Bitcoin Forum
March 19, 2024, 10:54:24 AM *
News: Latest Bitcoin Core release: 26.0 [Torrent]
 
   Home   Help Search Login Register More  
Pages: [1]
  Print  
Author Topic: non-partisan ASIC Q+A  (Read 2397 times)
shmadz (OP)
Legendary
*
Offline Offline

Activity: 1512
Merit: 1000


@theshmadz


View Profile
February 08, 2013, 06:40:09 AM
Last edit: February 09, 2013, 10:01:51 AM by shmadz
 #1

I would like to start a thread where we could share information about the upcoming ASIC revolution, without the name-calling and bickering.

I am simply looking for honest questions, and honest *and unbiased* answers. (or honest, calculated projections or predictions, since concrete answers to many of the most pertinent questions are simply not possible at this time)

I'd like to stick to strictly technical questions about ASIC technology. Please direct questions such as "when will it ship?" or whatever to the hardware speculation board.


ideally each question would bring differing opinion and spark even more questions,
For example:

What is the optimal chip placement for the BFL 'little singles'? -  - If you could choose any 4 out of the 8 spots, which would you select and why?


*my guess for optimal performance would be the 4 closest to the center, but I think you could dissipate heat more easily if you placed chips on the 4 corner positions*

if you put 4 small individual copper heatsinks instead of one big one, this would allow direct contact using thermal paste instead of thermal pads in order to deal with the problem of uneven chip height. -- further, you could have a heatpipe for each of the 4 that would connect to a fan-blown radiator similar to the design for the 'single' .. there might even be enough space to mount 4 little waterblocks?!

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

And, in the spirit of a balanced discussion, also a question about the competition: how difficult would it be for Avalon to switch their current design to a flip-chip? apparently BFL did that whole process in a matter of months... though I would imagine you'd need to re-design the boards I guess?...

Any questions, answers, debate, or other productive observations or suggestions would be greatly appreciated.

-----edit 2013.02.09 -- link "4 out of the 8 spots" now links directly to full-size image, not the forum post -- link to 'single' removed/redundant

"You have no moral right to rule us, nor do you possess any methods of enforcement that we have reason to fear." - John Perry Barlow, 1996
Advertised sites are not endorsed by the Bitcoin Forum. They may be unsafe, untrustworthy, or illegal in your jurisdiction.
crazyates
Legendary
*
Offline Offline

Activity: 952
Merit: 1000



View Profile
February 08, 2013, 06:48:41 AM
 #2

Is the Little Single using the same heatsink as the SC Single? I thought the SC Single was using a copper, 4 pipe design, but the Little Single was using an aluminum one? I can't remember if I had heard that or not.

Tips? 1crazy8pMqgwJ7tX7ZPZmyPwFbc6xZKM9
Previous Trade History - Sale Thread
bce
Sr. Member
****
Offline Offline

Activity: 756
Merit: 250



View Profile
February 08, 2013, 07:41:26 AM
Last edit: February 08, 2013, 08:06:42 AM by bce
 #3

I believe that the chip layout has already been determined.

Here is the Aluminum heatsink for the Mini-Single:



Imagine this being rotated 90 degrees when compared to the Copper heatpipe heatsink for the Single SC (with same screw hole layout on both heatsink versions).   This Aluminum heatsink seems designed to cool the four most centrally placed ASIC chips - this forms a diamond shaped pattern relative to the dimensions of the PCB.  Smiley   At stock speeds, this Aluminum block will most likely perform as well as the Copper beast on the 8-chip Single, but I can't wait to see what experimental improvements people come up with, now that these BFL ASIC chips won't be plastic-encased little QFN ovens!  Grin  These chips that BFL is shipping later this month will have real potential.
shmadz (OP)
Legendary
*
Offline Offline

Activity: 1512
Merit: 1000


@theshmadz


View Profile
February 09, 2013, 12:39:05 AM
 #4

Does anyone have confirmed dimensions of the base of the aluminum heatsink?

