Klondike - 16 chip ASIC Open Source Board - Preliminary

[marto74]

EMC-Lab confirmed the appointment.

And i have some more overclocking figures for you guys.
Note: I expect that future versions will be more stable due to a few changes.
341 Mhz 1.2V 57.5W - passed long term tests with 0.2% hw-errors.
409 Mhz 1.2V 68.04W - hw-errors increase to 3%
448 Mhz 1.2V 74.4W - not stable
448 Mhz 1.25V 81.6W - unstable 50% hw-errors
448 Mhz 1.28V 84.016W - unstable 50% hw-errors

Burnin made some overclocking tests, unfortunately without temperature. But maybe good for comparition.

Regarding the klondikes... will they have CE and other signs? Or should i ask the actual assembler for this?

CE mark is a test for safety.
It can be covered only with fully enclosed device.
The process of getting CE approval is quite time and money consuming ( at least 30-40 days) .

[SebastianJu]

EMC-Lab confirmed the appointment.

And i have some more overclocking figures for you guys.
Note: I expect that future versions will be more stable due to a few changes.
341 Mhz 1.2V 57.5W - passed long term tests with 0.2% hw-errors.
409 Mhz 1.2V 68.04W - hw-errors increase to 3%
448 Mhz 1.2V 74.4W - not stable
448 Mhz 1.25V 81.6W - unstable 50% hw-errors
448 Mhz 1.28V 84.016W - unstable 50% hw-errors

Burnin made some overclocking tests, unfortunately without temperature. But maybe good for comparition.

Regarding the klondikes... will they have CE and other signs? Or should i ask the actual assembler for this?

CE mark is a test for safety.
It can be covered only with fully enclosed device.
The process of getting CE approval is quite time and money consuming ( at least 30-40 days) .

I only wonder if it will be possible to host the miners in a normal datacentre, i mean if they accept hardware that doesnt have the CE-Sign or so.

[sensei]

Regarding the klondikes... will they have CE and other signs? Or should i ask the actual assembler for this?

I'm just assembling, not designing.

Unless we meet CE marking requirements, we shouldn't put a mark on the board.

As for assembling. If any component in an assembly is not marked, then the whole assembly cannot be marked, unless tested. For instance, if I put a fan, a board and a power supply in a box and each of those is CE marked, and wire it per directions supplied by the manufactures of those parts, then I can mark the assembly and self certify (if I follow safety guidelines). If either the fan, PS, or board are not CE marked, then I cannot simply self-certify (assuming the board is an active computing device) unless I myself perform some testing to back up the self-certification claim.

[SebastianJu]

Regarding the klondikes... will they have CE and other signs? Or should i ask the actual assembler for this?

I'm just assembling, not designing.

Unless we meet CE marking requirements, we shouldn't put a mark on the board.

As for assembling. If any component in an assembly is not marked, then the whole assembly cannot be marked, unless tested. For instance, if I put a fan, a board and a power supply in a box and each of those is CE marked, and wire it per directions supplied by the manufactures of those parts, then I can mark the assembly and self certify (if I follow safety guidelines). If either the fan, PS, or board are not CE marked, then I cannot simply self-certify (assuming the board is an active computing device) unless I myself perform some testing to back up the self-certification claim.

The thing is only that i most probably will host miners in a datacentre. Im not sure if i will use a bitcoiners offer, then it wouldnt matter, but normal datacentres might be a problem. So as long as the end-product has the signs that make it possible to host the miners in a datacentre its fine. A datacenter i contacted only mentioned the hardware has to apply the "actual security standards".

[sensei]

I will also note that most, if not all, PCBs are not CE marked. Let me explain:

The CE mark is REQUIRED on all products which are self-contained (possess an intrinsic function), such as a computer. Products which are not self-contained (components) must also be CE marked if they are freely available, such as a computer monitor. Components which are not freely available do not need a CE mark, such as a printed circuit board.

Therefore, alone, the Klondike does not need a CE mark.

However, if you put it in a box and sell it as a functional unit, it would need a CE mark.

Definitely needing directives:
 
Low Voltage Electrical Safety
Electromagnetic Compatibility (EMC for both Emission and Immunity)

If your power is supplied from a single transformer, the industrial directive applies. If from a shared public source (110/220) then it is residential/light industrial.

If the Klondikes (board assemblies only) are sent to end-users for installation, no CE mark is required. However, resulting EMC behavior is the responsibility of the installer.

This is IMHO, from reading the mess of CE guidelines, opinions, and directives.

[SebastianJu]

Ok, thanks for clearing this. I think that means i wont get problems with hosting.

