ASICMiner BE300S Samples Arrived, <0.2W/G Achieved at Board Level

[MrTeal]

The package turns from QFN to Flip-Chip-LGA,

Package propably will be the same.

These two are inherently incompatible statements

I was reffering to this:

...The MPW samples...final chips...

The ones that are mounted onto the boards in the pictures shown are the MPW samples, and they aren't FCLGA. If the production chips are to be FCLGA then it would make sense that the package is different between the MPW samples and final chips.

[1713566217]

1713566217

[1713566217]

1713566217

[1713566217]

1713566217

[Dexter770221]

... the MPW samples, and they aren't FCLGA....

No pictures of chip itself so we don't know. There's no reason that it couldn't be a FCLGA... It's not a big processor with over thousand pins so it dosn't need to look like one....

[MrTeal]

... the MPW samples, and they aren't FCLGA....

No pictures of chip itself so we don't know. There's no reason that it couldn't be a FCLGA... It's not a big processor with over thousand pins so it dosn't need to look like one....

There are pictures of the chip mounted to the board, as well as the board mounting pattern. While you could do a flip chip like that an encapsulate it, it would seem weird to do so with those massive power and ground pads at the bottom if you're planning on pulling most of the heat out the top. You can get overmolded FC packages, but the thermals out the top aren't usually great.

I will concede it could be FCLGA though.

[Dexter770221]

... the MPW samples, and they aren't FCLGA....

No pictures of chip itself so we don't know. There's no reason that it couldn't be a FCLGA... It's not a big processor with over thousand pins so it dosn't need to look like one....

There are pictures of the chip mounted to the board, as well as the board mounting pattern. While you could do a flip chip like that an encapsulate it, it would seem weird to do so with those massive power and ground pads at the bottom if you're planning on pulling most of the heat out the top. You can get overmolded FC packages, but the thermals out the top aren't usually great.

I will concede it could be FCLGA though.

I'm also curious how friedcat want to achive better performance in mass production chips. As for now we only know that with samples he achieved better specs than simulations. 6GH/s | 0.343w/g  vs.  7.2GH/s | 0.2495W/G

[antirack]

... the MPW samples, and they aren't FCLGA....

No pictures of chip itself so we don't know. There's no reason that it couldn't be a FCLGA... It's not a big processor with over thousand pins so it dosn't need to look like one....

There are pictures of the chip mounted to the board, as well as the board mounting pattern. While you could do a flip chip like that an encapsulate it, it would seem weird to do so with those massive power and ground pads at the bottom if you're planning on pulling most of the heat out the top. You can get overmolded FC packages, but the thermals out the top aren't usually great.

I will concede it could be FCLGA though.

Sometimes things are different than they seem, especially in China.

I have seen these small LGA packages in sensors (from Analog Devices) for instance.

[friedcat]

FC, you can probably dispel a little confusion here. The MPW samples you have and are testing are not in the same package as the final chips will be in. The production chips will look significantly different than the test chips, and the board package will be different.
Is that correct?

BE200 -- QFN 8x8 or 9x9
BE300S -- FCLGA 5x5
BE300 Final Version -- FCLGA nxn (n not determined yet)

QFN in a low voltage setting introduces too much overhead, so although we wanted compatibility hard, we chose a different package.

[AJRGale]

FC, you can probably dispel a little confusion here. The MPW samples you have and are testing are not in the same package as the final chips will be in. The production chips will look significantly different than the test chips, and the board package will be different.
Is that correct?

BE200 -- QFN 8x8 or 9x9
BE300S -- FCLGA 5x5
BE300 Final Version -- FCLGA nxn (n not determined yet)

QFN in a low voltage setting introduces too much overhead, so although we wanted compatibility hard, we chose a different package.

i thought Flip Chip Land Grid Array (FCLGA), would require pins, or solder balls on the PCB, to connect to the land grid at the bottom of the package. Something like them socket'ed intel cpus, you know, they flipped the chip for the bond pads to face down, for the bond wires to be all inside the substrate.

these BE300S's still look like Quad Flat No-leads (QFN)

now Flip Chip Ball Grid Array (FCBGA), like you'd see on DDR2/3/4 ram, is what i thought the setup would be like.
..if you want exposed dies, you could follow the path of what intel does with one of their gigabit Ethernet controller chips, have the dies backside flush with the top of the substrate. or you could copy knc or cointerra and have bare dies just sitting on the substrate and held by glue.

on the note of compatibility, there was no compatibility between the BE100 and the BE200.  we all like the "drop in and run" idea, but i've never seen too much backwards compatibility with packages like this. so don't kick yourself over it, make the package the way the chips need it, even if you have to go 40 pin DIP, people will work around it

... am i rambling on?

[Dexter770221]

....
even if you have to go 40 pin DIP, people will work around it
....

Yeah, I miss alot 40 pin DIP packages...
Those were good times of C51 uC...

[sidehack]

Heck yeah. PCB costs would suck but I would totally build a miner with DIP. BGA, not so much.

The backward compatibility idea was pretty great, especially when they open-sourced BE200 design, because a lot of folks were working on stuff hoping they'd be able to reuse the same design. Would have been nice, and I'm glad they at least tried to make it happen.