I am hoping that I might be able to use my old Thermaltake "Big Typhoon" to replace that aluminum one, but the base is only approx. 50X50mm :sadface:

 The copper one looks like it means business and I don't imagine needing to replace that beast  Grin

"You have no moral right to rule us, nor do you possess any methods of enforcement that we have reason to fear." - John Perry Barlow, 1996
MrTeal
Legendary
*
Offline Offline

Activity: 1274
Merit: 1004


View Profile
February 09, 2013, 02:29:49 AM
 #5

Does anyone have confirmed dimensions of the base of the aluminum heatsink?

I am hoping that I might be able to use my old Thermaltake "Big Typhoon" to replace that aluminum one, but the base is only approx. 50X50mm :sadface:

 The copper one looks like it means business and I don't imagine needing to replace that beast  Grin
If you look at the white silkscreened square around the ASIC area on the original PCB, it's about 40mmx40mm based on my pixel counting. 50x50 would cover all the asics.
bce
Sr. Member
****
Offline Offline

Activity: 756
Merit: 250



View Profile
February 09, 2013, 06:27:33 AM
 #6

Does anyone have confirmed dimensions of the base of the aluminum heatsink?

I am hoping that I might be able to use my old Thermaltake "Big Typhoon" to replace that aluminum one, but the base is only approx. 50X50mm :sadface:

 The copper one looks like it means business and I don't imagine needing to replace that beast  Grin

Maybe Butterfly Labs will be selling the copper heatsinks, or watercool options in the future? - But yeah, with a drill and some work, making new cooling will be a fun thing to wach in this community!  We've all got aftermarket GPU stuff- I wonder what the measurements were on that accelero xtreme of mine?  Tongue
shmadz (OP)
Legendary
*
Offline Offline

Activity: 1512
Merit: 1000


@theshmadz


View Profile
February 09, 2013, 10:34:22 AM
 #7

after doing some "ruler-on-the-screen" measurements, it appears that MrTeal's pixel counting skills are impeccable. Meaning "standard" sized cpu coolers (and water-blocks for that matter) *should* work... (depending on how they might interfere with surrounding components, nearly any common cpu heatsink should cover all the chips.)


the thing I worry about is the mounting procedure, I'd like to use something better than a bunch of elastic bands to fix the heatsink to the board  Cheesy

I'm curious about the retention bracket design, as the image provided for the copper/heatpipe design has 4 screwholes, but the pcb appears to have only 2... and even more troubling, those 2 holes look to be directly beneath the heatsink base plate. it's difficult for me to imagine how they plan to mount the heatsink to this thing.

"You have no moral right to rule us, nor do you possess any methods of enforcement that we have reason to fear." - John Perry Barlow, 1996
kjj
Legendary
*
Offline Offline

Activity: 1302
Merit: 1024



View Profile
February 09, 2013, 12:24:18 PM
 #8

For a multichip design, you must use a thick thermal pad, something like thermagon t-flex.  You can't just use paste (including the pre-applied paste you see on OEM coolers).

The reason is that the chip tops will not be planar.  Based on the colors of the BFL board pictures, it appears they may be using ENIG with good pad planarity, which is good, but variation in the solder thickness will still leave the chips at different heights and slightly different angles.

In these circumstances, you need spacers to properly position the heatsink surface a fixed height above the PCB, and a compressible thick pad to deal with the different heights and angles.  You can see this on a lot of video cards.  A Radeon 6870, to pick an example that I know fairly well, uses a thin film paste on the main GPU, because it is a single planar chip, but all other chips needing cooling (RAM and VRM, I think) use a ~1mm thermal pad to contact a second heatsink held at a fixed height over the PCB.

17Np17BSrpnHCZ2pgtiMNnhjnsWJ2TMqq8
I routinely ignore posters with paid advertising in their sigs.  You should too.
shmadz (OP)
Legendary
*
Offline Offline

Activity: 1512
Merit: 1000


@theshmadz


View Profile
February 13, 2013, 02:18:09 AM
 #9

For a multichip design, you must use a thick thermal pad, something like thermagon t-flex.  You can't just use paste (including the pre-applied paste you see on OEM coolers).

The reason is that the chip tops will not be planar.  Based on the colors of the BFL board pictures, it appears they may be using ENIG with good pad planarity, which is good, but variation in the solder thickness will still leave the chips at different heights and slightly different angles.