I will also note that most, if not all, PCBs are not CE marked. Let me explain:

The CE mark is REQUIRED on all products which are self-contained (possess an intrinsic function), such as a computer. Products which are not self-contained (components) must also be CE marked if they are freely available, such as a computer monitor. Components which are not freely available do not need a CE mark, such as a printed circuit board.

Therefore, alone, the Klondike does not need a CE mark.

However, if you put it in a box and sell it as a functional unit, it would need a CE mark.

Definitely needing directives:
 
Low Voltage Electrical Safety
Electromagnetic Compatibility (EMC for both Emission and Immunity)

If your power is supplied from a single transformer, the industrial directive applies. If from a shared public source (110/220) then it is residential/light industrial.

If the Klondikes (board assemblies only) are sent to end-users for installation, no CE mark is required. However, resulting EMC behavior is the responsibility of the installer.

This is IMHO, from reading the mess of CE guidelines, opinions, and directives.

[cardcomm]

Tell me about it. I have one planned but like the ceiling and floor it's on the waiting list. I'm sweating here most days but today was quite cool at 32C. We have 40C days here in March.

Good lord - i would in fact be dead and/or liquid!

Thanks for the pic - nice to solidify the image i had in my head  

I'm from south Texas, where we know heat. I wouldn't exactly call 32c cool!  We've had it pretty cool here so far this year - We've only topped 38c once   . We'll be hitting 40c + most days by the end of July. The real heat comes at the end of August here...

I've been thinking for a while that we need to take up a collection for an A/C unit for you! I couldn't take it...

Thanks for the updates, BTW. Awesome!!!!

[sensei]

I should also add that I think the EU prohibits any electronics to be manufactured with leaded solder (no more than 0.1% lead in any homogenous component). This means Lead-free solder must be used for anything shipped to the EU.

[freeworm]

Your clock signal is great with very sharp rising/falling edge. I am wondering how you made it.
I simply added a NOR gate after the RC delay (100 ohm, 220pF) of your original circuit, but the signal is not as good as yours. The rising / falling time is much larger than yours and the pulses are broader.
It's better than the original circuit but I still get HWs.

The errors I get now are almost certainly on the input side rather than result capture, except the infrequent overrun I see get counted.

I use a trailing edge delay circuit. The first NOR gate has P and N, and it's output goes to A of 2nd NOR gate, then through a 100R resistor to B of 2nd gate. B has the 30pF to GND. So the 2nd NOR is ORing the clock with a delayed clock. The UART is set for rising edge capture, and data is inverted when read out of FIFO (~RCREG). This will be updated in the schematic very soon.

Great! Thanks a lot for telling the details. Very helpful.

[AtomSea]

Just wanted to say "thank you" to everyone working hard on this project.
I have been following for a few weeks now - haven't watched Netflix since.... Ha! No lies
It's been a great drama seeing these boards come alive

[marto74]

I should also add that I think the EU prohibits any electronics to be manufactured with leaded solder (no more than 0.1% lead in any homogenous component). This means Lead-free solder must be used for anything shipped to the EU.

Actually almost every PCB shop in EU is working only RoHS compatible and assemblers are using lead free solder

[ecliptic]

If you want the number of HW errors like you are counting, it is simply the device HW - NMW
Avalon device HW is bad nonces + NMW

No need to change anything.

What/where is NMW?

That's an Avalon added stat
(No Matching Work)
If your version of the code isn't adding HW but just calling submit_nonce() with each one, then the HW counter will just be the number of submitted that were no good (and Diff1 Work will be the number submitted - that includes HW)

Ok. I see. I only get NMW when it's hashing before any work is sent so I no longer have it send nonces when it's not in W (working) state. So when a flush occurs it changes to R (ready) state until new work is pushed. I was getting them at start up sometimes when an old workid matched in the queue it would trigger as a HW error, so better to not even receive useless nonces.

For klondike it doesn't send midstate back, only workid and each device in a chain has it's own sequence. I store (dev*256 + workid &0xFF) in the work->subid when sent and find it when a nonce comes back. I remove completed work when sending new work.

Just started running with 4 chips. 17 WU/m, 1200 GH/s.

Nice!
Are the 4 chips all on the same side or is it 2 and 2.

Right now just on one side, U5,6,7,8 at 300MHz. I have to relocate my NOR gate hack board before I can put chips on the other bank. U8 seems to run a bit hotter than the others at 52C. Others <50C. With steamboats black heat sink and a fan the heat sink runs 37C (on that corner, barely warm to touch) in 32 ambient.

How are you measuring temperature?  I've noticed IR thermometers can often error on the low side (maybe 5-15 C below what they really are at).