So uh, so when do we get to see the Vcore/GH curve?

[Sine(X)]

The spring will be an interesting time for sure.  I hope FC and AM build a decent size self mine with their cheap power in China.

AMHash takes 47$/Ths/month (0.8W/Ghs) = 0.08 USD/kwh - is it a cheap power?

[Chris_Sabian]

The spring will be an interesting time for sure.  I hope FC and AM build a decent size self mine with their cheap power in China.

AMHash takes 47$/Ths/month (0.8W/Ghs) = 0.08 USD/kwh - is it a cheap power?

You have to figure maintenance, internet, staff, datacenter are all included in that $47 / Th/month.  Cheap power would be ~$0.03 / kwh.

[dogie]

The spring will be an interesting time for sure.  I hope FC and AM build a decent size self mine with their cheap power in China.

AMHash takes 47$/Ths/month (0.8W/Ghs) = 0.08 USD/kwh - is it a cheap power?

There is cheaper power in CN, however there are plenty of other reasons that manufacturers want farms in other countries even if they're more expensive.

[dogie]

So uh, so when do we get to see the Vcore/GH curve?

It might be misleading to post it for the BE300S if he thinks that they're going to be significantly different than BE300 - which its suggested it will be.

[MrTeal]

FC, you can probably dispel a little confusion here. The MPW samples you have and are testing are not in the same package as the final chips will be in. The production chips will look significantly different than the test chips, and the board package will be different.
Is that correct?

BE200 -- QFN 8x8 or 9x9
BE300S -- FCLGA 5x5
BE300 Final Version -- FCLGA nxn (n not determined yet)

QFN in a low voltage setting introduces too much overhead, so although we wanted compatibility hard, we chose a different package.

What do you expect the Tjc to be for the top and bottom on the production chips?

[sidehack]

So uh, so when do we get to see the Vcore/GH curve?

It might be misleading to post it for the BE300S if he thinks that they're going to be significantly different than BE300 - which its suggested it will be.

Right, but at least it'd be a good starting point. We have really good data on expected W/GH at various GH setpoints, but without knowing either the I or the V, the W is basically useless except for cooling concerns and overarching "efficiency" marketability. Sure we can plan heatsinks, but not string lengths or VRM outputs or anything. Even in accurate data is still data.

[raskul]

So uh, so when do we get to see the Vcore/GH curve?

It might be misleading to post it for the BE300S if he thinks that they're going to be significantly different than BE300 - which its suggested it will be.

Right, but at least it'd be a good starting point. We have really good data on expected W/GH at various GH setpoints, but without knowing either the I or the V, the W is basically useless except for cooling concerns and overarching "efficiency" marketability. Sure we can plan heatsinks, but not string lengths or VRM outputs or anything. Even in accurate data is still data.

the only reason FC is obliged to release any specs at all is due to the fact that the company is in perpetual ISO.
we all know the initial specs will be a wide variation of the finished product, it's always the case with AM.

[ensurance982]

The spring will be an interesting time for sure.  I hope FC and AM build a decent size self mine with their cheap power in China.

AMHash takes 47$/Ths/month (0.8W/Ghs) = 0.08 USD/kwh - is it a cheap power?

There is cheaper power in CN, however there are plenty of other reasons that manufacturers want farms in other countries even if they're more expensive.

They have to find really cheap power, if you ask me! i want to see 0.01$/KWh! That way they could build a really big farm an mine the hell out of the blockchain. AM isn't a small player, they should go as far as possible and re-claim those 40% in 2015! Also they could continue to offer a cutting-edge cloud mining!

[jdany]

The spring will be an interesting time for sure.  I hope FC and AM build a decent size self mine with their cheap power in China.

AMHash takes 47$/Ths/month (0.8W/Ghs) = 0.08 USD/kwh - is it a cheap power?

There is cheaper power in CN, however there are plenty of other reasons that manufacturers want farms in other countries even if they're more expensive.

They have to find really cheap power, if you ask me! i want to see 0.01$/KWh! That way they could build a really big farm an mine the hell out of the blockchain. AM isn't a small player, they should go as far as possible and re-claim those 40% in 2015! Also they could continue to offer a cutting-edge cloud mining!

I LOVE this talk.
Guns a'blazin.

[sidehack]

*cough*decentralize*cough*

[antirack]

They have to find really cheap power, if you ask me! i want to see 0.01$/KWh! That way they could build a really big farm an mine the hell out of the blockchain. AM isn't a small player, they should go as far as possible and re-claim those 40% in 2015! Also they could continue to offer a cutting-edge cloud mining!

Even if the network wouldn't grow much from today on, claiming 40% of the network possibly means building 100-150MW worth of new Bitcoin mines in 2015. That's about half of all of Google's capacity (265MW) by power combined. And twice the size of Facebook (78MW). And it remains a moving target during and after.

I had a post here with some numbers I put together recently:

Energy Consumption of the Bitcoin Network
http://bitcointalk.org/index.php?topic=520977.msg9825138#msg9825138