Klondike - 16 chip ASIC Open Source Board - Preliminary

[ryepdx]

Kinda interested in the k-1

I don't see anyone willing to make them though

I'm making them, assuming I get enough sales to fill a batch of 128.
www.stumptownminers.com
https://bitcointalk.org/index.php?topic=203388.0

[SoggyLettuce]

Kinda interested in the k-1

I don't see anyone willing to make them though

Go check out sensei's thread. He seems to have his head on straight and i'm probably going to get some k1s from him.

[BkkCoins]

So the bake went ok. There are a few things I need to fix up. Here's a few photos snapped with the USB microscope and not enough light (it's late night here). Notice a solder bridge on PIC, IR3895 reg. twisted out of position a bit and I guess I'll have to use my rework station and adjust that. Maybe a bit too much paste? And a couple 0402 caps just plain sat up right. I thought they'd popped or exploded but I didn't hear anything, but then with the cam I saw they had just turned upwards and so only looked like they were gone.





Notice the USB conn didn't have the locator pin holes drilled. Not sure why yet. That's why I didn't place it as I didn't want to drill with all the parts on there. Then I thought I should just cut the pins off so maybe I'll do that.

Anyway. I'm off to bed. Tomorrow I'm hoping this thing will see some power.

[marto74]

It's without holes on plot files , same here

[marto74]

For test I reserve the right to ask if something pops up

[BkkCoins]

For grins, I have tried this and it worked at 325, 350 and 375. The error rate at 375 meant the useful hashrate was lower than at 300, but it worked at 325 and 350 quite well and I got 81GH. The power usage went up by 5W (at the wall) at 325 and 10W at 350.

Does anyone know what this 5W and 10W increase is relative to? Is it usually around 600W at the wall or something else? He said after a few hours at 350 he got 82GH steady but that isn't much more than 80 for a 16% increase in clock.

[marto74]

For grins, I have tried this and it worked at 325, 350 and 375. The error rate at 375 meant the useful hashrate was lower than at 300, but it worked at 325 and 350 quite well and I got 81GH. The power usage went up by 5W (at the wall) at 325 and 10W at 350.

Does anyone know what this 5W and 10W increase is relative to? Is it usually around 600W at the wall or something else? He said after a few hours at 350 he got 82GH steady but that isn't much more than 80 for a 16% increase in clock.

a friend of mine is going to test this tonight.
Ill ask him tomorrow for update

[BkkCoins]

It's without holes on plot files , same here

I had a look and there are holes in the part footprint but they're marked as "NPTH Mechanical", instead of Thru-Hole. Which makes sense to me but I noticed that the holes aren't in the drill file but another NPTH drill file is created by Kicad. So I guess it depends on the PCB Fab and how they handle it. In my case I never sent the NPTH drill file so they obviously didn't drill it. But next time I'll ask them if they need it all in one file or two files, or if it should just be done as plated holes if they don't support a second NPTH file. So that's another change for v0.2.

[SebastianJu]

The 6 pin PCIE cables use 3 x 12V lines, providing 6.25A and a a maximum wattage of 75W. That should allow 2 devices to be powered per cable using a splitter.

Also, how many amps does a K16 use?

You should expect around 40W (3.3A) . It probably will use less. Using the max ratings for regulators, under a condition of over clocking, you could probably draw up to 55W (4.6A) beyond which it would be dicey and shut down should occur if the regs are doing their job. Off hand, I didn't go check at what rating the shutdown  kicks in. This is allowing an approximate 75% for regulator inefficiency but actually hope we get better.

Does this mean the overclocking capability is limited by hardware or what does this mean? Is it a protection of some kind?

[ecliptic]

What are the options/alternatives for the k16 heatsinks?

There are many options covered in this thread, I suggest you look at some rather than having people repeat them over and over.

Type the word heatsink into the search and you should get 6 pages of hits.

People shouldn't need to search through multiple pages of mostly irrelevant information to find 1 comment which makes a passing mention of the info you're looking for.

Any kit that is currently ready should be linked to in the initial post.

There are no kits in this thread, it's a design discussion thread. Go look at one of the board buy threads, there are many, like Burnin's or Terrahash threads.