Only being at ~50C for full speed (or slightly above fullspeed @ 300Mhz, i thought spec was 282Mhz for some reason) seems perhaps a bit cool

also i think the highest OC i've heard of was one of the 65GH/sec units stable @ 98 GH/sec which implies 150% OC or ~450 MHz clock.   Interesting to note burnin hasn't had luck at that yet, but i don't think his design is released to take a look at it.

[BkkCoins]

How are you measuring temperature?  I've noticed IR thermometers can often error on the low side (maybe 5-15 C below what they really are at).

Only being at ~50C for full speed (or slightly above fullspeed @ 300Mhz, i thought spec was 282Mhz for some reason) seems perhaps a bit cool

IR thermometers will give incorrect readings when the emissivity of the surface is not close to what it's set for. Typical setting is 0.95 and many surfaces are quite far from that. Metallic/reflective surfaces especially.

My main problem wasn't that but at close range the laser dot is off position due to parallax. What I do is measure just off the board edge and get readings that are ambient for the table below. This is within 2C of the current ambient in the room usually as compared to a regular digital thermometer. Then as I move the laser near board edge with gun oriented perpendicular I can see the reading change even though the laser is off position. It jumps up for the board edge to the 40s. As I move around the board I treat the highest value I see as the ASIC temp, even though it's usually about 8mm off position. The emissivity of the chips should be quite close as I checked that black matte surfaces are close to 0.95. If I move to just off the position of the thermistor then I get a value about 2C off from the thermistor reading. For the board temp. The green board surface may not have emissivity 0.95 so that could also be why it's off from the thermistor. Or it could be it's likely around 2C off generally. The board and heat sink are barley warm to touch so I don't think they are wrong in the 40Cs, and with a 10-15C differential to the chip it seems to pull heat out adequately.

When I measure the Erupter in a similar way with no cooling I find it runs hotter. More like 65C-70C, another indication it's not too far off. If the Erupter was actually at 85C I think it would shut down. Or I hope it would, as that's likely max temp for some of the chips even if the regs typically can handle higher. I have it sitting on and making contact with a larger aluminum heat sink to help dissipate heat more than the small attached bar.

[daemondazz]

Sorry if I've missed this up-thread. Have you taken any measurements as to the actual power consumption of the ASICs when running at the different speeds?

[BkkCoins]

Sorry if I've missed this up-thread. Have you taken any measurements as to the actual power consumption of the ASICs when running at the different speeds?

Not yet. I need to make up a PCIe tap cable for my meter and haven't got to it yet.

[sensei]

I should also add that I think the EU prohibits any electronics to be manufactured with leaded solder (no more than 0.1% lead in any homogenous component). This means Lead-free solder must be used for anything shipped to the EU.

Actually almost every PCB shop in EU is working only RoHS compatible and assemblers are using lead free solder

I'm not sure if you can find any non-RoHS components if you tried...

[daemondazz]

Sorry if I've missed this up-thread. Have you taken any measurements as to the actual power consumption of the ASICs when running at the different speeds?

Not yet. I need to make up a PCIe tap cable for my meter and haven't got to it yet.

No worries, thanks for your awesome work!

[notlist3d]

Truly amazing work.  Thank you for all the hard work and time, it really shows in project!

[Bicknellski]

When you get time more Klondike p0rn please... I do love a good board shot showing all the chips hashing. Ooo Ya Baby right there!

[gill83]

I thought maybe people would like to see my Franken K...

80mm fan just connected to molex for now. See I caught the red led showing data xfer. The red GND lead off back is for scope probes (why isn't it black? just me being retarded.)

Heat sink not attched to fan, just sitting on top as I disconnect often to take it to the solder station.

Right now just on one side, U5,6,7,8 at 300MHz. I have to relocate my NOR gate hack board before I can put chips on the other bank. U8 seems to run a bit hotter than the others at 52C. Others <50C. With steamboats black heat sink and a fan the heat sink runs 37C (on that corner, barely warm to touch) in 32 ambient.

Great Job!!!!  

BTW,  "32 ambient" !!!!!   LOL time for some AC.

Tell me about it. I have one planned but like the ceiling and floor it's on the waiting list. I'm sweating here most days but today was quite cool at 32C. We have 40C days here in March.

Dear BkkCoins,

Can you please elaborate a little about the changes that you made to the circuit, to make it work? I know about the missing power to the PIC, and the little jumper cable that you soldered to power it. What is that small red PCB doing? I also got a sample board made around the time you did. I also got 2 sample chips that I have mounted. But for some reason the board is not hashing.