Topic: Cairnsmore1 - Quad XC6SLX150 Board - page 123. (Read 286370 times)

The efficiency is much the same between the 2 approaches. But having 48A, say around a big regulator, in a localised area causes losses in the copper of the PCB and causes heat. In the structure we have 12V at say 4-5 amps max is the distribution and this is a relatively low loss. Similarly the max point at the output of the smaller regulator approach only has to deal with a max of 12A so less loss and heat. This is a simplistic analysis but reasonably accurate.

Distribution voltage also effects the efficiency of most regulator circuits. 12V going down to 1.2V tens to be less efficient than 5V going down to 1.2V. However this is offset by the higher distribution loss of 5V. So it's swings and roundabouts which way you go.

Efficiency of most regulators varies with output current. At low currents they are usually poor. At high currents they lose some efficiency. That's copper conduction losses mainly. The best place to operate tends to mid capacity e.g. about 6A in out 12A circuit case. In this there is an equation where operating your FPGA at maximum might not be the best return and you might be cheaper with more FPGAs doing the task. Remember FPGAs have metal leadframes and solder balls which also have a conduction loss.

Can you estimate the efficiency of the power subsystem? How would the efficiency compare between 4 small regulators vs. 1 large one?

We actually thought about the possibilities of a failure and that is why each FPGA has it's own 12A core voltage switching regulator. We have actually made it possible to shut these down independently for power saving when not in use. We can use that as a last resort in a major problem event. It also a lot better situation if a fault is detected at 12A rather than a combo 48A supply. If you don't "see" a fault until you hit 48A and that might be a burning board by then.

The controller runs off the USB feed and it can be operational for debug or any other useful purpose without powering the main array. This also lets us phase power up reducing big surges at switch on.

We have thought about modular systems and I can understand how some people sell them as a good idea and we might do one eventually. Depends what level you want to be modular at. The negative side of this is that it is more expensive to manufacture say a motherboard and 4 modules that's because you but 2 types of PCB, 5 actual PCBs, and the assembly cost and test costs are close to 5X that of a single board approach. You also add the cost of connector and the electrical losses in currents passing through them, Connectors also add another thing to fail and more thimgs to buy and assemble.

Thermal design of a modular system can be poorer as well if not done well that can increase the fail rate of boards as a result.

Unless I'm missing something, I don't see a reason why having one FPGA go out would take out the whole board anyway. Having daughterboards seems like an added expense for not a whole lot of benefit.

You should update the first page with a pic and put the prices to stand out. (maybe change its colours or make it bold or something.)

Update on our warranty policy. It has been extended to 1 year for our Cairnsmore1. We will review that further once we know a bit more about the board.

Yohan

another suggestion to yohan is .... why not modular design? each LX150 is on a small miniboard that can be replaced. This way if one chip has gone bad, we wouldnt be left with a handicap 3 FPGAs board

Beside, this might give us option to "upgrade". I dont know FPGA much to see if thats possible.

OK I get it, but even then there are more peripheral connectors than you will actually be able to use. Often, there are 3 to 5 4-pin connectors per cable, but the whole cable can only support about 50 watts. So you wouldn't be able to use them all without melting something.

I guess he meant the number of PCIe6 connectors per PSU is limited, i.e. a somewhat standard one has maybe 1-2*PCIe6 but 8*Peripheral. You'd need to split the PCIe to power up to 4 Quads.

we get it, you're an genius.

i will let others to comment, i dont care

What?
The "molex" connector (i.e. peripheral device connector) is only good for about 35 watts, if I remember correctly. The PCIe 6-pin connector is good for 150 watts, however.

please use molex connector.

the pcie connector is very limited per psu.

Best thing about this board is the PCIe 6 pin power connector and the low price and reputable UK company !

Very tempting ...

Ok, back to fans for the foreseeable future

spiccioli

The Novac and similar low boiling point dialectrics used for phase change cooling are very expensive.  Something like $700 / gal IIRC.

However !, these phase change coolants are very good at dissipating heat.  If low volume enclosures to house boards could be made a very small amount of these fluids could be used to effectively cool them.

Though, capture and cooling of vapor and redelivery to the enclosures would be a pain.

 

It should not it is a dielectric fluid which evaporates very quickly.

spiccioli.

It's nice that you aren't charging an arm and a leg for small order quantities

Yes, it does. thanks for the explanation.

Well out of the gate, since they are stating it will be able to run the Icarus bitstream natively, that would mean this thing is basically 2x icarus when running in that mode. So with the default icarus bitstream that's 760MHash/s but if you use the 200MHz icarus bitstream (if these boards can handle it without impacting the life of the chip, ngzhang didn't run it because he was worried about the chip lifetime on the actual icarus boards). That would bump it up to 800MHash/s per board.

That said they mention the Control chip can manage the clocks. So it's possible the control chip would allow some degree of throttling/overclocking based on temperature, or something like that. Meaning these could in theory (if cooled well) outperform the icarus even with the same bitstream.

Now that said, these guys have a professional FPGA design team on hand. So probably for a quick smooth release they will ship with the icarus firmware, but I would expect they plan to release an "improved" firmware which is designed/tuned specifically for this board in the future. But that's an unknown. Considering others are reporting success upwards of 250+MHash/s per LX150 FPGA, that would mean there is at least some headroom to grow with these chips. So if their team could pull it off there is the *possibility* that these boards could grow to be upwards of 1GHash/s or faster per board in the near future.

But ultimately until the boards are finished manufacturing. And testing, and have been actually benchmarked, we don't really know anything. It's all speculation based on the released info and specifications at this point.

Does that help?

What sort of hash rate will the board be able achieve (sorry if this has been asked before)