of going into a chip group buy is an interesting idea to get things moving, but now it seems several people have done bulk chip buys with the sole idea of building and on-selling completed boards as a commercial venture.

[0xFE]

Yes. I try to delay the extracted clock output long enough for the setup time on the data input. It may need some trial and error there - and the scope could be useful to see what it's doing.

Oh, I see - I thought the clock input is delayed. Thanks for the clarification.

[Bicknellski]

I'm following this thread with a lot of excitement, I do alot of hardware developing myself, and I would in future want to order pcb's for soldering and programming myself.
But my thoughts are about the chips. There are tons of group buy, but, how do we know those are legit and not scams? Which to choose ? there are alot too choose from. (Sweden here)

I'm a bit worried to order chips actually! Any advice?

Hard to know for sure. Check the reputations of the various people running the group buys. Read carefully check their reputation by their previous posts in the Forums.

Zefir https://bitcointalk.org/index.php?topic=177827.0 (closed not sure if he will get more or not)
Steamboat  https://bitcointalk.org/index.php?topic=192916.0

Both of them jump out as people I think you can check into and get some answers definitively about their history. Let me know how that goes for you.

Always remember don't put in anything you can't afford to lose. We are all taking some big risks even just hoping that BKKCoins will be able to deliver. Chips come from Avalon and who is to say they won't be delayed or damaged... lots of chances for this all to get sideways. I hope it all goes well for us but that is the risk we are taking.

[DaGreatRV]

I'm following this thread with a lot of excitement, I do alot of hardware developing myself, and I would in future want to order pcb's for soldering and programming myself.
But my thoughts are about the chips. There are tons of group buy, but, how do we know those are legit and not scams? Which to choose ? there are alot too choose from. (Sweden here)

I'm a bit worried to order chips actually! Any advice?

I ordered my chips at t13hydra's group buy. Looks legit so far. What convinced me to buy there was the fact that John K. supplied the escrow.
Anyway, I won't know for a few weeks if it was a good choice. There is always risk involved.
https://bitcointalk.org/index.php?topic=189582.0

[DobZombie]

I ordered my chips at t13hydra's group buy. Looks legit so far. What convinced me to buy there was the fact that John K. supplied the escrow.
Anyway, I won't know for a few weeks if it was a good choice. There is always risk involved.
https://bitcointalk.org/index.php?topic=189582.0

Same here

[SebastianJu]

Offtopic... Since i mentioned it before: Waiting and getting informed by ebay-email-search-alarm finally worked... i got a seasonic Platinum 1000W that costs 246€ at amazon for 117€... *g*

[Ente]

Not sure how much was mentioned here already.

What numbers concerning the thermal resistance do we already have?
I want to calculate through the temperature, to approximate how I will keep all of this cool.
I may go the watercooling-route here, but most of the calculation will be the same for airooling.

1) Junction to chippackage
2) Solder layer between chippackage and board
3) Board with 25 vias
4) Thermal paste between board and aluminum

Watercooling:
5a) X mm aluminum
6a) Aluminum to water
7a) Radiator with/without fan to air

Aircooling:
5b) Heatsink with/without fan to air


I will calculate with a conservative 2w per chip and an area of 5x5mm (correct?) only for 2) and 3), and a somewhat larger area for 5a) 6a) and 5b).

Anyone has some numbers handy already?

Ente

[Ente]

-collecting numbers in this post here-

1) Junction to chippackage 0.001K/W
2) Solder layer between chippackage and board 0.01K/w
3) Board with 25 vias 3.8K/W
4) Thermal paste between board and aluminum 2.5K/W

Watercooling:
5a) X mm aluminum
6a) Aluminum to water
--> I'll simplify this to "T_alu = T_water"
7a) Radiator with/without fan to air 0.25K/W


Aircooling:
5b) Heatsink with/without fan to air 0.5K/W


2w per chip
5x5mm for 2) and 3)
somewhat larger area for 5a) 6a) and 5b)



This tutorial was posted earlier, it has almost all answers already:
http://www.electronics-cooling.com/2004/08/thermal-vias-a-packaging-engineers-best-friend/

1) Silicon to chippackage: 0.001K/W

2) As shown on the thermal imaging here:
https://bitcointalk.org/index.php?topic=190731.msg2043405;topicseen#msg2043405
I calculate with a delta of 7K between chip and board (62°C and 55°C respectively).
This assumes the soldering and boardsetup is comparable between Avalon and Klondike.
--> Nah, 3.5K/W is waay too high. The tutorial says 0.008K/W, 0.01K/W seems more realistic here. We don't use the length of the board for heat transportation, but the thickness!

3) 5x5mm area with 5x5 grid of vias, each 0.3mm diameter.
Bottom layer is 1 oz. copper
electronics-cooling says 3.8K/W with a 5x5 grid of vias (at 2mm spacing though, ignored)

4) https://en.wikipedia.org/wiki/Thermal_grease says 0.5K/W @ 12.6cm², which calculates to 2.5K/W for our 5x5mm area.

5b) I randomly found a value of 0.3K/W for a CPU-cooler with a fan. Let's calculate with 0.5K/W for our non-sophisticated DIY solution.

7a) Found 0.21..0.35K/W for a 12cm radiator. The larger the radiator, the wore heat gettin' rid of. I'll use 0.25K/W here, although this value is pretty random - can be reduced at will with more/larger radiator.

..I'll calculate a bit with those values later.


Ente

[Bicknellski]

Nice work Ente... been waiting on someone with the right math skills to do this.

What specs for the heat sink do you think will work?

[cp1]

Where are you located? I might be able to loan you a scope if you're local.

[Ente]

Nice work Ente... been waiting on someone with the right math skills to do this.

What specs for the heat sink do you think will work?

Ah, that stuff is easy once you figure it out! :-)

"0.5K/W" means, for example, that for each W[att] you throw at it, you have a K[temperature] difference of 0.5 degree. Like a 30 watt CPU cooled with a 0.5K/W heatsink would be 30*0.5 = 15 degree above the air temperature.
You add up all thermal resistances between the heat producer (CPU or ASIC core) and the surrounding (air). This is the thermal resistance of the cooling system.

Here we are at around 7K/W combined. Interestingly, the difference between air- and watercooling is neglicible, at an additional 0.25 K only.
This 7K/W is for each chip, it doesn't matter that we have 16 chips combined (as long as the heatsink covers the whole 10x10cm, doh!). This means, at 2 watts, that the silicon will be 14 degree warmer than ambient. Best case. Calculate 20 degree for some margin.

So, air or water?
Depends on your setup. If it's just a few chips, you won't have problems with air. As soon as your room would warm up, like, in summer with a thousand chips, you might prefer watercooling. Not because it cools better (it does not), but because you can cool the warm water somewhere else than where the chips are. Like having the miners inside and the radiator outside. Or warming your pool with it (use two watersystemns for that!).

Air? Any regular heatsink with a regular, slow spinning fan should be more than enough. A 1K/W heatsink isn't anything special, and that's with no fan! So, basically, as long as you have any aluminum piece with fins on it and have a little air blowing at it too, everything is sweet! :-)

In hindsight the whole calculation isn't really necessary. I didn't expect it to work out that nicely.

Homework:
Calculate similar systems with a 90w CPU (20x20mm), regular thermal compound and a bad/calm 0.5K/W heatsink. Max allowed CPU temperature is 70°C. How hot may the air in summer be at maximum, before the CPU starts cycling down its speed?

edit: Homework #2: What performance must a better heatsink have at least to work up to 30°C with the same CPU?

edit#2: Thank you Sophokles, indeed I had a typo at 0.5K

Ente