Avalon Asic Design Discussion

[crazyates]

Has anyone seen this http://techcrunch.com/2012/06/26/this-fanless-heatsink-is-the-next-generation-in-cpu-cooling/



http://www.youtube.com/watch?feature=player_embedded&v=JWQZNXEKkaU

Those are still experimental, and would not be good for something with 80 chips @ 2W each.

On the other hand, have you guys looked into Water Cooling?

[cedivad]

You guys should investigate more on how server heat dissipation works. There is a reason if they are done that way. Avalon design is correct.

[PuertoLibre]

You guys should investigate more on how server heat dissipation works. There is a reason if they are done that way. Avalon design is correct.

Servers use fans with high static pressure. (I don't recall the correct term)

The current way seems to be more in line with a normal ventilation technique.

------------------------

I'll deposit a question:

Why did it matter what Jeffs Garziks environmental conditions were like if these units are (supposedly) rated up to 105F operation? (Has this changed?)

I assume Jeff wasn't roasting himself in his home. I myself am a fan of Avalon but keep in mind that does not mean I will overlook some glitches. Was Jeffs errors about software and not temperature?

I think Ngzhang seems to think it was just a bug. I would love some confirmation that this is indeed the case.

Edit: I live in an environment where temperatures reach 105F to 110F pretty regularly. I plan to move the device even further south where it will probably be just as equally hot.

BFL devices were last rated only for 95F operation.

[mrb]

A 280mm fan would have made the 4U (178mm) chassis way taller. Not good.

Wrong direction. Like a computer case intake on the side.

http://www.newegg.com/Product/Product.aspx?Item=N82E16835705056

This would not be proper airflow in a rack (and Avalon is designed to be racked).

[cedivad]

I've seen many racked servers working even without the jet-like fans. However, these fans are not slow and they draw a lot correct? I bet that they do their job correctly.

[PuertoLibre]

A 280mm fan would have made the 4U (178mm) chassis way taller. Not good.

Wrong direction. Like a computer case intake on the side.

http://www.newegg.com/Product/Product.aspx?Item=N82E16835705056

This would not be proper airflow in a rack (and Avalon is designed to be racked).

Strange, I have seen Server racks online with those kinds of configurations. (mostly 120mm fans)

[mrb]

Rackable server chassis pull air from the front, and push it to the back. Any opening on top or bottom of the chassis must assume it is obstructed by the chassis below or above.

[Photon939]

I've been asked to make a water cooling block for these, not sure how it will work out with the via's but it should give much more even cooling. With the way the boards are laid out it should also be fairly straightforward to cool both sides of the chips. Waiting on dimensions atm, should have it together fairly soon.

It will be interesting to see what kind of a price point you come up with since the chip arrays in the Avalon are so large, making it from copper would be very expensive and aluminum causes corrosion problems in water cooling loops.

[Aseras]

What I would at least like to see in the next batches are better use of forcing air in the fins of the heatsinks especially when adding more modules.I bet it would decrease temps by 5 to 10 degrees which would allow better overclocking.It should also be a more push pull fan design since the module heatsink is so long.

I actually agree,

They should have probably have gone with a large 280mm+ fan assembly. (IMO only)

Edit: Push/pull would have been better than what they did. I have to agree. I wonder how it will fare in summer to be honest. Jeffs machine is having issues. Some say it is heat related and other say it isn't.

I wish the guy with the thermal imager (Aseras I think it was) would get his machine soon.

Yep, that's me, and I am waiting. I can't wait to get mine and tear it open and look around. I can't help it, that's the scientist in me. ( I work for a very large university )

[Aseras]

I've been asked to make a water cooling block for these, not sure how it will work out with the via's but it should give much more even cooling. With the way the boards are laid out it should also be fairly straightforward to cool both sides of the chips. Waiting on dimensions atm, should have it together fairly soon.

It would be interesting to know if the blank space behind the chip has traces or if it could be drilled out. If you could make direct contact with the pad underneath you could cool them far better.

[Aseras]

A 280mm fan would have made the 4U (178mm) chassis way taller. Not good.

Wrong direction. Like a computer case intake on the side.

http://www.newegg.com/Product/Product.aspx?Item=N82E16835705056

This would not be proper airflow in a rack (and Avalon is designed to be racked).

Strange, I have seen Server racks online with those kinds of configurations. (mostly 120mm fans)

These fans you guys are looking at are crap. Server, rackmounts fans are substantial and HUGE. 120mm, but 2-3 inches thick with crazy high power motors.

[MrTeal]

There are a wide number of calculators available that will allow to find out what your thermal resistance is through a via. Simple ones are usually worst case since they only take into account the conduction of the plating and not the reflow of the solder in the untented via, but it's not rocket science to find the thermal resistance going from one side of the PCB to the other. I'd be shocked if ngzhang didn't know pretty accurately what kind of temperature rise he's expecting from the thermal pad to the heatsink.

[Aseras]

For this chip, at 100nm I don't think a QFN and vias is a problem. That's how they were all made back when the process was started. It's when you shrink the process and you start having chips with astronomical heat densities that you need better thermal conduction and BGA and heatspreaders come into play. Some of the current chips like the VRM on a 5000+ AMD or quad core chip are higher thermal densities than nuclear reactors per sq cm under full load. It's ridiculous. No we have exotic cooling and heatpipes and vapor phase cooling, along with water and other ideas.

[kev7112001]

BFL design is how it should have been done there should be no reason to need that many chips

[mrb]

How you define "should have been done" is arbitrary. Obviously Avalon has done it the right way for them: a design with many small 110nm chips was simpler and allowed them to ship before BFL who is struggling with fewer more complex 65nm chips.

[kev7112001]

How you define "should have been done" is arbitrary. Obviously Avalon has done it the right way for them: a design with many small 110nm chips was simpler and allowed them to ship before BFL who is struggling with fewer more complex 65nm chips.

using that many adds to complexity and we do know if they have actually shipped yet

[mrb]

The added PCB complexity of Avalon is completely negligible compared to the complexity of making a working 65nm chip.

[Nemesis]

How you define "should have been done" is arbitrary. Obviously Avalon has done it the right way for them: a design with many small 110nm chips was simpler and allowed them to ship before BFL who is struggling with fewer more complex 65nm chips.

by ship, you meant you received one? If not, STFU already.

[Syke]

How you define "should have been done" is arbitrary. Obviously Avalon has done it the right way for them: a design with many small 110nm chips was simpler and allowed them to ship before BFL who is struggling with fewer more complex 65nm chips.

by ship, you meant you received one? If not, STFU already.

Avalon has working chips, BFL doesn't. Plain and simple.

[Littleshop]

BFL design is how it should have been done there should be no reason to need that many chips

Each design has its merits.  The Avalon design is cheaper do design and has lower startup costs.  The BFL design is much more expensive to start but is cheaper to manufacture and cheaper to run long term.  The real prize goes to who builds first, and in this case it is Avalon.  If they can pump out two or three rounds of these they make a pretty good profit.  There also may be a secondary market for these chips outside of bitcoin.