Since when is a heatsink not part of a PCB assembly design?

[TomKeddie]

Since when is a heatsink not part of a PCB assembly design?

Sure but a heatsink is kinda useless without a working circuit to generate the heat, this thread is focused on the first step, comparatively the heatsink is easy.

You seem to expect more than R&D in an R&D thread, there is nothing for sale yet here.  Most assemblers are offering heatsinks, if you want to assemble yourself try to buy from one of the assemblers.

Heatsink : https://bitcointalk.org/index.php?topic=208381.0

[TomKeddie]

So the bake went ok. There are a few things I need to fix up. Here's a few photos snapped with the USB microscope and not enough light (it's late night here). Notice a solder bridge on PIC, IR3895 reg. twisted out of position a bit and I guess I'll have to use my rework station and adjust that. Maybe a bit too much paste? And a couple 0402 caps just plain sat up right. I thought they'd popped or exploded but I didn't hear anything, but then with the cam I saw they had just turned upwards and so only looked like they were gone.

One think I notice is no heat relief around the pads for the caps.  To get a better joint don't you usually use a pad attached to the plane rather than the plane itself?

[mike2kt]

The 6 pin PCIE cables use 3 x 12V lines, providing 6.25A and a a maximum wattage of 75W. That should allow 2 devices to be powered per cable using a splitter.

Also, how many amps does a K16 use?

You should expect around 40W (3.3A) . It probably will use less. Using the max ratings for regulators, under a condition of over clocking, you could probably draw up to 55W (4.6A) beyond which it would be dicey and shut down should occur if the regs are doing their job. Off hand, I didn't go check at what rating the shutdown  kicks in. This is allowing an approximate 75% for regulator inefficiency but actually hope we get better.

Does this mean the overclocking capability is limited by hardware or what does this mean? Is it a protection of some kind?

Yes and Yes.     The regulators are designed with a maximum current rating (two 16A in the K-16 design). They have built-in protection (Over Voltage, Over Current and Over Temp) to prevent the regulator from frying itself and possibly other components on the board.


This calculation is just on the 16 chips, it does not include the other items on the board. The two regulators are split in the design (each powering 8 chips) but I'll combine the two here for easier math. Also this does not compensate for inefficiency.

Voltage (E) * Current (I) = Power (P)

1.2volts * (2amps * 16chips) = 38.4watts     1.2volts Avalon Typical Core Voltage, 2amps/chip Avalon Maximum Operating Current  

Now let's bump the voltage up to 1.3volts to try for a faster overclock.

Voltage (E)^2 / Power (P) = Resistance (R)

1.2volts^2 / 38.4watts = 0.0375ohms

Voltage (E)^2 / Resistance (R) = Power (P)

1.3volts^2 / 0.0375ohms = 45.1watts      1.32volts Avalon Maximum Core Voltage

Power (P) / Voltage (E) = Current (I)

45watts / 1.3volts = 34.7amps

Translation: Increasing the voltage too possibly gain a higher hashrate will result in a fungible/ballpark increased current demand of ~2.7amps (~1.35amps per) on the regulators.


There are three different protection mechanisms for the two 16A regulators, though here are the protection types to focus on:

THERMAL SHUTDOWN

Temperature sensing is provided inside IR3895. The trip
threshold is typically set to 145C. When trip threshold is
exceeded, thermal shutdown turns off both MOSFETs and
resets the internal soft start.

Automatic restart is initiated when the sensed
temperature drops within the operating range. There is
a 20C hysteresis in the thermal shutdown threshold.

OVER CURRENT PROTECTION

The over current (OC) protection is performed by sensing
current through the RDS(on) of the Synchronous Mosfet.
The current limit is pre�-set internally and is compensated
according to the IC temperature. So at different ambient
temperatures, the over�-current trip threshold remains
almost constant.

The over current implementation details are complex. See the IR3895 datasheet for the details.

[ecliptic]

Since when is a heatsink not part of a PCB assembly design?

Sure but a heatsink is kinda useless without a working circuit to generate the heat, this thread is focused on the first step, comparatively the heatsink is easy.