In these circumstances, you need spacers to properly position the heatsink surface a fixed height above the PCB, and a compressible thick pad to deal with the different heights and angles.  You can see this on a lot of video cards.  A Radeon 6870, to pick an example that I know fairly well, uses a thin film paste on the main GPU, because it is a single planar chip, but all other chips needing cooling (RAM and VRM, I think) use a ~1mm thermal pad to contact a second heatsink held at a fixed height over the PCB.

thanks, this makes total sense, I'm familiar with these pads and I always thought they were crap for heat exchange. I've removed these pads and replaced with AS5 on many motherboards with good results, but it appears I may have been "playing with fire"  Cheesy  (never tried this on a GPU tho)

QUESTION:

Does this make the pad itself the "thermal bottleneck" in this system? and if so, isn't that copper heatsink kinda overkill?

In other words... Does the thermal substrate (in this case, the pad) have a limit to the amount of heat it can dissipate per second? If this is the case, then would it really matter as long as there's barely sufficient heatsink (i.e. able to remove the heat from the entire pad faster than the pad can transfer the heat from the chips)??

"You have no moral right to rule us, nor do you possess any methods of enforcement that we have reason to fear." - John Perry Barlow, 1996
crazyates
Legendary
*
Offline Offline

Activity: 952
Merit: 1000



View Profile
February 13, 2013, 04:01:39 AM
 #10

For a multichip design, you must use a thick thermal pad, something like thermagon t-flex.  You can't just use paste (including the pre-applied paste you see on OEM coolers).

The reason is that the chip tops will not be planar.  Based on the colors of the BFL board pictures, it appears they may be using ENIG with good pad planarity, which is good, but variation in the solder thickness will still leave the chips at different heights and slightly different angles.

In these circumstances, you need spacers to properly position the heatsink surface a fixed height above the PCB, and a compressible thick pad to deal with the different heights and angles.  You can see this on a lot of video cards.  A Radeon 6870, to pick an example that I know fairly well, uses a thin film paste on the main GPU, because it is a single planar chip, but all other chips needing cooling (RAM and VRM, I think) use a ~1mm thermal pad to contact a second heatsink held at a fixed height over the PCB.

thanks, this makes total sense, I'm familiar with these pads and I always thought they were crap for heat exchange. I've removed these pads and replaced with AS5 on many motherboards with good results, but it appears I may have been "playing with fire"  Cheesy  (never tried this on a GPU tho)

QUESTION:

Does this make the pad itself the "thermal bottleneck" in this system? and if so, isn't that copper heatsink kinda overkill?

In other words... Does the thermal substrate (in this case, the pad) have a limit to the amount of heat it can dissipate per second? If this is the case, then would it really matter as long as there's barely sufficient heatsink (i.e. able to remove the heat from the entire pad faster than the pad can transfer the heat from the chips)??

Umm why would you need a 3mm thick thermal pad? Whenever I've repasted a GPU in the past, I've always replaced the thermal pad covering the RAM chips and the VRMs, and those are all multi-chip enviroments. Usually 1mm pads work well, but sometimes you can get away with 0.5mm pads, depending on what the original pads were.

If they ever release a faster firmware for the SC Singles (80GH/s, maybe?), you can bet your ass I'll be running them caseless, fan blowing down, and one of these pads under the heatsink. From MrTeal's estimations, it should fit perfect.

Tips? 1crazy8pMqgwJ7tX7ZPZmyPwFbc6xZKM9
Previous Trade History - Sale Thread
mobodick
Hero Member
*****
Offline Offline

Activity: 840
Merit: 1000



View Profile
February 13, 2013, 08:18:13 PM
 #11


*my guess for optimal performance would be the 4 closest to the center, but I think you could dissipate heat more easily if you placed chips on the 4 corner positions*


Why would you think that placing them closer to the center would be optimal?
 Huh
shmadz (OP)
Legendary
*
Offline Offline

Activity: 1512
Merit: 1000


@theshmadz


View Profile
February 15, 2013, 01:13:10 AM
 #12

For a multichip design, you must use a thick thermal pad, something like thermagon t-flex.  You can't just use paste (including the pre-applied paste you see on OEM coolers).

