Topic: Cairnsmore1 - Quad XC6SLX150 Board - page 120. (Read 286370 times)

I should add that I doubt we will see a full 15W draw per chip. The ZTEX based opensource hashing core (fully pipelined) has a fairly high switching rate as it is. Any "improved" core will likely not increase that much beyond what it is (I suspect improved cores will use a non-pipelined, iterative approach, ie: fully rolled up. Many many small tight cores, instead of 2-3 big unrolled ones). So chances are we might see 12W or 13W but I doubt it will peak out to 15W (but who knows it is possible). I was just stating what I expect the max sustained draw per chip will be based on the (admittedly little) released specs on the power supply circuits.

@Yohan:

Can you please let us know if we can re-program or update the bitstream via USB?

You're being pretty pessimistic on the heatsink. You can get passive northbridge heatsinks with 2.5C/W thermal resistance. With airflow you should easily be able to get Rthca to 0.5C/W or lower if you go big/loud enough. You'll have quickly decreasing gains from going much lower than that, but 2.6 or 2.7 C/W Rthja should be reasonable and that would keep the junction at 60C in a 20C room with 15W.

Wow, you really have to be borred
Power handle will be a major factor with this chips. Datasheet says of 2.2C/W Rthjc. With good heatsink and fan this will be more like 4C/W (thermal resistance adds up). So when this chip will consume 15W, core temp will increase 60C (15*4) over ambient. That means 85C core temperature (25C ambient + 60C from dissipation) and thats more than guaranteed by Xillinx for commercial grade chips. So, Rthjc is blocking higher hashrates, not power consumption. Watercooling may push this babies a little bit more...

awesome, thanks for the reply. I'm still new to the fpga side of things and i struggled through programming class so i'm trying to get an idea of how difficult it's going to be for me to set up an fpga farm. i've already pre-ordered one but i'd really like to get some more solid numbers on the performance so i can make a decision between this and other options.

I think from reading previous posts (will have to wait for official answer from Yohan to be sure) the following might help answer some of this:

It's currently expected to perform identically to 2x icarus boards (perhaps a bit faster, but since no real testing on hardware has been done yet, no "official" performance numbers are released yet. Icarus currently does 380MHash/s per board, so that would put ths at 760MHash/board if it matches exactly. If it can beat Icarus performance a bit (via more efficient design/cooling and dynamic clock control) then it can maybe come in at 800+MHash using the Icarus bitstream.

As for power consumption we can safely assume these boards should use around the same power per chip as Icarus (it's the same chips, same bitstream). Their PSU circuit might be a little more or less efficient, but likely only by 5% each way at most I would expect. So "around 10W per chip" Whole board will probably come in under 50W I would expect. (40W in FPGA, 10W for the control FPGA, and other supporting circuitry at MOST). But until they've completed real hardware testing we likely won't see "actual" power usage numbers.

Also consider that in theory newer bitstreams may push the chips harder (or be more efficient), so in the end that could alter the power usage by a much higher degree than the hardware of the board itself.

I believe the chips are each being supplied by a 15A 1.2V regulator circuit, which is fairly high efficiency. So that would mean a maximum wattage delivered to the chips of around 18W (but that would likely burn out the regulators if done over a sustained period of time). So you can likely assume about 15W max sustained draw from a chip (as in that's the max the board is capable of supplying). So the board at MOST (with a super power hungry bitstream) could consume maybe 70W. But that would be a super power hungry bitstream, and realistically, I think there would be major problems cooling the chips due to the plastic packaging at that much power dissipation.

Yes, on day one they are planning on using the Icarus bitstream. And from what I understand the first round of boards will ship pre-flashed with Icarus bitstreams pre-installed.

It's been said that the Control chip (Which is just a smaller FPGA programmed to handle all communications, and sensor monitoring, and power/clock control, it looks like a Spartan3) in combination with the USB chip they use, can drive a JTAG chain. So in otherwords you *should* be able to program it using the USB. What is currently unclear, is if this will require custom firmware on the control FPGA, or special support on the device. And if it will require custom software to do the install. (and if that software needs to be written first). So the best i can give here is you SHOULD be able to do it with the USB (no other hardware). But there is a chance the first batch of boards may not support this feature, requiring a JTAG to reflash them to add the "USB FLASH" capability to them initially. (after which USB would work fine). This is all speculation based on released info, so take with a grain of salt, and hopefully Yohan can answer more clearly in the near future.

Disclaimer: as I said, this is all speculation based on what I've seen so far in this thread, and on their site. Take everything I say here with a grain of salt, and don't consider any of it official unless it came straight from Yohan's mouth

Hope that helps

forgive me if this has already been answered, but i've read through the thread a couple times and haven't found it.

the board is expected to do ~800-1000 mh/s yes? how much power?
if they are planning on using the icarus bitstream, will it come pre-programmed? and will i need any special software/hardware to flash to a new firmware when they become available?

Yohan,
Can you please write all the specifications of the board in 1 place,

everything that need to make the FGPA'(S) to work, exactly? like what exact cables are needed, how many of them, where to connect each one?etc,
i am a noob when it come to hardware, so someone will help me, but i do need to know what to buy, etc, or what cable(s) goes to each FPGA, and from where,

i am not asking you to tell me what motherboard to buy or such, but do tell what is needed to assemble,


Thank you

The mini-usb connector looks ot be off by a few microns. I may have to cancel my preorders.. 



All kidding aside, it looks awesome and is nice to see more competition in the market.



cheers

I would be interested in this too, as well as dimensions (total thickness of the board with heatsinks) which will be useful for designing high-density rackmount clusters.

Will the heatsink fins on the "stock" heatsinks be arranged to be optimal for push/pull cooling with the airflow flowing along the long edge of the board? (so flowing from the end with the power connectors toward the opposite end for example).

Daaaaammmnn. Does someone have a stopwatch running for a start to first board? That seemed incredibly fast.


Kind of wish there was a higher resolution/more clear photo so I could print it out and pretend to have a stack sitting at my desk 

Yeah, they even can't do a decent quality photo, even shutter speed is fast

The speed and consistency with which these guys do it is amazing.

Photos of the fully populated Issue1 now on http://www.enterpoint.co.uk/cairnsmore/cairnsmore1.html.

We are still pretty much on schedule at the moment. Our biggest restriction is still the arrival of our main batch of heatsinks which arrive 14th June currently from China. We are trying to get enough of that batch air freighted for when we want them. We do have a small supply already procured and in our stores that will let some boards ship in May. We are also looking at some alternatives that we can get quickly if air freighting doesn't work. The lack of heatsinks won't affect our ability to assemble boards or do the basic testing of the finished boards and we can stockpile a large pile of boards ready to fit the heatsinks as soon as they arrive. A large pile can then ship just after the 14th.

Other than the heatsinks we are in a good component supply situation for the May/June pre-orders and the few things we might be short of in July most look like they will arrive in plenty of time. It does always take a little while to get major builds like this going but it is something that we specilise in within our general board building business.

Yohan

Are you guys at Enterpoint sure that you can deliver boards in mid June?

I'm probably just used to BFL lead time so that's why four weeks sounds impossible.

Nice to see some serious progress on this. Thanks for all the information. Look forward to seeing the first run of boards running. I guess we'll see the first assembled boards powered up and running after May 21s?

seconded.

 Bit-Pay will directly convert the BTC to the USD/EUR or whatever at the cash price you are asking and depsoit it for you..  Much cheaper than your CC provider does even.


cheers

Sadly no direct bitcoin payments. I suggested to them that they sign-up for a bit-pay account.

Subbed, ordered, excited, and hoping that bitcoins will be accepted as payment.