You seem to expect more than R&D in an R&D thread, there is nothing for sale yet here.  Most assemblers are offering heatsinks, if you want to assemble yourself try to buy from one of the assemblers.

Heatsink : https://bitcointalk.org/index.php?topic=208381.0

It's not the design, it's the sourcing.  Of course getting all the circuity going is critical.  But if you find out you can't get a heatsink for 2 months, you're screwed

I'll work on it this weekend, maybe even setup a group buy or something

this is supposed to be a DIY project, (saying a critical component has to be bought from an assembler?)

[turtle83]

Since when is a heatsink not part of a PCB assembly design?

Sure but a heatsink is kinda useless without a working circuit to generate the heat, this thread is focused on the first step, comparatively the heatsink is easy.

You seem to expect more than R&D in an R&D thread, there is nothing for sale yet here.  Most assemblers are offering heatsinks, if you want to assemble yourself try to buy from one of the assemblers.

Heatsink : https://bitcointalk.org/index.php?topic=208381.0

It's not the design, it's the sourcing.  Of course getting all the circuity going is critical.  But if you find out you can't get a heatsink for 2 months, you're screwed

I'll work on it this weekend, maybe even setup a group buy or something

this is supposed to be a DIY project, (saying a critical component has to be bought from an assembler?)

The heatsink doesn't need much specifications. Just get a heatsink of about the same size (or bigger) . Look earlier in the thread for positioning of the mounting holes. Goto some workshop with CNC machine, and have them drill the mounting holes needed. Buy thermal pad 10cm x 10cm seems to be a common size perfect for K16. Mount, screw, , profit.

[mike2kt]

Mount, screw, profit.

Writing down that one for later use...   

[BkkCoins]

It's not the design, it's the sourcing.  Of course getting all the circuity going is critical.  But if you find out you can't get a heatsink for 2 months, you're screwed

I'll work on it this weekend, maybe even setup a group buy or something

this is supposed to be a DIY project, (saying a critical component has to be bought from an assembler?)

I have a bunch of heat sinks here for testing. Some from ebay, some from local shops (cheap), and some custom ones from assemblers who want to test their "pro" versions. I think a number of different approaches will work of varying professionalism. You could do a proper CNC tap or just drill a thru hole and use a longer bolt. I think ideally a spring mounted bolt would ensure the board isn't stressed but also judging by the Avalon just screwing directly to the board works. I don't think the heat sink will be too critical though having a good one, and especially good thermal mating, will help with any over clocking. That opinion is just based on seeing how the Avalon worked despite what looked like mediocre compound spread and mating. Certainly the approach is different depending on whether you need to source a few or a few thousand heat sinks.

[chaoztc]

regarding I2C:

Hardware supported (which is the only possibilty at the slave) I2C is a pain - all states must be handled correctly else I2C will stall.
This means testing & debugging each step on master and slave (which is a lot of work because I have only 1 pickit).

Today I have finished the I2C code, seems to work now very well and without race conditions - fixed them today.
For tomorrow a test with 3 slave pics (currently only 1 during development) is planned.
If everything works - I will send the code to bkkcoins.

[southerngentuk]

regarding I2C:

Hardware supported (which is the only possibilty at the slave) I2C is a pain - all states must be handled correctly else I2C will stall.
This means testing & debugging each step on master and slave (which is a lot of work because I have only 1 pickit).

Today I have finished the I2C code, seems to work now very well and without race conditions - fixed them today.
For tomorrow a test with 3 slave pics (currently only 1 during development) is planned.
If everything works - I will send the code to bkkcoins.

+1 G8 others are helping bkk where they can. I would help too if I understood half of what was said here...

[marto74]

regarding I2C:

Hardware supported (which is the only possibilty at the slave) I2C is a pain - all states must be handled correctly else I2C will stall.
This means testing & debugging each step on master and slave (which is a lot of work because I have only 1 pickit).

Today I have finished the I2C code, seems to work now very well and without race conditions - fixed them today.
For tomorrow a test with 3 slave pics (currently only 1 during development) is planned.
If everything works - I will send the code to bkkcoins.

we have 10+ delivered here.
DO you need help