Topic: Cairnsmore1 - Quad XC6SLX150 Board - page 115. (Read 286370 times)

How many FPGAs do you have?

LMAO! That was unexpected and awesome

Bitcoin-kitteh is on the prowl!

The designer is watching.

Ever heard about "sarcasm" ?

Let me get it strait. I spent quite some time with my quad and my single boards to find out the best way to cool them. It doesn't really matter if you blow them with a tiny 40mm fan direct on the heatsink or with one or two 120mm fans from above or from the side. It also depends on the heatsink. The "gain" between a 120mm and a well placed 40mm is ridiculous. But I wish everyone that wants to build the perfect case good luck

@yohan looks great

 

Beautiful!

hearsay?
 lol

If you take that as some sort of "expert" knowledge, i wonder what your "common sense" is

i don't have time to find any published work supporting this but this is my understanding/experience concerning heatsink layout/airflow.

the absolute best configuration is to create a "wind tunnel" effect using two fans in a push/pull configuration on either side of the heatsink. this creates an air channel across the fins which helps to keep air moving in a consistent direction with minimal turbulence, which increases the cooling capacity of the heatsink. it is necessary to point out that both fans need to be rated for the same cfm (or the pull fan slightly stronger) in order to avoid turbulence across the fins caused by air from the push fan getting slowed by the pull fan upon exit.

second up is a fan pushing through the fins with open air on the other side. a single push configuration with a high cfm fan will generally give good airflow across the fins, with any turbulence created happening upon exiting the other side of the heatsink. this is good for open air designs where airflow inside a case does not have to be considered. the same concept works for pull configurations, though reduced efficiency may occur if air is not channeled through the heatsink, and instead pulled from around it where there is less resistance. this configuration is good for closed cases where a "wind tunnel" effect can be created and air can be directed through the heatsink by a shroud.

the worst configuration is a "fan down" style mount where the fan blows air directly down on the heatsink. this causes an "air cushion" to form at the base of the heatsink which creates a bubble of turbulent air causing resistance, which reduces the amount of air actually flowing across the fins. also, fan down configurations are often open air, which further reduces efficiency by allowing air to bypass the heatsink completely by flowing around the edges.

tl;dr. he's right, it's cooling 101.

Real world experience or hearsay ? You sound like an expert but do you own a board so you can prove all your theories ?

Not sure if people caught this earlier, but if a Spartan 6 maxes out at 1 amp each (assuming ~250 mhash per Spartan 6 with the most optimized bitstream possible), you could power 28 Spartan 6s off a single EPS12 plug (28 amps == 336 watts), which would really reduce the wiring mess inside of such a flat case.

I wouldn't be against someone selling a 16-24 FPGA board that fits in a 1U case (slightly custom or not) with 6 or so 10krpm 40mm fans in the front and comes with a chopped down 1U PSU inside of it that just has the EPS12 plug and no other wiring and autostarts without ATX green wire signal.

There will be a market for this, so get it made please!

Well, I imagine a lot of people would be interested in an ATX/EATX sized board had 16 or more FPGAs that came with copper 1U heatsinks so they could drop it into cheap 1U cases. Except I think it'd require a redesign, you couldn't just fab 4 of your quad boards on one giant board and have it still fit in the ATX/EATX specs.

I'm just putting the idea out there, because there clearly is a market for things like the BFL MiniRig and I don't think the MiniRig is the right answer to the solution.

If you were interested in say mounting in a 3U rack a bolt-on mechanical board edge could be made to make the board the right width. A wider or longer variant of Cairnsmore1 would also be possible. That's only a few minutes of design time. It's then more a manufacturing thing where it splits the volume and we buy in 2 PCBs rather than 1. However if there is a serious requirement for this and enough orders say 25+ units we could do it for a small increase in cost. Other than the PCB all the other components can stay the same and we get the volume pricing advantages of that. Worst case (small numbers) the pricing delta might be about GBP £25 / US $40 per board. If that is of serious interest to anyone it is worth contacting us. It's something we can do without a serious effect on the current batch build. The current heatsinks should work well in a rack config if suitable runner spacing is used as we considered side blow when we choose them as much as the downward standard configuration.

I don't understand your point. BFL came up with a good idea to do interconnections inside of their MiniRig. This has nothing to do with cooling. BFL does not use straight through cooling in their MiniRig either, which is a complete engineering design failure.

Putting boards in a case does not cause cooling problems as the case acts as a channel. Please do not spread FUD, and look how enterprise computers are typically designed.

So which is it then BFL is great because it has got case or it sucks because the cooling is no good in that case. I for one am failing to see the point your trying to make here putting these boards in a case is going to lead to the same cooling problems as with the BFL.

I was planning on a setup similar to this, with two 120mm fans per pair of boards.


A couple brackets to hold the boards to the fans and duct the air down to them, and a couple sheets of plexi on the top and bottom to complete the channel. Wires through grommets in the plexi. Hopefully the fins on the heatsinks are orientated in the right direction.

Edit: The next time you guys update the website, do you think it would be possible to post a picture of the backside of the Issue 1.1 board? Thanks yohan.

I didnt mention BFL's cooling as a good idea. I mentioned their use of a SATA plug for the USB connection on their Minirigs. That actually IS a good idea. You can get enterprise SATA locking connectors that won't wiggle loose during transit.

Don't know why your even bother mentioning the BFL labs as a good idea then they use a similar design, I've seen more than a few my BFL is throttling posts on here.

Cases double as air channels, really. The way most of these FPGAs are built are not meant for straight through airflow, which is kind of silly. Typical "fan down" heatsinks are fail, too much turbulence to be effective cooling.

Valid points

Though I doubt BFL are buying BFL because they are complete. (sure maybe some are) the majority are buying them because currently their $/MHash are the best. And their MHash/Watt are "pretty good" (especially compared to GPU).

These boards will hopefully match (maybe beat) BFL on $/MHash, and will definitely beat them on MHash/Watt.

Enterpoint does not appear to be in the "polished shiny case" product line. That's why earlier in the thread I suggested they find system integrators to partner with, who can take their boards, and resell them in nice packaged mining rigs ready to roll (to cater to that crowd).

Ultimately I don't see mounting 16 cards in a rackmount as a "hacked" solution (especially if they release a carrier board for them that powers and controls them, and allows "solo" mining for structured cluster building).

When you start talking about generic FPGA products, or a high density board that fits in a standard case, look at the Merrick 3 (24x Spartan6 LX150s on a single standard PCIe card formfactor) They already have products that serve that niche. The problem is their cost per chip actually isn't any cheaper. (without it being heavily optimized for bitcoin mining, which reduces it's usefulness for other applications)

That's why I say, even if they did build a custom FPGA board with higher density (like an ATX or EATX drop in board with say 16 FPGAs on it for mining, though look at the Merrick1, they can get a lot more than 16 chips on an ATX or EATX size board). I highly doubt we will see much more than 10% drop in cost per chip. And with that, you're still better off scaling in a fluid modular fashion (as far as long term ROI is concerned).