The reason is that the chip tops will not be planar.  Based on the colors of the BFL board pictures, it appears they may be using ENIG with good pad planarity, which is good, but variation in the solder thickness will still leave the chips at different heights and slightly different angles.

In these circumstances, you need spacers to properly position the heatsink surface a fixed height above the PCB, and a compressible thick pad to deal with the different heights and angles.  You can see this on a lot of video cards.  A Radeon 6870, to pick an example that I know fairly well, uses a thin film paste on the main GPU, because it is a single planar chip, but all other chips needing cooling (RAM and VRM, I think) use a ~1mm thermal pad to contact a second heatsink held at a fixed height over the PCB.

thanks, this makes total sense, I'm familiar with these pads and I always thought they were crap for heat exchange. I've removed these pads and replaced with AS5 on many motherboards with good results, but it appears I may have been "playing with fire"  Cheesy  (never tried this on a GPU tho)

QUESTION:

Does this make the pad itself the "thermal bottleneck" in this system? and if so, isn't that copper heatsink kinda overkill?

In other words... Does the thermal substrate (in this case, the pad) have a limit to the amount of heat it can dissipate per second? If this is the case, then would it really matter as long as there's barely sufficient heatsink (i.e. able to remove the heat from the entire pad faster than the pad can transfer the heat from the chips)??

Umm why would you need a 3mm thick thermal pad? Whenever I've repasted a GPU in the past, I've always replaced the thermal pad covering the RAM chips and the VRMs, and those are all multi-chip enviroments. Usually 1mm pads work well, but sometimes you can get away with 0.5mm pads, depending on what the original pads were.

If they ever release a faster firmware for the SC Singles (80GH/s, maybe?), you can bet your ass I'll be running them caseless, fan blowing down, and one of these pads under the heatsink. From MrTeal's estimations, it should fit perfect.

- not sure where you got the 3mm number from, was that spec mentioned by BFL?

and thanks for the link! I was wondering where to find high-performance pads.

"You have no moral right to rule us, nor do you possess any methods of enforcement that we have reason to fear." - John Perry Barlow, 1996
shmadz (OP)
Legendary
*
Offline Offline

Activity: 1512
Merit: 1000


@theshmadz


View Profile
February 15, 2013, 01:21:48 AM
 #13


*my guess for optimal performance would be the 4 closest to the center, but I think you could dissipate heat more easily if you placed chips on the 4 corner positions*


Why would you think that placing them closer to the center would be optimal?
 Huh

no reason really, just a guess on where the manufacturer might decide to place them, based on possibly shorter connections to those resistors? in the middle, (I know next to nothing about how this works, but I remember running into ram timing issues when overclocking when I had to use the further away pair of ram slots because my heatsink got in the way)

-- and also "cost-optimal" i.e. less costly for the manufacturer by allowing a slightly smaller heatsink base to cover all 4 chips?



My personal choice would obviously be the 4 on the corners to spread the heat out as much as possible  Grin

"You have no moral right to rule us, nor do you possess any methods of enforcement that we have reason to fear." - John Perry Barlow, 1996
crazyates
Legendary
*
Offline Offline

Activity: 952
Merit: 1000



View Profile
February 15, 2013, 02:23:11 AM
 #14

Umm why would you need a 3mm thick thermal pad? Whenever I've repasted a GPU in the past, I've always replaced the thermal pad covering the RAM chips and the VRMs, and those are all multi-chip enviroments. Usually 1mm pads work well, but sometimes you can get away with 0.5mm pads, depending on what the original pads were.

If they ever release a faster firmware for the SC Singles (80GH/s, maybe?), you can bet your ass I'll be running them caseless, fan blowing down, and one of these pads under the heatsink. From MrTeal's estimations, it should fit perfect.
- not sure where you got the 3mm number from, was that spec mentioned by BFL?

and thanks for the link! I was wondering where to find high-performance pads.
No, to my knowledge the thickness of the thermal pads has not been released. I pulled that 3mm figure out of my ass cuz when I googled "thermagon t-flex" that kjj mentioned, the first result was a 3mm filler pad, which just sounds a bit too thick. A little more research shows they come in all different thicknesses.

And ya, I have some of the Fujipoly 11W/mK pads that I bought for my crappy 7970 VRM heatsink, and they definitely do their job well. This was back when the 11W/mK was the best they had. They're very soft, and you need to be careful cutting and aligning them to not rip them. I havn't tried the 17W/mK ones yet, but I'd be interested to see if they actually drop a Single's temp at all.

Tips? 1crazy8pMqgwJ7tX7ZPZmyPwFbc6xZKM9
Previous Trade History - Sale Thread
kjj
Legendary
*
Offline Offline

Activity: 1302
Merit: 1024



View Profile
February 15, 2013, 02:27:12 AM
 #15

It shouldn't matter much.  As long as the chips make good contact with the heatsink (thermally speaking), all subsets of 4 spots should be nearly equivalent.  I'm sure that some are measurably better, but not meaningfully.

If the heatsink is mounted at a fixed height from the PCB, the matter is mostly aesthetic.  If the heatsink is floating on the chips (hopefully with a compressible pad) then you want the 4 corners to evenly distribute mechanical stress.  NSEW would be just a tiny bit worse.

17Np17BSrpnHCZ2pgtiMNnhjnsWJ2TMqq8
I routinely ignore posters with paid advertising in their sigs.  You should too.
MrTeal
Legendary
*
Offline Offline

Activity: 1274
Merit: 1004


View Profile
February 15, 2013, 03:11:59 AM
 #16

I believe Josh said on the forum that there could be up to a 3 mil variance between chips. Not of course that 1 mil !=1 mm.
mobodick
Hero Member
*****
Offline Offline

Activity: 840
Merit: 1000



View Profile
February 15, 2013, 07:08:01 AM
 #17


*my guess for optimal performance would be the 4 closest to the center, but I think you could dissipate heat more easily if you placed chips on the 4 corner positions*


Why would you think that placing them closer to the center would be optimal?
 Huh

no reason really, just a guess on where the manufacturer might decide to place them, based on possibly shorter connections to those resistors? in the middle, (I know next to nothing about how this works, but I remember running into ram timing issues when overclocking when I had to use the further away pair of ram slots because my heatsink got in the way)

-- and also "cost-optimal" i.e. less costly for the manufacturer by allowing a slightly smaller heatsink base to cover all 4 chips?



My personal choice would obviously be the 4 on the corners to spread the heat out as much as possible  Grin

Aah ok.,
Well, i don't think that connection length will do much on that scale. The algorithm requires little outside communication and most work is done on the chips.
This means that heat will be the number one problem in the design.
So propably the best solution would be to start at the corners.
Good cooling alows higher clocking and more performance!
Cheesy
shmadz (OP)
Legendary
*
Offline Offline

Activity: 1512
Merit: 1000


@theshmadz


View Profile
February 21, 2013, 05:57:31 AM
 #18

This thread, while pretending to be "non-partisan" has tended strongly into the bfl camp off the bat, and now I'd like to pose some avalon questions (torn between bumping this old thread or starting a new one, oh well, here goes)


Question
regarding QFN design: is the entire back of the pcb grounded? couldn't you place a big slab of copper with some heatpipes to some fins? to take care of some of this heat? or even better a big slab of copper with channels for running fluid for water-cooling?

I'm not convinced that the avalon cooling design is optimized at all, and I think there is a great deal of room for improvement (and thus, hopefully room for better performance/overclocking!)

Even just running some copper tubing squashed in between those aluminum fins, and running a chiller (think liquid cooling with icecubes)

?? is it worth it to even try to cool these chips better? or just attach a few extra fans and some venting/channeling and just leave it at that?

"You have no moral right to rule us, nor do you possess any methods of enforcement that we have reason to fear." - John Perry Barlow, 1996
Pages: [1]
  Print  
 
Jump to:  

Powered by MySQL Powered by PHP Powered by SMF 1.1.19 | SMF © 2006-2009, Simple Machines Valid XHTML 1.0! Valid CSS!